Intel® Debugger (IDB) Manual

Legal Notices

Table of Contents


About This Manual

What Capabilities Does The Debugger Provide?

The debugger provides a command-line interface.

The debugger provides extensive support for debugging programs written in C, C++, and Fortran (77 and 90).

Obtaining an Install Kit

Kits, manuals, and answers to the frequently asked questions (FAQs) are available from the following sources:

  • TBD

Audience

This manual is intended for programmers who have a basic understanding of one of the programming languages that Intel IDB supports (C, C++, Fortran).

Organization

This manual is organized as follows:

  • Part I contains a quick introduction to the debugger.
    • Chapter 1 contains all the information you need to make simple use of the debugger.
  • Part II contains most of the information you need to make expert use of the debugger.
    • Chapter 2 describes preparing your program for debugging.
    • Chapter 3 describes starting the debugger.
    • Chapter 4 describes giving commands to the debugger.
    • Chapter 5 describes context for executing commands.
    • Chapter 6 describes running your program under debugger control.
    • Chapter 7 describes locating the site of the problem.
    • Chapter 8 describes how to examine the code, the data, and previously obtained information
    • Chapter 9 describes modifying the process.
    • Chapter 10 describes continuing execution of the process.
    • Chapter 11 describes snapshots as an undo mechanism.
    • Chapter 12 describes debugging optimized code.
    • Chapter 13 describes support limitations.
  • Part III contains advanced reference information.
  • The appendixes contain the following information:

Related Documentation

The following documents contain related information:

  • Man pages for the various compilers

Reporting Problems

TBD.

What to Report

Please provide the following information when you enter your problem report. Doing so will make it easier for us to reproduce and analyze your problem. If you do not provide this information, we may have to ask you for it.

  • A description of the problem. The clearer and more detailed the description, the easier it will be for us to reproduce and analyze your problem.

  • A transcript of the debugger output. You can obtain this by using the record io debugger command or by using the script(1) system command.

  • Operating system and version information. The output of uname -a is best.

  • Version information. The version number is in the welcome banner that displays when you invoke the debugger. You can also obtain the version number by invoking the debugger with the idb -V command.

  • The smallest source code example possible; build instructions (a Makefile is preferable); source languages, compiler versions, and so forth; and a pointer to a tar file containing sources or binaries that reproduce the problem. To obtain compiler versions, you can use the -V option if your compiler supports it (see the reference page for your compiler). Alternatively, you can generate the output of /usr/sbin/setld -i showing the installed compiler subsets.

  • The exact debugger commands that cause the problem to occur.

  • Any other information that you think would be helpful.

The debugger development team can use ftp to fetch sources and executables if you can place them in an anonymous FTP area. If not, you may be asked to use another method.

Conventions

The following conventions are used in this manual:

Convention Meaning
% A percent sign represents the C shell system prompt.
# A pound sign represents the default superuser prompt.
UPPERCASE
lowercase
The operating system differentiates between lowercase and uppercase characters. On the operating system level, you must type examples, syntax descriptions, function definitions, and literal strings that appear in text exactly as shown.
Ctrl/C This symbol indicates that you must press the Ctrl key while you simultaneously press another key (in this case, C).
monospaced text This typeface indicates a routine, partition, pathname, directory, file, or non-terminal name. This typeface is also used in interactive examples.
monospaced bold text In interactive examples, this typeface indicates input that you enter. In syntax statements and text, this typeface indicates the exact name of a command or keyword.
monospaced italic text Monospaced italic type indicates variable values, place holders, and function argument names.
In syntax definitions, monospaced italic text indicates non-terminal names. When a non-terminal name consists of more than one word, the words are joined using the underscore (_), for example, breakpoint_command.
italic text Italic type indicates book names or emphasized terms.
foo_bar
    : item1
    | item2
    | item3
A colon (:) starts the syntax definition of a non-terminal name (in this example, foo_bar. Vertical bars (|) separating items that appear in syntax definitions indicate that you choose one item from among those listed.
[ ] In syntax definitions, brackets indicate items that are optional.
option ;...
option ,...
option  ...
A set of three horizontal ellipses indicates that you can enter additional parameters, options, or values. A semicolon, comma, or space preceding the ellipses indicates successive items must be separated by semicolons, commas, or spaces.
idb(1) Cross-references to online reference pages include the appropriate section number in parentheses. For example, idb(1) indicates that you can find the material on the idb command in Section 1 of the reference pages. The man command % man 1 idb shows the reference page for this command.

Part I
A Quick Introduction to Using Intel Debugger

Part I provides all the information you need to make simple use of the debugger.

Chapter 1 — Overview

Intel IDB supports DBX and GDB modes. In the GDB mode, Intel IDB operates like the GNU* Debugger, GDB*. See the Starting the Debugger section to get to know how to launch the debugger in the required mode.

You look for a bug by doing the following:

  1. Find a repeatable reproducer of the bug - the simpler the reproducer is, the simpler the following steps will be to do.
  2. Prepare your program for debugging.
  3. Start the debugger.
  4. Give commands to the debugger.
    • Command the debugger to either
      • Prepare to create a process running the program, or
      • Attach to and interrupt a process that you created using normal operating system specific methods.
    • Command the debugger to create breakpoints that will pause the process as close as possible to where the bug happened.
    • If you are using the debugger to create the process, tell it to create the process now.
  5. Do whatever it takes to reproduce the bug, so that the breakpoints will stop the process close to where the bug has caused something detectably wrong to happen.
  6. Look around to determine the location of the bug:
    • If the bug is in code where the debugger has stopped the process, exit the debugger and fix the bug.
    • If the bug has not happened yet, remove any breakpoints that are triggering too often, create other breakpoints that work better at locating the problem, and continue the process.
    • If the bug has already occurred, take the same steps of creating breakpoints and so on, except with the process running backward. Unfortunately, reverse execution is a difficult problem (how do you un-erase that disk?) so the compilers and the debugger do not support it. Instead, you have to rerun from an earlier position (a snapshot if you made one, or else the beginning of the program), first creating breakpoints that stop the process sooner.

Preparing a Program for Debugging

Compile and link your program using the -g switch.

% icc -g tmp.c

For more details see Preparing the Compiler and Linker Environment.

Starting the Debugger

Before you start the debugger, make sure that you have correctly set the size information for your terminal; otherwise, the debugger's command line editing support may act unpredictably. For example, if your terminal is 25x80, you may need to set the following:

% stty rows 25 ; setenv LINES 25
% stty cols 80 ; setenv COLS 80

There are four basic alternatives for running the debugger on a process (see examples below):

  1. Have the debugger create the process using the shell command line to identify the executable to run. (dbx) (gdb)
  2. Have the debugger create the process using the debugger commands to identify the executable to run. (dbx) (gdb)
  3. Have the debugger attach to a running process using the shell command line to identify the process and the executable file that process is running. (dbx) (gdb)
  4. Have the debugger attach to a running process using the debugger commands to identify the process and the executable file that process is running. (dbx) (gdb)
DBX Mode

Intel IDB starts operating in DBX mode by default, so you do not have to specify any special options in the shell command line.

Examples:

  1. Creating the process using the shell command line.

    % idb a.out
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    object file name: a.out
    Reading symbolic information ...done
    (idb) stop in main
    [#1: stop in int main(void) ]
    (idb) run
  2. Creating the process using the debugger commands.

    % idb
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    (idb) load a.out
    Reading symbolic information ...done
    (idb) stop in main
    [#1: stop in int main(void) ]
    (idb) run
  3. Ataching to a running process using shell command line.

    % ./a.out &
    [1] 27859
    % jobs
    [1]+  Running                 ./a.out &
    % idb a.out -pid 27859
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    Reading symbolic information ...done
    Attached to process id 27859  ....

    Press Ctrl/C to interrupt the process.

  4. Attaching to the process using the debugger commands.

    % ./a.out &
    [1] 27859
    % jobs
    [1]+  Running                 ./a.out &
    % idb
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    (idb) attach 27859 a.out
    Reading symbolic information ...done
    Attached to process id 27859  ....

    Press Ctrl/C to interrupt the process.

GDB Mode

To start the debugger in the GDB mode, specify -gdb option in the shell command line.

Examples:

  1. Creating the process using the shell command line.

    % idb -gdb a.out
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    object file name: a.out
    Reading symbols from a.out...done
    (idb) break main
    Breakpoint 1 at 0x80484f6: file qwerty.c, line 9.
    (idb) run
    
  2. Creating the process using the debugger commands.

    % idb -gdb
    Linux Application Debugger for ..., Version ..., Build ...
    -------------------
    (idb) file a.out
    Reading symbols from a.out...done.
    (idb) break main
    Breakpoint 1 at 0x80484f6: file qwerty.c, line 9.
    (idb) run
  3. Ataching to a running process using shell command line.

    % ./a.out &
    [1] 27859
    % jobs
    [1]+  Running                 ./a.out &
    % idb -gdb a.out -pid 27859
    Linux Application Debugger for ..., Version ..., Build ...
    ------------------
    object file name: a.out
    Reading symbols from a.out...done.
    Attached to process id 27859  ....

    Press Ctrl/C to interrupt the process.

  4. Attaching to the process using the debugger commands.

    % ./a.out &
    [1] 27859
    % jobs
    [1]+  Running                 ./a.out &
    % idb -gdb
    Linux Application Debugger for ..., Version ..., Build ...
    (idb) file a.out
    Reading symbols from a.out...done.
    (idb) attach 27859
    Attached to process id 27859  ....

    Press Ctrl/C to interrupt the process.

Note: In the case of Fortran, routine main at which your program stops is not your main program unit. Rather, it is a main routine supplied by the Fortran system that performs some initialization and then calls your code. Just step forward using the step command a couple of times (probably twice) and you will soon step into your code.

Entering Debugger Commands

The debugger issues a prompt when it is ready for the next command from the terminal:

(idb) you type here

When you enter commands, you use the left and right arrow keys to move within the line and the up and down arrow keys to recall previous commands for editing. When you finish entering a command, press the Enter key to submit the completed line to the debugger for processing.

You can continue a line by ending the line to be continued with a backslash (\) character.

On a blank line, press the Enter key to re-execute the most-recent valid command.

Following are two very useful commands available in both modes:

(idb) help
(idb) quit

Scripting or Repeating Previous Commands

DBX Mode

To execute debugger commands from a script, use the source command as follows:

(idb) source filename

The source command causes the debugger to read and execute debugger commands from filename.

GDB Mode

The source command is not yet available in the GDB mode.

Context for Executing Commands

Although the debugger supports debugging multiple processes, it operates only on a single process at a time, known as the current process.

Processes contain one or more threads of execution. The threads execute functions. Functions are sequences of instructions that come from source lines within source files.

As you enter debugger commands to manipulate your process, it would be very tedious to have to repeatedly specify which thread, source file, and so on you wish the command to be applied to. To prevent this, each time the debugger stops the process, it re-establishes a static context and a dynamic context for your commands. The components of the static context are independent of this run of your program; the components of the dynamic context are dependent on this run.

  • The static context consists of the following:
    • A current program
    • A current file
    • A current line
  • The dynamic context consists of the following:
    • A current call frame
    • A current thread
    • The particular thread executing the event that caused the debugger to gain control of the process

You can change most of these individually to point to other instances, as described in the relevant portions of this manual, and the debugger will modify the rest of the static and dynamic context to keep the various components consistent.

Running a Program Under Debugger Control

As was shown previously, you can tell the debugger to create a process or to attach to an existing process.

After you specify the program (either on the shell command line or by using the load(dbx) or file(gdb) command), but before you have requested the debugger to create the process, you can still do things that seem to require a running process; for example, you can create breakpoints and examine sources. Any breakpoints that you create will be inserted into the process as soon as possible after it executes your program.

To have the debugger create a process (rather than attach to an existing process), you request it to run, specifying, if necessary, any arguments and input and output redirection as follows:

% idb a.out
Linux Application Debugger for ..., Version ..., Build ...  
...Preparing the Compiler and Linker Environment>
(idb) run

or

(idb) run arguments

or

(idb) run arguments > output-file

or

(idb) run arguments < input-file > output-file

The result of using any of the preceding command variations is similar to having attached to a running process.

DBX Mode

The rerun command repeats the previous run command with the same arguments and file redirection.

GDB Mode

The run command without arguments repeats the previous run (with the same arguments, input and output redirections).

r is a shortcut for the run command.

Pausing the Process at the Problem

Following are the four most common ways to pause a process:

  1. Press Ctrl/C. (dbx) (gdb)
  2. Wait until the process raises some signal. It will do this when there is an arithmetic exception, an illegal instruction, or an unsatisfiable memory access, such as an attempt to write to memory for which protection is set to read-only. (dbx) (gdb)
  3. Create a breakpoint before you run or continue the process. (dbx) (gdb)
  4. Create a watchpoint before you run or continue the process. (dbx) (gdb)
DBX Mode
  1. Pressing Ctrl/C.

    (idb) run
    ^C
    Interrupt (for process)
    
    Stopping process localhost:27903 (a.out).
    Thread received signal INT
    stopped at [int main(int):5 0x120001138]
          5     while (argc < 2 && i < 10000000)
    
  2. Waiting until the process raises some signal.

    (idb) run
    Thread received signal SEGV
    stopped at [void buggy(char*, char*):13 0x8048b79]
         13         output[k] = input[k];
    
  3. Creating a breakpoint before running or continuing the process.

    (idb) stop in main
    [#1: stop in int main(void) ]
    (idb) run
    [1] stopped at [int main(void):182 0x1200023f8]
        182     List<Node> nodeList;
    
  4. Creating a watchpoint before running or continuing the process.

    (idb) watch variable nodeList._firstNode write
    [#2: watch variable nodeList._firstNode write ]
    (idb) cont
    [2] Address 0xbffff0fc was accessed at:
    List<Node>::List(void): x_list.cxx
     [line 121, 0x8057edd]	_ZN4ListI4NodeEC1Ev(...)+0xf:         movl     $0x0, (%eax)
    	0xbffff0fc: Old value = 0x080b0ba4
    	0xbffff0fc: New value = 0x00000000
    [2] stopped at [List<Node>::List(void):123 0x8057ee3]
        123 }
    
GDB Mode
  1. Pressing Ctrl/C.

    (idb) run
    ^C
    Interrupt (for process)
    
    Stopping process localhost:27903 (a.out).
    Thread received signal INT
    main(argc=1) at x_whatHappensOnControlC.cxx: 5
    5     while (argc < 2 && i < 10000000)
    
  2. Waiting until the process raises some signal.

    (idb) run
    Starting program: /usr/examples/x_segv
    Thread received signal SEGV
    buggy (input=0xbffff2f1 "/usr/examples/x_segv", output=0x0) at x_segv.cxx:13
    13	        output[k] = input[k];
    
  3. Creating a breakpoint before running or continuing the process.

    (idb) break main
    Breakpoint 1 at 0x8049f70: file x_list.cxx, line 182.
    (idb) run
    Starting program: /usr/examples/x_list
    
    Breakpoint 1, main () at x_list.cxx:182
    182	    List<Node> nodeList;
    
  4. Creating a watchpoint before running or continuing the process.

    (idb) watch nodeList._firstNode
    Hardware watchpoint 2: nodeList._firstNode
    (idb) continue
    Continuing.
    Old value = (class Node *) 0xbffff21c
    New value = (class Node *) 0x0
    Hardware watchpoint 2: nodeList._firstNode
    List<Node>::List (this=0xbffff1fc) at x_list.cxx:123
    123	}
    

Examining the Paused Process

This section describes how to examine components of the paused process.

Looking at the Source Files

You can perform the following operations on source files:

  • Tell the debugger where your sources are, if it cannot find them.
  • Find out the name of the current source file.
  • Switch to a different source file.
  • List lines in a source file.
  • Search within a source file.
DBX Mode

Following is an example that shows listing lines and using the / command to search for a string:

(idb) file
x_list.cxx
(idb) list 180: 10
    180 main()
    181 {
    182     List<Node> nodeList;
    183
    184     // add entries to list
    185     //
>   186     IntNode* newNode = new IntNode(1);
    187     nodeList.append(newNode);
    188
    189     CompoundNode* cNode = new CompoundNode(12.345, 2);
(idb) /CompoundNode
    192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);

Aliases are shorthand forms of longer commands. This example shows using the W alias, which lists up to 20 lines around the current line. Note that a right bracket (>) marks the current line.

(idb) alias W 
W       list $curline - 10:20
(idb) W 
    176
    177
    178 //  The driver for this test
    179 //
    180 main()
    181 {
    182     List<Node> nodeList;
    183
    184     // add entries to list
    185     //
>   186     IntNode* newNode = new IntNode(1);
    187     nodeList.append(newNode);
    188
    189     CompoundNode* cNode = new CompoundNode(12.345, 2);
    190     nodeList.append(cNode);
    191
    192     nodeList.append(new IntNode(3));
    193
    194     IntNode* newNode2 = new IntNode(4);
    195     nodeList.append(newNode2);
GDB Mode

Use info source, info line, and list comands for looking at source files:

(idb) info source
Current source file is x_list.cxx
(idb) list 180,+10
180	main()
181	{
182	    List<Node> nodeList;
183
184	    // add entries to list
185	    //
186	    IntNode* newNode = new IntNode(1);
187	    nodeList.append(newNode);
188
189	    CompoundNode* cNode = new CompoundNode(12.345, 2);
(idb) forward-search CompoundNode
192	    CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
Looking at the Threads

In a multithreaded application, you can obtain information about the thread that stopped or about all the threads, and you can then change the context to look more closely at a different thread. Note that a right bracket (>) marks the current thread. And the asterisk (*) marks the thread with return status.

DBX Mode
(idb) thread
  Thread Name                      State           Substate    Policy       Pri
  ------ ------------------------- --------------- ----------- ------------ ---
>*     1 default thread            running VP 3                SCHED_OTHER  19

(idb) show thread
  Thread Name                      State           Substate    Policy       Pri
  ------ ------------------------- --------------- ----------- ------------ ---
>*     1 default thread            running VP 3                SCHED_OTHER  19
      -1 manager thread            blk SCS                     SCHED_RR     19
      -2 null thread for slot 0    running VP 1                null thread  -1
      -3 null thread for slot 1    ready VP 3                  null thread  -1
      -4 null thread for slot 2    new             new         null thread  -1
      -5 null thread for slot 3    new             new         null thread  -1
       2 threads(0x140000798)      blocked         cond 3      SCHED_OTHER  19
       3 threads+8(0x1400007a0)    blocked         cond 3      SCHED_OTHER  19
       4 threads+16(0x1400007a8)   blocked         cond 3      SCHED_OTHER  19
       5 threads+24(0x1400007b0)   blocked         cond 3      SCHED_OTHER  19
       6 threads+32(0x1400007b8)   blocked         cond 3      SCHED_OTHER  19

You can select any thread to be the focus of commands that show things. For example:

(idb) thread 2
  Thread Name                      State           Substate    Policy       Pri
  ------ ------------------------- --------------- ----------- ------------ ---
>      2 threads(0x140000798)      blocked         cond 3      SCHED_OTHER  19

GDB Mode
Within the GDB mode, you can have a look at a particular thread by specifying the internal debugger thread number. The asterisk (*) marks the current thread. Or you can observe all threads while your program is running:
(idb) info threads
  0 Thread 1024 (LWP 19513)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  1 Thread 2049 (LWP 19514)  0x805a42a in __clone from /tmp/pthread_manythreads
* 2 Thread 1026 (LWP 19515)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  3 Thread 2051 (LWP 19516)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  4 Thread 3076 (LWP 19517)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  5 Thread 4101 (LWP 19518)  0x8048288 in prime_search at pthread_manythreads.c:79
This command provides the following information about known threads:
  • Number: Number of a thread from the debugger point of view. This number is not used for numbering again if a thread dies.
  • Thread TID: Thread identifier. TID is an identifier (non duplicated) assigned to each thread inside the thread library.
  • LWP PID: Light weight process identifier (non duplicated) assigned by the Linux kernel to each process in the system. Architecturally, LinuxThreads has a PID for each thread in a multithreaded application.
  • Location: Location where the thread stopped. Its output is very similar to the backtrace location field.

You can select any thread to be the focus of commands that show things. For example:

(idb) thread 2
* 2 Thread 1026 (LWP 19515)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
Looking at the Call Stack

You can examine the call stack of any thread. Even if you are not using threads explicitly, your process will have one thread running your code. You can move up and down the stack, and examine the source being executed at each call.

DBX Mode
(idb) where 4
>0  0x804a519 in ((Node*)0x80c38f0)->Node::Node() "x_list.cxx":79
#1  0x804a568 in ((IntNode*)0x80c38f0)->IntNode::IntNode(data=2) "x_list.cxx":88
#2  0x804a61f in ((CompoundNode*)0x80c38f0)->CompoundNode::CompoundNode(fdata=12.3450003, idata=2) "x_list.cxx":103
#3  0x804a09e in main() "x_list.cxx":189
(idb) up 2
>2  0x804a61f in ((CompoundNode*)0x80c38f0)->CompoundNode::CompoundNode(fdata=12.3450003, idata=2) "x_list.cxx":103
    103     IntNode(idata),
(idb) list $curline - 10: 20
     93     cout << " type is integer, value is ";
     94     cout << _data << endl;
     95 }
     96
     97
     98 //=============================================================================
     99 // CompoundNode definition
    100 //
    101 CompoundNode::CompoundNode(float fdata, int idata)
    102   :
>   103     IntNode(idata),
    104     _fdata (fdata)
    105 {
    106 }
    107 void CompoundNode::printNodeData() const
    108 {
    109     cout << " type is compound, value is ";
    110     cout << _fdata << endl;
    111     cout << "       parent ";
    112     IntNode::printNodeData();
(idb) down 1
>1  0x804a568 in ((IntNode*)0x80c38f0)->IntNode::IntNode(data=2) "x_list.cxx":88
     88 IntNode::IntNode(int data) : _data(data)
GDB Mode
(idb) backtrace 4
#0  0x804a681 in ((Node*)(class Node *) 0x80b98f0)->Node::Node (this=(class Node *) 0x80b98f0) at x_list.cxx:79
#1  0x804a6d0 in ((IntNode*)(class IntNode *) 0x80b98f0)->IntNode::IntNode (this=(class IntNode *) 0x80b98f0, data=2) at x_list.cxx:88
#2  0x804a787 in ((CompoundNode*)(class CompoundNode *) 0x80b98f0)->CompoundNode::CompoundNode (this=(class CompoundNode *) 0x80b98f0, fdata=12.345, idata=2) at x_list.cxx:103
#3  0x804a206 in main () at x_list.cxx:189
(idb) up 2
#2  0x804a787 in ((CompoundNode*)(class CompoundNode *) 0x80b98f0)->CompoundNode::CompoundNode (this=(class CompoundNode *) 0x80b98f0, fdata=12.345, idata=2) at x_list.cxx:103
103	    IntNode(idata),
(idb) list 93,112
93	    cout << " type is integer, value is ";
94	    cout << _data << endl;
95	}
96
97
98	//=============================================================================
99	// CompoundNode definition
100	//
101	CompoundNode::CompoundNode(float fdata, int idata)
102	  :
103	    IntNode(idata),
104	    _fdata (fdata)
105	{
106	}
107	void CompoundNode::printNodeData() const
108	{
109	    cout << " type is compound, value is ";
110	    cout << _fdata << endl;
111	    cout << "       parent ";
112	    IntNode::printNodeData();
(idb) down 1
#1  0x804a6d0 in ((IntNode*)(class IntNode *) 0x80b98f0)->IntNode::IntNode (this=(class IntNode *) 0x80b98f0, data=2) at x_list.cxx:88
88	IntNode::IntNode(int data) : _data(data)
Looking at the Data

You can look at variables and evaluate expressions involving them by using the print command.

DBX Mode
(idb) print fdata
12.3450003
(idb) print idata
2
(idb) print idata + 59
61
(idb) print this
0x8068f38
(idb) print *this
class CompoundNode {
  _fdata = 12.3450003;
  _data = 2;                      // class IntNode
  _nextNode = 0x0;                // class IntNode::Node
}
GDB Mode
(idb) print fdata
$2 = 12.345
(idb) print idata
$3 = 2
(idb) print idata + 59
$4 = 61
(idb) print this
$5 = (class CompoundNode *) 0x8068f38
(idb) print *this
$6 = {<IntNode> = {<Node> = {_nextNode = 0x0}, _data = 2}, _fdata = 12.345}

The p is a shortcut, and the inspect command is a synonym for the print command.

Looking at the Signal State

The debugger shows you the signal that stopped the thread.

DBX Mode
(idb) run
Thread received signal SEGV
stopped at [void buggy(char*, char*):13 0x8048b79]
     13         output[k] = input[k];

Information: idb allows you to restart the execution of your program
from saved positions. Enter "help snapshot" for details.

GDB Mode
(idb) run
Starting program: /usr/examples/x_segv
Thread received signal SEGV
buggy (input=0xbffff2f1 "/usr/examples/x_segv", output=0x0) at x_segv.cxx:13
13	        output[k] = input[k];

Information: idb allows you to restart the execution of your program
from saved positions. Enter "help snapshot" for details.

Looking at the Generated Code

You can print memory as instructions or as data.

DBX Mode

In the following example, the wi alias lists machine instructions before and after the current instruction. Note that the asterisk (*) marks the current instruction.

(idb) alias wi 
wi	($curpc - 20)/10 i
(idb) wi
CompoundNode::CompoundNode(float, int): x_list.cxx
 [line 105, 0x120002348]	cpys	$f17,$f17,$f0
 [line 105, 0x12000234c]	bis	r31, r18, r8
 [line 101, 0x120002350]	bis	r31, r19, r16
 [line 101, 0x120002354]	bis	r31, r8, r17
 [line 101, 0x120002358]	bsr	r26, IntNode::IntNode(int)
*[line 101, 0x12000235c]	ldq	r18, -32712(gp)
 [line 101, 0x120002360]	lda	r18, 48(r18)
 [line 101, 0x120002364]	stq	r18, 8(r19)
 [line 101, 0x120002368]	sts	$f0, 24(r19)
 [line 106, 0x12000236c]	bis	r31, r19, r0
(idb) $pc/10x 
0x12000235c: 0x8038 0xa65d 0x0030 0x2252 0x0008 0xb653 0x0018 0x9813
0x12000236c: 0x0400 0x47f3
(idb) $pc/6xx 
0x12000235c: 0xa65d8038 0x22520030 0xb6530008 0x98130018
0x12000236c: 0x47f30400 0x47f5041a
(idb) $pc/2X 
0x12000235c: 0x22520030a65d8038 0x98130018b6530008
GDB Mode

Use the x command to dump memory in various formats. The disassemble command also provides disassembling capability.

(idb) x /10i $pc
0x8049973 <main+27>:         pushl    %edi
0x8049974 <main+28>:         leal     -212(%ebp), %eax
0x804997a <main+34>:         movl     %eax, (%esp)
0x804997d <main+37>:         call     0x804a2f8 <_ZN4ListI4NodeEC1Ev>
0x8049982 <main+42>:         addl     $0x4, %esp
0x8049985 <main+45>:         movl     $0x0, -200(%ebp)
0x804998f <main+55>:         pushl    %edi
0x8049990 <main+56>:         movl     $0xc, (%esp)
0x8049997 <main+63>:         call     0x8049790 <_init+792>
0x804999c <main+68>:         addl     $0x4, %esp
(idb) x /10xh $pc
0x8049973 <main+27>: 0x8d57 0x2c85 0xffff 0x89ff 0x2404 0x76e8 0x0009 0x8300
0x8049983 <main+43>: 0x04c4 0x85c7
(idb) x /6xw $pc
0x8049973 <main+27>: 0x2c858d57 0x89ffffff 0x76e82404 0x83000009
0x8049983 <main+43>: 0x85c704c4 0xffffff38
(idb) x /2xg $pc
0x8049973 <main+27>: 0x89ffffff2c858d57 0x8300000976e82404

To examine individual registers, use the print command with name of register prepended with the dollar sign ($). Commands showing all (or subset of) the registers, are specific for the mode, see examples below.

DBX Mode

To look at all the registers, use the printregs command. For example:

(idb) print $eax
1
(idb) printx $eax
0x1
(idb) printregs
$eax           0x1	1
$ecx           0x8068438	134644792
$edx           0xbfffd52c	-1073752788
$ebx           0x4213030c	1108542220
$esp [$sp]     0xbfffd3e0	-1073753120
$ebp           0xbfffd4b8	-1073752904
$esi           0x40013020	1073819680
$edi           0xbfffd524	-1073752796
$eip [$pc]     0x8049973	134519155
$eflags        0x200282	2097794
$cs            0x23	35
$ss            0x2b	43
$ds            0x2b	43
$es            0x2b	43
$fs            0x0	0
$gs            0x0	0
$orig_eax      0xffffffff	-1
$fctrl         0x37f	895
$fstat         0x0	0
$ftag          0x0	0
$fiseg         0x0	0
$fioff         0x0	0
$foseg         0x0	0
$fooff         0x0	0
$fop           0x0	0
$f0            0x00000000000000000000000000000000 0
$f1            0x00000000000000000000000000000000 0
$f2            0x00000000000000000000000000000000 0
$f3            0x00000000000000000000000000000000 0
$f4            0x00000000000000000000000000000000 0
$f5            0x00000000000000000000000000000000 0
$f6            0x00000000000000000000000000000000 0
$f7            0x00000000000000000000000000000000 0
$xmm0          0x00000000000000000000000000000000
$xmm1          0x00000000000000000000000000000000
$xmm2          0x00000000000000000000000000000000
$xmm3          0x00000000000000000000000000000000
$xmm4          0x00000000000000000000000000000000
$xmm5          0x00000000000000000000000000000000
$xmm6          0x00000000000000000000000000000000
$xmm7          0x00000000000000000000000000000000
$mxcsr         0x1f80	8064
$vfp           0xbfffd4b8	0xbfffd4b8
GDB Mode

Following commands allow you to examine sets of registers:

  • info registers

For example:

(idb) print $eax
$11 = 1
(idb) print/x $eax
$12 = 0x1
(idb) info registers
$eax           0x1	1
$ecx           0x8068438	134644792
$edx           0xbfffea8c	-1073747316
$ebx           0x4213030c	1108542220
$esp [$sp]     0xbfffe940	-1073747648
$ebp           0xbfffea18	-1073747432
$esi           0x40013020	1073819680
$edi           0xbfffea84	-1073747324
$eip [$pc]     0x8049973	134519155
$eflags        0x200282	2097794
$cs            0x23	35
$ss            0x2b	43
$ds            0x2b	43
$es            0x2b	43
$fs            0x0	0
$gs            0x0	0
$orig_eax      0xffffffff	-1
$fctrl         0x37f	895
$fstat         0x0	0
$ftag          0x0	0
$fiseg         0x0	0
$fioff         0x0	0
$foseg         0x0	0
$fooff         0x0	0
$fop           0x0	0
$f0            0x00000000000000000000000000000000 0
$f1            0x00000000000000000000000000000000 0
$f2            0x00000000000000000000000000000000 0
$f3            0x00000000000000000000000000000000 0
$f4            0x00000000000000000000000000000000 0
$f5            0x00000000000000000000000000000000 0
$f6            0x00000000000000000000000000000000 0
$f7            0x00000000000000000000000000000000 0
$xmm0          0x00000000000000000000000000000000
$xmm1          0x00000000000000000000000000000000
$xmm2          0x00000000000000000000000000000000
$xmm3          0x00000000000000000000000000000000
$xmm4          0x00000000000000000000000000000000
$xmm5          0x00000000000000000000000000000000
$xmm6          0x00000000000000000000000000000000
$xmm7          0x00000000000000000000000000000000
$mxcsr         0x1f80	8064
$vfp           0xbfffea18	0xbfffea18

Continuing Execution of the Process

After you are satisfied that you understand what is going on, you can move the process forward and see what happens. The following table shows the aliases and commands you can use to do this.

Desired Behavior Alias Command Can Take Repeat Count
Continue until another interesting thing happens c cont Yes*
Single step by line, but step over calls n next Yes
Single step to a new line, stepping into calls s step Yes
Continue until control returns to the caller None return(dbx), finish(gdb) No
Single step by instruction, over calls ni nexti Yes
Single step by instruction, into calls si stepi Yes

* In GDB mode repeat count has a different meaning for the cont command. For the other commands repeat count has the same meaning in both modes.

The following examples demonstrate stepping through lines of source code (dbx) (gdb) and stepping at the instruction level (dbx) (gdb).

DBX Mode

Stepping through lines of source code:

(idb) list $curline - 10: 20
    172
    173     if (i == 1) cout << "The list is empty ";
    174     cout << endl << endl;
    175 }
    176
    177
    178 //  The driver for this test
    179 //
    180 main()
    181 {
>   182     List<Node> nodeList;
    183
    184     // add entries to list
    185     //
    186     IntNode* newNode = new IntNode(1);
    187     nodeList.append(newNode);
    188
    189     CompoundNode* cNode = new CompoundNode(12.345, 2);
    190     nodeList.append(cNode);
    191
(idb) next
stopped at [int main(void):186 0x804825e]
    186     IntNode* newNode = new IntNode(1);
(idb) next 10
stopped at [int main(void):201 0x80486bc]
    201     nodeList.append(cNode2);
(idb) step
stopped at [void List<Node>::append(class Node* const):148 0x8048bf6]
    148     if (!_firstNode)
(idb) list $curline - 2: 6
    146 {
    147
>   148     if (!_firstNode)
    149         _firstNode = node;
    150     else {
    151         Node* currentNode = _firstNode;
(idb) step
stopped at [void List<Node>::append(class Node* const):151 0x8048c09]
    151         Node* currentNode = _firstNode;
(idb) list $curline - 2: 5
    149         _firstNode = node;
    150     else {
>   151         Node* currentNode = _firstNode;
    152         while (currentNode->getNextNode())
    153             currentNode = currentNode->getNextNode();
(idb) return
stopped at [int main(void):201 0x80486d4]
    201     nodeList.append(cNode2);
(idb) step
stopped at [int main(void):203 0x80486d7]
    203     nodeList.print();
(idb) step
stopped at [void List<Node>::print(void):162 0x8048c7e]
    162     Node* currentNode = _firstNode;
(idb) list $curline: 8
>   162     Node* currentNode = _firstNode;
    163
    164     int i = 1;
    165     cout << "The list is: " << endl;
    166     while (currentNode) {
    167         cout << "Node " << i ;
    168         currentNode->printNodeData();
    169 	currentNode = currentNode->getNextNode();
(idb) step 2
stopped at [void List<Node>::print(void):165 0x8048c8d]
    165     cout << "The list is: " << endl;

Stepping at the instruction level:

(idb) $curpc - 20/14i
void List<Node>::print(void): x_list.cxx
 [line 161, 0x8048c79]	print+0x1:            movl     %esp, %ebp
 [line 161, 0x8048c7b]	print+0x3:            subl     $0x1c, %esp
 [line 162, 0x8048c7e]	print+0x6:            movl     0x8(%ebp), %eax
 [line 162, 0x8048c81]	print+0x9:            movl     (%eax), %eax
 [line 162, 0x8048c83]	print+0xb:            movl     %eax, -12(%ebp)
 [line 164, 0x8048c86]	print+0xe:            movl     $0x1, -8(%ebp)
*[line 165, 0x8048c8d]	print+0x15:           addl     $-8, %esp
 [line 165, 0x8048c90]	print+0x18:           movl     $0x8125798, (%esp)
 [line 165, 0x8048c97]	print+0x1f:           movl     $0x80e0054, 0x4(%esp)
 [line 165, 0x8048c9f]	print+0x27:           call     std::operator<<(class basic_ostream<char,std::char_traits<char>>&, const char*)
 [line 165, 0x8048ca4]	print+0x2c:           addl     $0x8, %esp
 [line 165, 0x8048ca7]	print+0x2f:           movl     %eax, -28(%ebp)
 [line 165, 0x8048caa]	print+0x32:           addl     $-8, %esp
 [line 165, 0x8048cad]	print+0x35:           movl     -28(%ebp), %eax
(idb) stepi 2
stopped at [void List<Node>::print(void):165 0x8048c97]	print+0x1f:           movl     $0x80e0054, 0x4(%esp)
(idb) nexti
stopped at [void List<Node>::print(void):165 0x8048c9f]	print+0x27:           call     std::operator<<(class basic_ostream<char,std::char_traits<char>>&, const char*)
(idb) nexti
stopped at [void List<Node>::print(void):165 0x8048ca4]	print+0x2c:           addl     $0x8, %esp
GDB Mode

Stepping through lines of source code:

(idb) list
172
173	    if (i == 1) cout << "The list is empty ";
174	    cout << endl << endl;
175	}
176
177
178	//  The driver for this test
179	//
180	main()
181	{
182	    List<Node> nodeList;
183
184	    // add entries to list
185	    //
186	    IntNode* newNode = new IntNode(1);
187	    nodeList.append(newNode);
188
189	    CompoundNode* cNode = new CompoundNode(12.345, 2);
190	    nodeList.append(cNode);
191
(idb) next
186	    IntNode* newNode = new IntNode(1);
(idb) next 10
201	    nodeList.append(cNode2);
(idb) step
List<Node>::append (node=(class Node *) 0x80b4f40) at x_list.cxx:148
148	    if (!_firstNode)
(idb) list -2,+6
146	{
147
148	    if (!_firstNode)
149	        _firstNode = node;
150	    else {
151	        Node* currentNode = _firstNode;
(idb) step
151	        Node* currentNode = _firstNode;
(idb) list -2,+5
149	        _firstNode = node;
150	    else {
151	        Node* currentNode = _firstNode;
152	        while (currentNode->getNextNode())
153	            currentNode = currentNode->getNextNode();
(idb) finish
main () at x_list.cxx:201
201	    nodeList.append(cNode2);
(idb) step
203	    nodeList.print();
(idb) step
List<Node>::print () at x_list.cxx:162
162	    Node* currentNode = _firstNode;
(idb) set listsize 8
(idb) list ,+8
162	    Node* currentNode = _firstNode;
163
164	    int i = 1;
165	    cout << "The list is: " << endl;
166	    while (currentNode) {
167	        cout << "Node " << i ;
168	        currentNode->printNodeData();
169		currentNode = currentNode->getNextNode();
(idb) step 2
165	    cout << "The list is: " << endl;

Stepping at the instruction level:

(idb) x /14i $pc
0x804aae9 <print+21>:         addl     $-8, %esp
0x804aaec <print+24>:         movl     $0x80b4b94, (%esp)
0x804aaf3 <print+31>:         movl     $0x80869ec, 0x4(%esp)
0x804aafb <print+39>:         call     0x805036c <std::operator<<>
0x804ab00 <print+44>:         addl     $0x8, %esp
0x804ab03 <print+47>:         movl     %eax, -28(%ebp)
0x804ab06 <print+50>:         addl     $-8, %esp
0x804ab09 <print+53>:         movl     -28(%ebp), %eax
0x804ab0c <print+56>:         movl     %eax, (%esp)
0x804ab0f <print+59>:         movl     $0x8050808, 0x4(%esp)
0x804ab17 <print+67>:         call     0x804c864 <operator<<>
0x804ab1c <print+72>:         addl     $0x8, %esp
0x804ab1f <print+75>:         movl     -12(%ebp), %eax
0x804ab22 <print+78>:         testl    %eax, %eax
(idb) stepi 2
165	    cout << "The list is: " << endl;
(idb) nexti
165	    cout << "The list is: " << endl;
(idb) nexti
165	    cout << "The list is: " << endl;

Snapshots as an Undo Mechanism

Often when you move the process forward, you accidentally go too far. For example, you may step over a call that you should have stepped into.

In a program that does not use multiple threads, you can use snapshots to save your state before you step over the call. Then clone that snapshot to position another process just before the call so you can step into it.

The following example shows the stages of a snapshot being used in this way:

  1. The first stage is to build the program and start debugging.

  2. The next stage is to stop the process just before the call and take a snapshot. You can see you are just before the call because the right bracket (>) to the left of the source list shows the line about to be executed.

    (idb) next 2
    stopped at [int main(void):187 0x1200024b8]
        187     nodeList.append(newNode);
    (idb) list $curline - 10: 20
        177
        178 //  The driver for this test
        179 //
        180 main()
        181 {
        182     List<Node> nodeList;
        183
        184     // add entries to list
        185     //
        186     IntNode* newNode = new IntNode(1);
    >   187     nodeList.append(newNode);
        188
        189     CompoundNode* cNode = new CompoundNode(12.345, 2);
        190     nodeList.append(cNode);
        191
        192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
        193     nodeList.append(cNode1);
        194
        195     nodeList.append(new IntNode(3));
        196
    (idb) save snapshot
    # 1 saved at 08:41:46 (PID: 1012).
        stopped at [int main(void):187 0x1200024b8]
        187     nodeList.append(newNode);
    
  3. You now step over the call. The execution is now after the call, shown by the right bracket (>) being on the following source line.

    (idb) next
    stopped at [int main(void):189 0x1200024d0]
        189     CompoundNode* cNode = new CompoundNode(12.345, 2);
    (idb) list $curline - 10: 20
        179 //
        180 main()
        181 {
        182     List<Node> nodeList;
        183
        184     // add entries to list
        185     //
        186     IntNode* newNode = new IntNode(1);
        187     nodeList.append(newNode);
        188
    >   189     CompoundNode* cNode = new CompoundNode(12.345, 2);
        190     nodeList.append(cNode);
        191
        192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
        193     nodeList.append(cNode1);
        194
        195     nodeList.append(new IntNode(3));
        196
        197     IntNode* newNode2 = new IntNode(4);
        198     nodeList.append(newNode2);
    
  4. Oh, how you wish you hadn't done that! No problem, just clone that snapshot you made.

    (idb) clone snapshot
    Process has exited
    Process 1009 cloned from Snapshot 1.
    # 1 saved at 08:41:46 (PID: 1012).
        stopped at [int main(void):187 0x1200024b8]
        187     nodeList.append(newNode);
    
  5. Now you are in a new process before the call is executed.

    (idb) list $curline - 10: 20
        177
        178 //  The driver for this test
        179 //
        180 main()
        181 {
        182     List<Node> nodeList;
        183
        184     // add entries to list
        185     //
        186     IntNode* newNode = new IntNode(1);
    >   187     nodeList.append(newNode);
        188
        189     CompoundNode* cNode = new CompoundNode(12.345, 2);
        190     nodeList.append(cNode);
        191
        192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
        193     nodeList.append(cNode1);
        194
        195     nodeList.append(new IntNode(3));
        196
    

Note: fork() was used by the debugger both to create the snapshot and to clone it.


Part II
A Guide to Using Intel Debugger

Part II provides most of the information needed to make expert use of the debugger.

Some additional details have been moved to Part III: Advanced Topics so they do not hinder the reading of this section.

Chapter 2 — Preparing a Program for Debugging

To facilitate debugging, you can prepare your source code and the compiler and linker environment.

Preparing Your Source Code

You do not need to make changes to the source code to debug the program. However, you can do the following to make debugging easier:

  • If the source code has functions that can be called to output data structures, you can call them from the debugger; you may want to create such functions.
  • It is a good idea to make the following items part of your source code:
    • An initial stall point if you cannot create the process easily from within the debugger.
    • Assertions sprinkled liberally through the sources to help locate errors early.

Preparing the Compiler and Linker Environment

Debugging information is put into .o files by compilers. The level and format of information is controlled by compiler options. Use the -g option with the Intel® Compiler, for example:

% icc -g hello.c
...
% icpc -g hello.cpp
...

With GNU* Compiler, GCC*, use -gdwarf-2 option:

% gcc -gdwarf-2 hello.c
...
% g++ -gdwarf-2 hello.cpp
...

See your compiler's reference page for more details.

The debugging information is propagated into the a.out (executable) or .so (shared library) by the ld command. It is removed by the strip command. If you strip your programs, keep the unstripped version to use with the debugger.

The debugging information can cause .o files to be very large, causing long link times, but even so it can also be incomplete.

If you are debugging optimized code, refer to the appropriate compiler documentation for information about various -g switches and their relationship to optimization.

Chapter 3 — Starting the Debugger

You can start the debugger in the following ways:

This chapter also discusses the following topics:

Starting the Debugger from a Shell

When you invoke the debugger from a shell, you can bring a program or core file under debugger control, or you can attach to a running process.

The following is the shell syntax to invoke the debugger using the idb command:

idb
        [ -c file ]
        [ -gui ] 
        [ -gdb ]
        [ -i file ]
        [ -I directory ]
        [ -interactive ]
        [ -k ]      
        [ -line serial_line ] 
        [ -nosharedobjs ]
        [ -parallel ]  
        [ -pid process_id  ]
        [ -prompt string ]
        [ -remote ]  
        [ -rp remote_debug_protocol ]
        [ -tty terminal_device ] 
        [ -quiet ]
        [ -V ]
        [ executable_file [ core_file ] ]

The following table describes the idb command options and parameters:

Options and Parameters Description
-c file Specifies an initialization command file. The default initialization file is .dbxinit. During startup, the debugger searches for this file in the current directory. If it is not there, the debugger searches your home directory. This file is processed after the target process has been loaded or attached to.
-gdb Start working in the GDB mode.
-i file Specifies a pre-initialization command file. The default pre-initialization file is .idbrc. The debugger searches for this file during startup, first in the current directory and then in your home directory. This file is processed before the debugger has connected to the application being debugged, so that commands such as set $stoponattach = 1 will have taken effect when the connection is made.
-I directory Specifies the directory containing the source code for the target program, in a manner similar to the use command. Use multiple -I options to specify more than one directory. The debugger searches directories in the order in which they were specified on the command line.
-interactive Causes the debugger to act as though stdin is isatty(), regardless of whether or not it is. This flag is sometimes useful when using rsh to run the debugger. Currently, the only effect is to cause the debugger to output the prompt to stdout when it is ready for the next line of input.
-nosharedobjs Prevents the reading of symbol table information for any shared objects loaded when the process executes. Later in the debug session, you can enter the readsharedobj command to read the symbol table information for a specified object.
-pid process_id Specifies the process ID of the process to be debugged.
-prompt string

Specifies a debugger prompt. If the prompt argument contains spaces or special characters, enclose the argument in quotes (""). You can specify a debugger prompt when you start the debugger from a shell with the -prompt option. The default debugger prompt is (idb).

% idb -prompt ">> "
>> quit

DBX Mode

You can also change the prompt by setting the $prompt debugger variable. For example:

(idb) set $prompt = "newPrompt>> "
newPrompt>>

GDB Mode

Use set prompt prompt to specify a new prompt to use henceforth. To see the prompt used by the debugger, type the show prompt command.

(idb) set prompt (gdb mode) 
(gdb mode) show prompt
Intel IDB's prompt is "(gdb mode) ".
(gdb mode) 

Note: There is a space at the end of the first line of the example above. If space is missed, example will look like following:

(idb) set prompt (gdb mode)
(gdb mode)show prompt
Intel IDB's prompt is "(gdb mode)".
(gdb mode)
-quiet Causes the debugger to start but not to print sign-on message.
-V Causes the debugger to print its version number and exit without starting a debugging session.
executable_file Specifies the program executable file.
core_file Specifies the core file.

For example, to invoke the debugger on an executable file named a.out:

% idb a.out

To invoke the debugger on a core file:

% idb a.out core

To invoke the debugger and attach to a running process:

% idb -pid 8492 a.out

To invoke the debugger and attach to a running process when you do not know what file it is executing:

% idb -pid 8492

Starting the Debugger Using Emacs*

You can control your debugger process entirely through the Emacs Grand Unified Debugger (GUD) buffer mode, which is a variant of shell mode. All the debugger commands are available, and you can use the shell mode history commands to repeat them.

The debugger supports:

  • GNU* Emacs* Version 19 and higher.
  • XEmacs* Version 19.14 and higher.

The information in the following sections assumes you are familiar with Emacs and are using the Emacs notation for naming keys and key sequences.

For each Emacs session, before you can invoke the debugger, you must load the Intel Debugger-specific Emacs LISP code, as follows:

M-x load-file

At the Load file: prompt, type the path to the Intel Debugger-specific Emacs LISP file, which is located in the Intel IDB installation directory.  For example:

/opt/intel_idb_72/bin/idb.el

You can also place a load-file call in your Emacs initialization file (~/.emacs). For example:

(load-file "/opt/intel_idb_72/bin/idb.el")

To start the debugger with Emacs, type:

M-x idb

The following invocation line displays:

Run the debugger (like this): idb

Edit the invocation line by typing the target program and pressing Return. Emacs remembers the invocation. To debug the same program again, you need only press Return.

Emacs displays the GUD buffer and runs the debugger within it; the debugger starts and displays its (idb) prompt, indicating readiness. The GUD buffer saves all of the commands you type and the program output for you to edit. In general, interact with the debugger in the GUD buffer as you would with a debugger started from a shell.

One of the benefits of running the debugger from within Emacs is a closer correlation between program execution and source. When your program stops, for example at a breakpoint, Emacs displays the source of your program in a second buffer (source buffer) and indicates the current execution line with =>.

Note: If the source is already loaded into a buffer, Emacs often finds that buffer. However, in some NFS mounting situations, Emacs may use an alternate name for some directories and will create a second buffer for your source (often with <2> appended to the name). Be careful that you do not modify the original buffer or kill it outright.

By default, Emacs sets its current working directory to be the directory containing the target program. Because the debugger does not do this when invoked directly, you may need to change the source code search path when using the debugger from within Emacs. To set an alternate source code search path, use the debugger map source directory command.

All Emacs editing functions and GUD key bindings are available. For example:

  • You can execute a step command by typing the command in the GUD buffer.

  • You can select a line of code in the current source buffer and type a command to set a breakpoint at that position:

    C-x SPC

For more information on Emacs functionality and key bindings, see the Emacs documentation. For example:

M-x info

Then select the Emacs menu, then the Debuggers menu.

XEmacs will come up with the source buffer displayed. Use C-x 2 and a buffer menu to select the control buffer.

Starting the Debugger Using DDD*

GNU* DDD* is a graphical front-end for command-line debuggers, such as GDB*. You can use it with Intel IDB as well.

DBX Mode

Specify --ladebug and --debugger idb options in the shell command line, for example:

$ ddd --ladebug --debugger idb a.out

If idb is not accessible trough PATH environment variable, specify path to the debugger, absolute or relative, for example:

$ ddd --ladebug --debugger /opt/intel_idb_72/bin/idb a.out
GDB Mode

Specify --debugger "idb -gdb" options in the shell command line, for example:

$ ddd --debugger "idb -gdb" a.out

Alternatively, you can include set $cmdset="gdb" command to ~/.idbrc file to start Intel IDB in GDB mode. In this case -gdb option is redundant:

$ ddd --debugger idb a.out

If idb is not accessible trough PATH environment variable, specify path to the debugger, absolute or relative, for example:

$ ddd --debugger "/opt/intel_idb_72/bin/idb -gdb" a.out

Ending a Debugging Session

DBX Mode

To exit the debugger, use the quit command:

quit_command
        : quit

Alternatively, you can type exit, which is a pre-defined alias for quit.

GDB Mode

Use the quit command (q is a shorter equivalent) to exit the debugger.

quit_command
        : quit [ exit_status ]
        | q [ exit_status ]
exit_status
        : expression

Optionally you can specify debugger exit status. Example:

(idb) quit 3
% echo $?
3
%

Getting Help

To access the online help about debugger commands, use the help command.

DBX Mode
help_command
        : help [ topic ]

Enter help to see a list of help topics. Enter help command to see a list of debugger commands. Enter help idb to see a list of function-oriented debugger commands.

GDB Mode
help_command
        : help [ topic ]
        | h [ topic ]
        | complete [ args ]

If you do not specify any arguments, the help (or h) command will display a short list of named classes of commands. Use the help command with a name of a class or a name of a particular command to get more detailed help.

You may also use the complete args command to list all the possible completions for the beginning of a command. Use args to specify the beginning of the command you want completed.

Chapter 4 — Giving Commands to the Debugger

The debugger has several different mechanisms you can use to direct its behavior. It receives input from:

  • Environment variables
  • Shell command line
  • stdin, which is usually one of the following:
    • A terminal
    • A file
    • A pipe connecting the debugger to an editor (usually Emacs)
  • Other files:
    1. At startup, before attaching to or starting the target executable and before processing command line qualifiers, commands in:
      1. .idbrc, if available, otherwise
      2. ~/.idbrc, if available
    2. Just before accepting command input from you:
      1. ./dbxinit, if available, otherwise
      2. ~/.dbxinit, if available
    3. Files specified in the source command

Some examples of the difference between .idbrc and .dbxinit are shown in the following table:

Example Command If Used in .idbrc If Used in .dbxinit
Assume the command "set $stoponattach = 1" is in one of these files and you invoked the debugger as:
% idb -pid process_id executable_file
The debugger attaches and stops. The debugger attaches and waits for you to press Ctrl/C; subsequent attaches will stop.
Assume the command "stop in main" is in one of these files. The debugger generates a message that there is no main in which to place a breakpoint, because there is no target yet. The debugger sets the breakpoint (assuming there is a main in the target).

This chapter discusses the following topics:

Debugger's Command Processing Structure

The debugger processes commands as follows:

  1. Prompts for input.
  2. Obtains a complete line from the input file and performs:
  3. Parses the entire line according to the parsing rules for the current language.
  4. Executes the commands.

Interrupting a Debugger Action

To interrupt program execution or to abort a debugger action, press Ctrl/C. This returns the debugger to the prompt.

Entering and Editing Command Lines

The debugger reads lines from stdin. The debugger supports command line editing when processing stdin if stdin is a terminal and the debugger variable $editline is non-zero (the default; see the set command to change it). For this to work correctly, you must set the terminal width to the correct value. After editing, press the Return key to send the line to the debugger.

  • Use the left and right arrow keys to edit parts of the line.
  • Use the up and down arrow keys to recall and edit earlier commands.

Note: When you use the up and down arrow keys, the debugger skips duplicate commands. To see a complete list of the commands you have entered, use the history command.

The debugger copies each line from stdin to the record input file, if you have requested that file.

The debugger scans each line from the beginning, looking for backslash (\) characters, which 'quote' the immediately following character. If the line ends in a quoted newline, then another line is similarly processed from stdin and appended to the first one, with the quoted newline removed.

Whether or not command line editing is enabled, you can always use your terminal's cut-and-paste function to avoid excessive typing while entering input.

History Replacement of the Line

Leading spaces and tabs are removed from the assembled line.

For assembled lines that begin with an exclamation point (!), the following rules apply:

  • If the second character is also an exclamation point (!), the assembled line is replaced by the most-recent entry from the history list. Any remaining characters after the digits or ! are appended to the assembled line.
  • Otherwise, spaces and tabs are skipped, and one of the following actions occurs:
    • If the next character is a digit, then the digits are read as a decimal number, and the assembled line is replaced by that line from the history list, with 1 being the oldest entry.
    • If the next character is a hyphen (-), then the digits following it are read as a decimal number, and the assembled line is replaced by that line from the history list, with -1 being the most-recent entry.
    • Otherwise, the rest of the line is used to find the most-recent command that starts with those characters, and the assembled line is replaced by that line from the history list.
    In the first two cases, any remaining characters after the digits are appended to the assembled line.

For lines that begin with a caret (^), these rules apply:

  • The line is analyzed to extract the following:
    1. The characters following the first caret but before a second caret, or until the end of line. These characters are the target string.
    2. If there is a second caret, the characters following it but before a third caret, or until the end of line. These characters are the replacement string.
    3. If there is a third caret, the characters following it to the end of the line. These characters are the append string.
  • The most-recent entry from the history list is checked to see if it has an occurrence of the target string. If it does not, an error is reported.
  • The assembled line is replaced by this most-recent entry, except that the first occurrence of the target string is replaced by the replacement string (possibly zero length), and the append string is appended to the assembled line.

The assembled line is now appended to the history list.

Exclamation points and carets cannot be used in command lists built with braces ({}); for example, {print3; !!3} will not parse. They may be used in scripts.

History in a command list is not limited by braces, but goes all the way back. For example:

(idb) print 1
1
(idb) stop at 182 { print 2; history 3}
[#1: stop at "x_list.cxx":182 { print 2; history 3} ]
(idb) run
2
11: print 1
12: stop at 182 {print 2; history 3}
13: run
[1] stopped at [int main(void):182 0x804a0a0]
    182     List<Node> nodeList;

Note: Commands in breakpoint action lists are not entered into the history list.

Alias Expansion of the Line (DBX Mode only)

The assembled line is now subjected to alias expansion. This is done by scanning the line, looking for pound (#), semicolon (;), and left brace ({) characters that are not inside strings.

  • Strings are recognized by their opening and closing double or single quotes. Backslash quotation causes a quote character not to terminate the string.

  • Pound (#) characters and all that follow to the end of the line are discarded, unless the pound character is the very first character in the line. If that is the case, the pound character is not discarded because a completely empty line has special meaning. An exception is made for pound (#) characters that are surrounded by non-whitespace characters, such as "file#name". This is needed because the tmpnam standard library function generates file and directory names containing pound (#) characters.

The debugger performs alias expansion as follows:

  1. At the beginning of the line, and immediately after semicolon (;) or left brace ({) characters not inside strings, the debugger checks for the occurrence of an alias identifier.

  2. If it finds an alias identifier, it associates the formal parameters of the alias with the specified actual parameters.

    If the alias has no formal parameters, this match consumes no more of the input.

    1. If there are formal parameters, white space is skipped, and then a '(' character is checked for and skipped. The characters following the '(' up to the first non-nested ',' or ')' character are associated with the formal parameter.

      Again, the characters within strings are not tested. Nesting is caused by '(' and ')' characters outside of strings.

    2. If there are more formal parameters, the ',' character is treated as the terminator of the actual parameter. It is skipped and processing continues as for the first parameter.
  3. After the alias and the correct number of actuals have been identified, all the characters from the start of the alias identifier to its end (no parameters) or the trailing ')' (one or more parameters) are replaced by the expansion.

  4. Within the definition of the alias, all occurrences of the formal parameter are replaced by the actual parameter, regardless of whether or not it is in a string.

Environment Variable Expansion

The debugger expands environment variables and the leading tilde (~) in the following cases:

  • As part of a command in which a file name or a directory is expected.
  • In the arguments to run or rerun(dbx).

As in any shell, you can group an environment variable name using a pair of curly braces ({}), and quote a dollar sign ($) by preceding it with a backslash (\).

The following table shows how various environment variables expand. It assumes that the home directory is /usr/users/hercules and the environment variable BIN is /usr/users/hercules/bin.

Command with Environment Variable Expands into
load ~/a.out load /usr/users/hercules/a.out
load $BIN/a.out load /usr/users/hercules/bin/a.out
load ${BIN}2/a\$b load /usr/users/hercules/bin2/a$b
map source directory $BIN ${BIN}2 map source directory /usr/users/hercules/bin /usr/users/hercules/bin2
stop at "$BIN/a.out":20 stop at "/usr/users/hercules/bin/a.out":20
run $BIN/a.out ~/core run /usr/users/hercules/bin/a.out /usr/users/hercules/core

Syntax of Commands

The debugger has different parsing rules for each of the different languages it supports. A line is processed according to the current language, even if executing the line will change the current language.

Lexical Elements of Commands

For the debugger to parse the line, it must first turn the line into a sequence of tokens, a process called "tokenizing" or "lexical analysis". Tokenizing is done with a state machine.

As the debugger starts tokenizing a line into a command, it starts processing the characters using the lexical state LKEYWORD. It uses the rules for lexical tokens in this state, recognizing the longest sequence of characters that forms a lexical token.

After the lexical token is recognized, the debugger appends it to the tokenized form of the line, perhaps changes the state of the tokenizer, and starts on the next token.

For more detailed information on lexical elements, see Lexical Elements of Commands in Part III.

Grammar of Commands

Each command line must parse as one of the following:

input
        : command_list
        | comment

A command_list is a sequence of commands that are executed one after the other.

command_list
        : command ;...
        | command ;
        | command

A comment is a line that begins with a pound (#) character.

comment
        : #

Any text after an unquoted pound character is ignored by the debugger. If the first non-whitespace character on a line is a pound character, the whole line is ignored.

Note: The difference between a blank command line and a command line that is a comment is that a blank line entered from the keyboard will cause the debugger to repeat the previous command and the comment line will not. Blank lines not entered from the keyboard are treated as comment lines.

Keywords Within Commands

If the identifiers thread, in, at, and if occur within the expression in the following commands, the debugger treats them as keywords unless they are enclosed within parentheses (()).

For example, if your program has thread defined as an integer, enter the following command to inspect the first thread levels of the stack.

For example:

(idb) where 3
>0  0x8049a6c in c() "x_whereAmbigParse.c":7
#1  0x8049a8f in b() "x_whereAmbigParse.c":12
#2  0x8049aa2 in a() "x_whereAmbigParse.c":13
(idb)
(idb)
(idb)
(idb) where three(3)
>0  0x8049a6c in c() "x_whereAmbigParse.c":7
#1  0x8049a8f in b() "x_whereAmbigParse.c":12
#2  0x8049aa2 in a() "x_whereAmbigParse.c":13
(idb)
(idb)
(idb)
(idb) where thread (1)
Stack trace for thread 1
>0  0x8049a6c in c() "x_whereAmbigParse.c":7
#1  0x8049a8f in b() "x_whereAmbigParse.c":12
#2  0x8049aa2 in a() "x_whereAmbigParse.c":13
#3  0x8049abe in main() "x_whereAmbigParse.c":17
#4  0x400dd177 in __libc_start_main(...) in /lib/i686/libc.so.6
#5  0x8049929 in _init(...) in /usr/examples/x_whereAmbigParse

(idb)
(idb)
(idb)
(idb) where three(3) thread (1)
Stack trace for thread 1
>0  0x8049a6c in c() "x_whereAmbigParse.c":7
#1  0x8049a8f in b() "x_whereAmbigParse.c":12
#2  0x8049aa2 in a() "x_whereAmbigParse.c":13

(idb)
(idb)
(idb)
(idb) where (thread(3))
>0  0x8049a6c in c() "x_whereAmbigParse.c":7
#1  0x8049a8f in b() "x_whereAmbigParse.c":12
#2  0x8049aa2 in a() "x_whereAmbigParse.c":13
(idb)
(idb)
(idb)
Using Braces to Make a Composite Command

It is possible to surround a command_list with braces to make it work like a single command. Some places require a braced_command_list just for readability, or to assist the debugger in understanding your input.

braced_command_list
        : { command_list }
Conditionalizing Command Execution

The debugger provides the if command, whose behavior depends on the value of an expression.

if_command
        : if expression braced_command_list [ else_clause ]

else_clause
        : else braced_command_list

In this command, the first braced_command_list is executed if expression evaluates to a non-zero value; otherwise, the braced_command_list in the else_clause is executed, if specified.

For example:

(idb) set $c = 1
(idb) assign pid = 0
(idb) if (pid < $c) { print "Greater" } else { print "Lesser" }
Greater
Debugger Variables

Debugger variables are pseudovariables that exist within the debugger instead of within your program. They have the following uses:

  • Support some limited programming capabilities within the debugger command language
  • Allow you to examine and change various debugger options
  • Allow you to find out exactly what various debugger commands did

The following table lists the three different varieties of debugger variables:

Kind of variable Purpose
User-defined variables You create these and can set them to a value of any type.
Preference variables You modify these to change debugger behavior. You can only set a preference variable to a value that is valid for that particular variable.
Display/state variables These variables display the parts of the current debugger state. You cannot modify them.

For more information about debugger variables, see Appendix 1 — Debugger Variables.

The following commands deal specifically with debugger variables:

dbgvar_command
        : set dbgvar_name = expression
        | set dbgvar_name
        | set
        | unset dbgvar_name

The dbgvar_name should not exist anywhere in your program, or you may confuse yourself about which of the occurrences you are actually dealing with. The predefined debugger variables all start with a dollar sign ($), to help avoid this confusion. It is strongly recommended that you follow the same practice; in a future release, all debugger variables will be required to start with a dollar sign.

Note: If a debugger variable exists that shares a name with a program variable, and you print an expression involving that name, which of the two variables the debugger finds is undefined.

The first form creates the debugger variable if it does not already exist. It then sets the value of the debugger variable to the result of evaluating the expression. For example:

(idb) set $myLoopCounter = 0
(idb) print $myLoopCounter
0

The second form is equivalent to the command set dbgvar_name = 1. For example:

(idb) print $stoponattach
0
(idb) set $stoponattach
(idb) print $stoponattach
1

The set form shows all the debugger variables and their values:

(idb) set
$ascii = 0
$beep = 1
$catchexecs = 0
$catchforkinfork = 0
$catchforks = 0
$childprocess = 0
$cmdset = "idb"
$curevent = 0
$curfile = "x_list.cxx"
$curfilepath = "../src/x_list.cxx"
$curline = 182
$curpc = 0x8049f4c
$curprocess = 30078
$cursrcline = 182
$cursrcpc = 0x8049f4c
$curthread = 1
$dbxoutputformat = 0
$dbxuse = 0
$debuggerpid = 30076
$decints = 0
$disasm_shows_unwind = 0
$doverbosehelp = 1
$editline = 1
$eventecho = 1
$floatshrinking = 1
$framesearchlimit = 0
$funcsig = 1
$givedebughints = 1
$hasmeta = 0
$hexints = 0
$historylines = 20
$indent = 1
$lang = "C++"
$lasteventmade = 0
$lc_ctype = "en_US.ISO8859-1"
$listwindow = 20
$main = "\"x_list.cxx\"`main"
$maxarrlen = 1024
$maxstrlen = 128
$memorymatchall = 0
$myLoopCounter = 0
$octints = 0
$overloadmenu = 1
$page = 1
$pagewindow = 0
$parentprocess = 0
$pimode = 1
$prompt = "(idb) "
$readtextfile = 0
$regstyle = 1
$repeatmode = 1
$reportsotrans = 0
$showlineonstartup = 0
$showwelcomemsg = 1
$stackargs = 1
$statusargs = 1
$stepg0 = 0
$stoponattach = 1
$stopparentonfork = 0
$symbolsearchlimit = 100
$threadlevel = "native"
$usedynamictypes = 1
$verbose = 0

To see the value of just one debugger variable, print it. For example:

(idb) print $catchexecs
0

The unset form deletes the debugger variable. Some predefined debugger variables either cannot be deleted or are automatically recreated in the future when needed. For example:

(idb) unset $myLoopCounter
(idb) print $myLoopCounter
Symbol "$myLoopCounter" is not defined.
(idb) unset $catchforks
Warning: The debugger variable "$catchforks" was not unset because it is an idb predefined variable

Scripting or Repeating Previous Commands

To repeat the last command line, enter two exclamation points (!) or press the Return key. You can also enter !-1.

command_repetition_command
        : !!
        | ! integer
        | !- integer
        | ! string

To repeat a command line entered during the current debugging session, enter an exclamation point followed by the integer associated with the command line. (Use the history command to see a list of commands used.) For example, to repeat the seventh command used in the current debugging session, enter !7. Enter !-3 to repeat the third-to-the-last command. See also History replacement of the line.

To repeat the most-recent command starting with a string, use the last form of the command. For example, to repeat a command that started with bp, enter !bp.

Following are other ways to reuse old commands and save typing effort:

  • Use a completely empty line to repeat the last command but not the last line, which could have been a comment or a syntactically invalid attempt at a command. Immediately pressing the Return key is the recommended way of doing this.
  • Use command line editing to recall and modify commands you have already entered.
  • It is often useful to have a text editor up and running while debugging, and use it to assemble short scripts that you can copy and paste to the debugger. Keep a separate text file that has such scripts in it, as well as other notes you wish to keep. This provides continuity from one debugging session to the next, and from one day to the next.

If you place commands in a file, you can execute them directly from the file rather than cutting and pasting them to the terminal. For example:

execute_commands_from_file_command
        : source filename
        | playback input filename

Use the source command to read and execute commands from a file. (You can also execute debugger commands when you invoke the debugger by creating an initialization file named .dbxinit.) These commands can be nested, and as each comes to an end, reading resumes from where it left off in the previous file.

Be aware, however, that blank lines in these files do not repeat the last command, unlike what blank lines do when entered from the terminal. Format the commands as if they were entered at the debugger prompt.

Use the pound character (#) to create comments to format your scripts.

The following is an example debugger script:

(idb) sh cat ../src/myscript
step
where 2

The following example shows how to execute it:

(idb) run
[1] stopped at [int main(void):182 0x8049f38]
    182     List<Node> nodeList;
(idb) source ../src/myscript
stopped at [List<Node>::List(void):121 0x8058256]
    121 List<NODETYPE>::List() : _firstNode(NULL)
>0  0x8058256 in ((List<Node>*)0xbffff35c)->List<Node>::List() "x_list.cxx":121
#1  0x8049f47 in main() "x_list.cxx":182

When a command file is executed, the value of the $pimode debugger variable determines whether the commands are echoed. If the $pimode variable is set to 1, commands are echoed; if $pimode is set to 0 (the default), commands are not echoed. The debugger output resulting from the commands is always echoed.

Recording Input and Output

To help you make command files, as well as to help you see what has happened before, the debugger can write both its input and its output to files, as follows:

record_command
        : record io  [ filename ]
        | record input [ filename ]
        | record output [ filename ]
        | unrecord io
        | unrecord input
        | unrecord output

Use record input to save debugger commands to a file. The commands in the file can be executed using the source command or the playback input command.

If no file name is specified, the debugger creates a file with a random file name in /tmp as the record file. The debugger issues a message giving the name of that file.

To stop recording debugger input or output, redirect as shown in the following example, use the appropriate version of the unrecord command, or exit the debugger:

(idb) record input /dev/null
(idb) record output /dev/null

The following example shows how to use the record input command to record a series of debugger commands in a file named myscript:

(idb) record input myscript
(idb) stop in main
[#1: stop in int main(void) ]
(idb) run
[1] stopped at [int main(void):182 0x804a0a0]
    182     List<Node> nodeList;
(idb) record input /dev/null

This example results in the following recorded input in myscript:

(idb) sh cat myscript
stop in main
run
record input /dev/null

The record output command saves the debugger output to a file. The output is simultaneously written to stdout (normal output) or stderr (error messages). For example:

(idb) record output myscript
(idb) stop in List<Node>::append
[#2: stop in void List<Node>::append(class Node* const) ]
(idb) cont
[2] stopped at [void List<Node>::append(class Node* const):148 0x804aa4a]
    148     if (!_firstNode)
(idb) next
stopped at [void List<Node>::append(class Node* const):149 0x804aa53]
    149         _firstNode = node;

After the above commands are executed, myscript contains the following:

(idb) sh cat myscript
[#2: stop in void List<Node>::append(class Node* const) ]
[2] stopped at [void List<Node>::append(class Node* const):148 0x804aa4a]
    148     if (!_firstNode)
stopped at [void List<Node>::append(class Node* const):149 0x804aa53]
    149         _firstNode = node;

The record io command saves both input to and output from the debugger. For example:

(idb) record io myscript
(idb) stop in main
[#1: stop in int main(void) ]
(idb) run
[1] stopped at [int main(void):12 0x120001130]
     12 int i;
(idb) quit
% cat myscript
(idb) stop in main
[#1: stop in int main(void) ]
(idb) run
[1] stopped at [int main(void):12 0x120001130]
     12 int i;
(idb) quit

If input or output is already being recorded, a new record input command will close the old file and record to a new one, rather than record simultaneously to two files. In that connection, note that record io is equivalent to the combination of record input and record output, and will cause any open recording files to be closed.

Note that the prompt itself is only recorded for record io.

Viewing the Command History

You can see all the commands you have already entered by using the history command. Use history_number to indicate how many commands to show, starting with the most recent. If you do not specify $historylines, the 20 previous commands are shown. See also History replacement of the line.

history_command
        : history [ integer_constant ]

For example:

(idb) history 6
18: stop at 182
19: run
20: stop at 103
21: cont
22: print "history_EXAMPLE START"
23: history 6

Defining Aliases (DBX mode only)

You can extend the set of debugger commands by defining aliases.

When the debugger is tokenizing a command line, it expands aliases and then retokenizes the expansion.

alias_command
        : alias [ alias_name ]
        | alias alias_name [ ( argument_name, ...) ] string
        | unalias alias_name

The following example shows how to define and use an alias:

(idb) alias cs 
alias cs is not defined
(idb) alias cs "stop at 186; run" 
(idb) cs
[#1: stop at "x_list.cxx":186 ]
[1] stopped at [int main(void):186 0x120002420]
    186     IntNode* newNode = new IntNode(1);

The following example further modifies the cs alias to specify the breakpoint's line number when you enter the cs command:

(idb) alias cs (x) "stop at x; run" 
(idb) cs(186)
[#2: stop at "x_list.cxx":186 ]
Process has exited
[2] stopped at [int main(void):186 0x120002420]
    186     IntNode* newNode = new IntNode(1);

Note: No warning is given if the alias_name already has a definition as an alias. The old definition will be replaced by the new one.

Use the unalias command followed by an alias name to delete the specified alias.

Executing Shell Commands

You can have the debugger execute a call to the operating system's system function. This function is documented in system(3). The call results in the sh(dbx) or shell(gdb) commands.

DBX Mode
execute_shell_command
        : sh string

For example, you can execute a system command through a shell from the debugger by issuing the following command:

(idb) sh uname -m
i686
(idb)

To execute more than one command at the specified shell, spawn a shell as follows, for example:

(idb) sh csh -f
% ls out
out
% ls *.b
recio.b
stdio.b
% exit
(idb)
GDB Mode
execute_shell_command
        : shell string

For example:

(idb) shell uname -m
i686
(idb)

To execute more than one command at the shell, spawn a shell as follows:

(idb) shell bash --norc
$ ls out
out
$ ls *.b
recio.b
stdio.b
$ exit
(idb)

Invoking Your Editor (DBX Mode only)

You can use the edit command to invoke the editor defined by the EDITOR environment variable.

edit_file_command
        : edit [ string ]

The editor is given the string as the file name to edit. If no file name is specified, the editor is given the current file. If no current file exists, the editor is started without a file.

If the EDITOR environment variable is undefined, the debugger invokes the vi editor.

The following example invokes the Emacs editor on the file chars.c:

(idb) sh printenv EDITOR
emacs
(idb) file
chars.c
(idb) edit

The following example invokes the nedit editor on the file ~/foo/bar.f:

(idb) sh printenv EDITOR
nedit
(idb) edit ~/foo/bar.f

Chapter 5 — Context for Executing Commands

This chapter discusses the following topics:

Multiple Processes

The debugger supports debugging multiple processes at the same time, but at any given time is only operating on a single process, known as the current process. The debugger variable $curprocess contains the process id for this process. Naming and switching the debugger between processes is described in Multiprocess Debugging.

Creating Processes

The debugger can find and control the following:

  • Processes that you may request it to create later
  • Processes that are currently running

Specifying an executable file on the shell command line or executing the load(dbx) or file(gdb) command causes the debugger to gain control of a process that you may request it to create later.

Note: In the background, the debugger immediately creates a process executing the program, stalls it, and uses it to answer questions about which shared libraries are mapped, and so on. This process never continues, and is killed when:

  • The debugger exits.
  • You unload this executable file.

Using the run command on a loaded process causes the debugger to create a process that is identified as currently running and that it is not recreatable.

Specifying a pid on the shell command line or executing the attach command causes the debugger to know about the process as currently running and not recreatable.

Catching a fork() causes the new child process to be identified as currently running and not recreatable.

Multiple Call Frames, Threads, and Sources

Processes contain one or more threads of execution. The threads execute functions. Functions are sequences of instructions that are generated by compilers from source lines within source files.

As you enter the debugger commands to manipulate your process, it would be very tedious to have to repeatedly specify which thread, source file, and so on, you wish the command to be applied to. To prevent this, each time the debugger stops the process, it re-establishes a static context and a dynamic context for your commands. The components of the static context are independent of this run of your program; the components of the dynamic context are dependent on this run.

Some pieces of these contexts are available as debugger variables.

  • The static context consists of the following:
  • The dynamic context consists of the following:
    • Current call frame
    • Current process - $curprocess
    • Current thread - $curthread
    • The thread executing the event that caused the debugger to gain control of the process

You can switch most of these individually to point to other instances, as described in the relevant portions of this manual, and the debugger will modify the rest of the static and dynamic context to keep the various components consistent.

Chapter 6 — Running the Program Under Debugger Control

Often, running the program in a process just requires forking a process and executing the program within it with the right environment variables, argc/argv, file descriptors, and so on. This is what usually happens when you run your program from a shell command line .

However, sometimes the program requires more context, or a process may already have been created. Perhaps it is part of a pipe, perhaps it is a long-running process, or perhaps it is created from a shell script or makefile .

Hence, the following situations are possible:

  • Running your program as a child process of the debugger process.
  • Using the debugger's ability to attach to any process it has access to.

Running the Program as a Child Process

If your program has a simple command line, and only requires stdin, stdout, and stderr connected, you can run it as a child process of the debugger process. For example:

DBX Mode
% idb a.out

or

% idb
(idb) load a.out
GDB Mode
% idb -gdb a.out

or

% idb -gdb
(idb) file a.out

Attaching to a Process

If your program is any of the following, you can use the debugger's ability to attach to any process to which it has access:

  • Already running in a process
  • Has a complex command line
  • Is part of a pipe
  • Is started by a script that is difficult to modify

Examples:

DBX Mode
% idb -pid 8492 a.out

or

% idb
(idb) attach 8492 a.out
GDB Mode
% idb -gdb -pid 8492 a.out

or

% idb -gdb
(idb) file a.out
(idb) attach 8492

Once you attach to a process, the process continues execution until it raises a signal that the debugger intercepts, for example, SEGV. If you have set the $stoponattach preference variable, it stops immediately.

One method you can use to attach to a process work at a predictable location is to add a looping function to your program that keeps executing until the debugger takes control and you interrupt it. For example:

  1. Add code such as the following to your application:

    volatile int endStallForDebugger=0;
    
    void stallForDebugger()
    {
            while (!endStallForDebugger) ;
    }
    
    int main()
    {
            ...
            stallForDebugger();
            ...
    }
  2. Run this version of your program.

  3. Attach the debugger to the running process as described above.

  4. Stop the program with Ctrl/C or by use of $stoponattach.

  5. Use the debugger to assign to the stallForDebugger variable, and continue the execution of the process, so that it exits from the loop:

    (idb) assign endStallForDebugger = 1
    (idb) # set any needed breakpoints, and so on
    (idb) cont
    

The load, unload, and file Commands

Using the load(dbx) and file(gdb) commands, you specify which executable file you intend to execute iunder control of the debugger. These commands read the symbolic information for an executable file and the shared libraries it uses (if available). Objects compiled without debug information will not have symbols to load. The load(dbx) command can optionally load a core file. (This is done automatically when you give the debugger a file name on the shell command line.)

DBX Mode
load_command
        : load filename [ filename ]

The second file name is used to specify a core file. If you specify a core file, the debugger acts as though it is attached to the process at the point just before it died, except that you cannot execute commands that require a runnable process, such as commands that try to continue the process or evaluate function calls.

Examples:

% idb /usr/examples/x_list
(idb) listobj
Program is not active
(idb) load /usr/examples/x_list
Reading symbolic information ...done
(idb) listobj
    section         Start Addr           End Addr
------------------------------------------------------------------------------
/usr/examples/x_list
     .text          0x8048000          0x805adf3
     .data          0x805b000          0x806855b
      .bss          0x806855c          0x8068ad3

/lib/i686/libm.so.6
     .text         0x40026000         0x40046a02
     .data         0x40047a10         0x40047bd3
      .bss         0x40047bd4         0x40047c33

/opt/intel-compilers/cc-8.0.016/ia32/lib/libcxa.so.3
     .text         0x40049000         0x40068579
     .data         0x40069580         0x4007847f
      .bss         0x40078480         0x400785cf

/lib/i686/libc.so.6
     .text         0x42000000         0x4212bae2
     .data         0x4212c000         0x42130b63
      .bss         0x42130b64         0x42134d3f

/lib/ld-linux.so.2
     .text         0x40000000         0x40012e08
     .data         0x40013000         0x4001350b
      .bss         0x4001350c         0x400136cf

GDB Mode
file_command
        : file [ filename ]

If filename is specified, the debugger loads the specified executable. Without an argument, the debugger unloads the current executable file.

Example:

% idb -gdb /usr/examples/x_list

or:

(idb) info files
(idb) file /usr/examples/x_list
Reading symbols from /usr/examples/x_list...done.
(idb) info files
Symbols from "/usr/examples/x_list".
Unix child process:
	Using the running image of child process 5595.
	While running this, IDB does not access memory from...
Local exec file:
	'/usr/examples/x_list', file type <unknown>
	0x8048000 - 0x805adf4 is .text
	0x805b000 - 0x806855c is .data
	0x806855c - 0x8068ad4 is .bss

Loading a process both creates the debugger's knowledge of it and makes it the current process that the debugger is controlling.

The opposite of loading an executable file is unloading an executable file, when the debugger removes all related symbol table information that the debugger associated with the process being debugged.

DBX Mode
unload_command
        : unload [ pid ,... ]
        | unload [ filename ]

pid
 	: integer_constant

Process for unloading can be specified by either a process id or an executable file name.

(idb) listobj
    section         Start Addr           End Addr
------------------------------------------------------------------------------
/usr/examples/x_list
     .text          0x8048000          0x805adf3
     .data          0x805b000          0x806855b
      .bss          0x806855c          0x8068ad3

/lib/i686/libm.so.6
     .text         0x40026000         0x40046a02
     .data         0x40047a10         0x40047bd3
      .bss         0x40047bd4         0x40047c33

/opt/intel-compilers/cc-8.0.016/ia32/lib/libcxa.so.3
     .text         0x40049000         0x40068579
     .data         0x40069580         0x4007847f
      .bss         0x40078480         0x400785cf

/lib/i686/libc.so.6
     .text         0x42000000         0x4212bae2
     .data         0x4212c000         0x42130b63
      .bss         0x42130b64         0x42134d3f

/lib/ld-linux.so.2
     .text         0x40000000         0x40012e08
     .data         0x40013000         0x4001350b
      .bss         0x4001350c         0x400136cf

(idb) unload
Process has exited
(idb) listobj
Program is not active
GDB Mode

Use the file command without an argument to unload an executable file.

(idb) info files
Symbols from "/usr/examples/x_list".
Unix child process:
	Using the running image of child process 5594.
	While running this, IDB does not access memory from...
Local exec file:
	'/usr/examples/x_list', file type <unknown>
	0x8048000 - 0x805adf4 is .text
	0x805b000 - 0x806855c is .data
	0x806855c - 0x8068ad4 is .bss
(idb) file
Program terminated normally
No executable file now.
No symbol file now.
(idb) info files

The run and rerun Commands

After you have loaded a program, you can create a process executing this program using either of the following forms of the run command:

DBX Mode
run_command
        : run   [ argument_string ] [ io_redirection ... ]
        | rerun [ argument_string ] [ io_redirection ... ]

If the rerun command is specified without arguments, the arguments and io_redirection arguments of the most recent run command entered with arguments are used. If there was no previous run command, the rerun command defaults to run.

GDB Mode
run_command
        : run [ argument_string ] [ io_redirection ... ]
        | r   [ argument_string ] [ io_redirection ... ]
arg_commands
        : set_args_command
        | show_args_command

set_args_command
        : set args [ argument_string ] [ io_redirection ... ]

show_args_command
        : show args

If the run (or r) command does not specify any arguments, default arguments are used. Default arguments are specified by the previous run command with arguments or by set args command. To view default arguments, use the show args command.

Note: The set args commands does not affect process currently running. New arguments will affect only the next run.

If the last modification time or size of the binary file or any of the shared objects used by the binary file has changed since the last run or rerun(dbx) command was issued, the debugger automatically rereads the symbol table information. If this happens, the old breakpoint settings may no longer be valid after the new symbol table information is read.

The argument_string provides both the argc and argv for the created process in the same way a shell does.

The debugger breaks up the argument_string into words, and supports several shell features, including tilde (~) and environment variable expansion, wildcard substitution, single quote ('), double quote ("), and single character quote (\).

The io_redirection argument allows you to change stdin, stdout, and stderr, which are otherwise inherited from the debugger process:

io_redirection
        : <  filename
        | >  filename
        | 1> filename
        | 2> filename
        | >& filename

The various forms have the same effect as in the csh(1) shell.

Note: Although the grammar currently allows more than the following forms of redirection, you should only use the following forms because the grammar may change in a future release of the debugger.

     > filename               Redirect stdout
    1> filename               Redirect stdout
    2> filename               Redirect stderr
    >& filename               Redirect stdout and stderr
    1> filename 2> filename   Redirect stdout and stderr to different files

Examples:

DBX Mode
(idb) stop at 182
[#1: stop at "x_list.cxx":182 ]
(idb) run -s > prog.output
[1] stopped at [int main(void):182 0x804a0a0]
    182     List<Node> nodeList;
GDB Mode
(idb) break main
Breakpoint 1 at 0x804a0a0: file x_list.cxx, line 182.
(idb) show args
Argument list to give program being debugged when it is started is "".
(idb) run
Starting program: /usr/examples/x_list

Breakpoint 1, main () at x_list.cxx:182
182	    List<Node> nodeList;
(idb) continue
Continuing.
The list is:
Node 1 type is integer, value is 1
Node 2 type is compound, value is 12.345
       parent  type is integer, value is 2
Node 3 type is compound, value is 3.1415
       parent  type is integer, value is 7
Node 4 type is integer, value is 3
Node 5 type is integer, value is 4
Node 6 type is compound, value is 10.123
       parent  type is integer, value is 5


Destroying nodes...
All nodes destroyed

Program terminated normally with exit code 0
(idb) set args -s > prog.output
(idb) show args
Argument list to give program being debugged when it is started is " -s > prog.output".
(idb) run
Starting program: /usr/examples/x_list

Breakpoint 1, main () at x_list.cxx:182
182	    List<Node> nodeList;

Information: idb allows you to restart the execution of your program
from saved positions. Enter "help snapshot" for details.

The kill Command

You can kill the current process:

kill_command
        : kill

Killing a process leaves the debugger running. Any breakpoints previously set are retained. You can execute the program again later using the rerun(dbx) or the run(gdb) commands without loading it again. For example:

DBX Mode
(idb) show process
Current Process: localhost:6653 (/usr/examples/x_list) paused.
(idb) kill
Process has exited
(idb) rerun
[1] stopped at [int main(void):182 0x8049973]
    182     List<Node> nodeList;

Information: idb allows you to restart the execution of your program
from saved positions. Enter "help snapshot" for details.

GDB Mode
(idb) info program
	Using the running image of child process 5596.
Program stopped at 0x8049973.
It stopped at breakpoint 1.
(idb) kill
Program terminated normally
(idb) info program
The "info program" command has failed because there is no running program.
(idb) run
Starting program: /usr/examples/x_list

Breakpoint 1, main () at x_list.cxx:182
182	    List<Node> nodeList;

Information: idb allows you to restart the execution of your program
from saved positions. Enter "help snapshot" for details.

The attach and the detach Commands

If a process already exists, you can have the debugger attach to it:

DBX Mode
attach_command
        : attach pid [ filename ]
GDB Mode
attach_command
        : attach pid

Note: The attach command requires the name of executable to be specified before attaching to the process. Use the file command or shell command line to specify the filename.

The process is specified by its pid:

pid
        : expression

For example:

DBX Mode
(idb) attach 12345 a.out
GDB Mode
(idb) file a.out
Reading symbols from a.out...done.
(idb) attach 12345

Note that you must specify the file name. The file name should be the executable file that is executing or a duplicate copy of it. You may omit the executable file name only if the debugger already has the file loaded. It means that if a file name is not specified, the current executable is used. If the executable contains symbolic debug information it will be read by the debugger during the attach.

Attaching to a process both creates the debugger's knowledge of it and makes it the current process that the debugger is controlling. The process continues execution until it raises a signal that the debugger intercepts. Usually you do this by pressing Ctrl/C or by using the shell command kill in another window. Any other mechanism for raising a signal within the process will also do. You can set the debugger variable $stoponattach to 1 to direct the debugger to immediately stop any process that it attaches to.

(idb) ^C
Interrupt (for process)

Stopping process localhost:16077 (loop.out).
Thread received signal INT
stopped at [int main(void):3 0x120001100]
      3     while (1) ;

The opposite of attaching to a process is detaching from a process. When you detach the debugger from a process, all breakpoints are removed and the process continues to run, but the debugger can no longer identify or control it:

DBX Mode
detach_command
        : detach pid ,...

For example:

(idb) detach 12345, 789
GDB Mode
detach_command
        : detach

The detach command detaches the debugger from a current process and, therefore, does not require pid.

Controlling the Process Environment

You can set and unset environment variables for processes created in the future to set up an environment different from the environment of the debugger and from the shell from which the debugger was invoked. When set, the environment variables apply to all new processes you debug.

Note: The environment commands have no effect on the environment of any currently running process. The environment commands do not change or show the environment variables of the debugger or of the current process. They only affect the environment variables that will be used when a new process is created.

environment_variable_command
        : show_environment_variable_command
        | set_environment_variable_command
        | unset_environment_variable_command

To print either all the environment variables that are currently set or a specific one, use a show_environment_variable_command.

DBX Mode
show_environment_variable_command
        : printenv [ environment_variable_name ]
        | export
        | setenv

Note: The export and setenv commands without any arguments are equivalent.

GDB Mode
show_environment_variable_command
        : show environment [ environment_variable_name ]
        | show env         [ environment_variable_name ]

The show env is a synonym for the show environment command.

If you do not specify the name of the environment variable to show, the debugger will print all the environment variables.

To add or change an environment variable, use a set_environment_variable_command. If the environment_variable_value is not specified, the environment variable value is set to "".

GDB Mode
set_environment_variable_command
        : set environment environment_variable_name [ [ = ] environment_variable_value ]
        | set env         environment_variable_name [ [ = ] environment_variable_value ]
environment_variable_value
        : string

To remove an environment variable, use the following commands:

DBX Mode
unset_environment_variable_command
        : unsetenv environment_variable_name
        | unsetenv *

If an asterisk (*) is specified, all environment variables are removed.

GDB Mode
unset_environment_variable_command
        : unset environment environment_variable_name
        | unset env         environment_variable_name

Note: There is no command to simply return to the initial state the environment variables had when the debugger started. You must use set_environment_variable commands and unset_environment_variable commands appropriately.

For example:

DBX
(idb) printenv TOOLDIRECTORY
Error: Environment variable 'TOOLDIRECTORY' was not found in the environment.
(idb) setenv TOOLDIRECTORY /usr/examples/tools
(idb) printenv TOOLDIRECTORY
TOOLDIRECTORY=/usr/examples/tools
GDB
(idb) show environment TOOLDIRECTORY
Environment variable "TOOLDIRECTORY" not defined.
(idb) set environment TOOLDIRECTORY /usr/examples/tools
(idb) show environment TOOLDIRECTORY
TOOLDIRECTORY=/usr/examples/tools
(idb) unset environment TOOLDIRECTORY
(idb) show environment TOOLDIRECTORY
Environment variable "TOOLDIRECTORY" not defined.

Multiprocess Debugging

The debugger can find and control more than one process at a time. The debugger can find and control a process for one of the following reasons:

At any one time, you can examine or execute only one of the processes that the debugger controls. The rest are stalled. You must explicitly switch the debugger to the process you want to work with, stalling the one it was controlling:

multiprocess_command
        : show_process_command
        | switch_process_command

You can show the processes the debugger controls:

show_process_command
        : show process [ all ]
        | process

all
        : all
        | *

For example:

(idb) show process
>localhost:5351 (/usr/examples/x_list) loaded.

You can explicitly command the debugger to control a different process:

switch_process_command
        : process pid
        | process filename

The process you are switching away from remains stalled until either the debugger exits or until you switch to it and continue it.

The following example creates two processes and switches from one to the other:

(idb) process
There is no current process.
You may start one by using the `load' or `attach' commands.
(idb) load /usr/examples/x_list
Reading symbolic information ...done
(idb) process
>localhost:5352 (/usr/examples/x_list) loaded.
(idb) set $old_process = $curprocess
(idb) printf "$old_process=%d", $old_process
$old_process=5352
(idb) load /usr/examples/x_segv
Reading symbolic information ...done
(idb) process
 localhost:5352 (/usr/examples/x_list) loaded.
>localhost:5353 (/usr/examples/x_segv) loaded.
(idb) process 5352
(idb) process
>localhost:5352 (/usr/examples/x_list) loaded.
 localhost:5353 (/usr/examples/x_segv) loaded.

Both the load(dbx) command and the attach(dbx) command switch the debugger to the process on which they operate.

Processes That Use fork()

The debugger has the following predefined variables that you can set for debugging a program that forks:

  • $catchforks — When set to a non-zero value, this variable instructs the debugger to stop the child process on exit out of the fork() or vfork() calls. The parent process continues to run. The default is 0 (zero).
  • $stopparentonfork — When set to a non-zero value, this variable instructs the debugger to stop the parent process on exiting out of the fork() or vfork() calls after it forks a child process. The child process continues to run if $catchforks is 0; otherwise, it does not. The default is 0 (zero).
  • $catchforkinfork — When set to a non-zero value, this variable instructs the debugger to stay in the fork routine after the fork and notifies you as soon as the forked process is created; otherwise, you are notified when the call finishes. You can debug forking processes before any "atfork" handlers are run by setting $catchforkinfork. Because the target stops inside the system call, you will need to issue up commands to get to user-written code. The default is 0 (zero).

When a fork occurs, the debugger sets the debugger variables $childprocess and $parentprocess to the child and parent process IDs, respectively.

In the following example, the debugger notifies you that the child process has stopped. The parent process continues to run.

(idb) set $catchforks = 1
(idb) run
Process 29027 forked.  The child process is 29023.
Process 29023 stopped on fork.
stopped at [int main(void):6 0x120001178]
      6  int pid = fork();
fork.c: I am the parent.
Process has exited with status 0
(idb) show process
>localhost:29028 (/usr/examples/fork) loaded.
 localhost:29023 (/usr/examples/fork) paused.

In the preceding example, note the following:

  • The debugger indicates that the child process has stopped, and shows the line number at which it is stopped.
  • The last two lines show that the child process has stopped and that the parent process has completed execution.

Continuing the previous example, the following shows how to switch the debugger to the child process. Listing the source code shows the source for the child process.

(idb) process $childprocess
(idb) show process
 localhost:29028 (/usr/examples/fork) loaded.
>localhost:29023 (/usr/examples/fork) paused.
(idb) list
      7
      8   if (pid == 0)
      9    {
     10      printf("fork.c: I am the child.\n");
     11    }
     12   else
     13    {
     14      printf("fork.c: I am the parent.\n");
     15    }
     16 }

In the preceding example, note the following:

  • The first line switches the current process context to the child process.
  • The right angle bracket indicates the current process.
  • The list command lists the source code for the current process.

Note: If you catch the child but not the parent, and the parent code tries to execute a wait on the child, the target will get stuck if you don't let the child run to completion. This happens because the parent will be running but making no progress, and the child is stopped by the debugger. For example:

(idb) set $catchforks = 1
(idb) set $stopparentonfork = 0
(idb) list
     10     int new_pid = 0;
     11
     12     if (pid == 0) {
     13         printf( "fork.c: I am the child.\n" );
     14 	fflush( stdout );
     15
     16     } else {
     17         printf( "fork.c: I am the parent, about to wait.\n" );
     18 	fflush( stdout );
     19
     20 	new_pid = wait( &status );
     21
     22         printf( "fork.c: I am the parent, and my wait is finished\n" );
     23
     24         if (new_pid != pid )
     25             printf( "\tthere was some error\n" );
     26         else {
     27 	    if (WIFEXITED(status))
     28                 printf( "\tthe child terminated normally\n" );
     29
     30             else if (WIFSIGNALED(status))
(idb) sh cat ./x.c_fork_hang.txt
  If we 'cont' now, the process will fork; the child will be
  caught and the parent will run to the 'wait' call and wait
  for the child to terminate.

  At that time, the child will be under debugger control,
  but the current process will be the parent, which will be
  running but making no progress.  Only a Ctrl/C will allow
  further progress.

  The example program has set up another process to simulate
  a Ctrl/C by the user.  It will send SIGINT to the parent.

(idb) cont
Process 580893 forked.  The child process is 580851.
Process 580851 stopped on fork.
stopped at [void test(void):9 0x120001318]
      9     int pid = fork();
fork.c: I am the parent, about to wait.

		:
	User is waiting here
		:
		:
	Sending SIGINT to parent process
		:

Thread received signal INT
stopped at [<opaque> __wait4(...) 0x3ff800d0918]

Information:  An <opaque> type was presented during execution of the previous command.  For complete type information on this symbol, recompilation of the program will be necessary.  Consult the compiler man pages for details on producing full symbol table information using the '-g' (and '-gall' for cxx) flags.

(idb) where
>0  0x3ff800d0918 in __wait4(...) in /usr/shlib/libc.so
#1  0x3ff800d668c in __wait(...) in /usr/shlib/libc.so
#2  0x120001398 in test() "c_fork_hang.c":20
#3  0x120001528 in main() "c_fork_hang.c":71
#4  0x1200012a8 in __start(...) in /usr/examples/c_fork_hang
(idb) show process
>localhost:580893 (/usr/examples/c_fork_hang) paused.
   \_localhost:580851 (/usr/examples/c_fork_hang) paused.

Processes That Use exec()

Set $catchexecs to 1 to instruct the debugger to stop the process and notify you when an exec occurs. The process stops before executing any user program code or static initializations. You can debug the newly executed process. The debugger keeps a history of the progression of the executed files.

In the following scenario, you set the predefined variables $catchforks and $catchexecs to 1. The debugger will notify you when an execution occurs. Because $catchforks is set, you will also be tracking the child process and, therefore, you will be notified of any exec in the child process.

The following example shows an exec occurring on the current context and the child process stopped on the run-time loader entry point:

(idb) set $catchforks = 1
(idb) set $catchexecs = 1
(idb) run
Process 14839 forked.  The child process is 14835.
Process 14835 stopped on fork.
stopped at [int main(void):8 0x1200011f8]
      8   if ((pid = fork()) == 0)
x_exec.c: I am the parent.
Process has exited with status 0
(idb) show process
>localhost:14918 (x_exec) loaded.
 localhost:14835 (x_exec) paused.
(idb) process $childprocess
(idb) list 6: 13
      6   int pid;
      7
>     8   if ((pid = fork()) == 0)
      9       {
     10       printf("About to exec \n");
     11       fflush(stdout);  /* Make sure the output gets out! */
     12       execlp("announcer", "announcer", NULL);
     13       printf("After exec \n");
     14       }
     15   else
     16      {
     17       printf("x_exec.c: I am the parent.\n");
     18      }
(idb) cont
About to exec
The process 14835 has execed the image "./announcer".
Reading symbolic information ...done
stopped at [ 0x3ff8001bf48]
      5     printf("announcer.c: I am here!! \n");

Note the following:

  • Use process $childprocess to set the current process context to the child process.
  • Listing the source code, you can see the process is almost ready to execute.
  • The debugger notifies you when the exec occurs.
  • The child process is stopped on the run-time loader entry point. The source display shows the code in the main routine.

Core File Debugging

When the operating system encounters an unrecoverable error, for example, a segmentation violation (SEGV), the system creates a file named core and places it in the current directory. The core file is not an executable file; it is a snapshot of the state of your process at the time the error occurred. It allows you to analyze the process at the point it crashed. For more information on core file debugging, see Core File Debugging in Part III.

Chapter 7 — Locating the Site of a Problem

To determine why a problem is happening, you usually want to execute your program up to or just before the point at which you observe the first evidence of the problem. Then you can examine the internal state of your program and try to identify something that explains the visible problem. Possibly you will see right away how the problem occurs, in which case you are finished debugging. You then correct your program, recompile, relink, and confirm that the correction works as intended.

Often, you will see something about the program state that is wrong but you will not see how it got that way. In that case, you need to make a guess at where the mistake might have occurred. Then, repeat this whole process, trying to stop at or just before the possible trouble point.

For simple problems, it may be easy to describe the conditions under which you want to stop the program; for example, "the first time traverse is called" or "when division_by_zero occurs". Other situations may require either more complex descriptions or repeated trial-and-error attempts to discover the critical information needed to solve your problem.

Breakpoints provide the means by which you specify to the debugger an event or condition under which you want to intervene in the execution of your program and what actions you want the debugger to take when that event is detected.

You can define breakpoints based on:

  • Reaching a certain place in your program (such as entering a certain function or reaching code for a particular source line number)
  • Accessing the contents of a variable or other memory when it is either read or written
  • Raising a specified signal

You can also enable, disable, or delete breakpoints.

Breakpoint commands include the following:

breakpoint_command
        : breakpoint_definition_command
        | simple_stop_command
        | signal_command
        | obsolete_breakpoint_definition_command
        | breakpoint_table_command

This chapter discusses the following topics:

Breakpoint Definitions

The following is a particularly common breakpoint:

DBX Mode
(idb) stop in main
[#1: stop in int main(void) ]
GDB Mode
(idb) break main
Breakpoint 1 at 0x8049973: file x_list.cxx, line 182.

This command tells the debugger that when execution enters the function main, you want the debugger to suspend execution and return control to you.

The debugger responds to a breakpoint command by displaying how it recorded the request internally. The debugger assigns a number to the breakpoint (in this case, it is 1), which it uses later to refer to that breakpoint. The debugger does not just repeat the command as you entered it; it provides a more complete description of the function main to help you confirm that it has correctly identified the function you meant.

Later, after you cause the program to execute, if that event occurs, the debugger reports the event and then prompts you for what to do next. For example:

DBX Mode
(idb) run
[1] stopped at [int main(void):182 0x8049f48]
    182     List<Node> nodeList;
GDB Mode
(idb) run
Starting program: /usr/examples/x_list

Breakpoint 1, main () at x_list.cxx:182
182	    List<Node> nodeList;

Both the event part and the action part of a breakpoint definition command consist of several subparts:

breakpoint_definition_command
        : disposition
            [ quiet ]
              detector
            [ thread_filter ]
            [ logical_filter ]
            [ breakpoint_actions ]

where the detector, thread_filter (if specified), and logical_filter (if specified) collectively specify the event part, and the disposition, quiet (if specified) and breakpoint_actions (if specified) collectively specify the action part.

Note: Additional obsolete forms of breakpoint definition are retained only for backward compatibility with earlier versions of the debugger. These forms are explained later. The obsolete forms may be eliminated in a future release.

There are three distinct points in time at which a breakpoint definition has an effect:

  1. When the command is entered

    The command is parsed, names and expressions that occur in any of the event parts are evaluated, and the breakpoint actions are parsed and checked for correctness (but not evaluated).

  2. When the debugger initiates program execution

    For each breakpoint that is not disabled, appropriate modifications are made to the program to enable detection of the specified event.

  3. When a detector triggers during program execution

    The thread filter specification (if present) and logical filter (if present) are evaluated to determine whether the breakpoint as a whole has triggered. If not, then execution is resumed (silently). If so, the breakpoint actions are performed, after which execution stops or resumes according to the specified disposition.

Disposition
disposition
        : stop
        | when

The stop command specifies that when the event specified by the breakpoint occurs and all processing for that breakpoint has been completed, the debugger should prompt for further commands.

The when command specifies that when the event specified by the breakpoint occurs and all processing for that breakpoint has been completed, the debugger may resume execution of the program. See the section When Multiple Breakpoints Trigger at Once for an explanation of how the debugger determines when to resume execution.

The quiet Specifier

By default, when an event is detected and the debugger determines that the breakpoint actions should be performed, the debugger prints a line that identifies the breakpoint, for example:

(idb) when in main { stop }
[#1: when in int main(void) { stop } ]
(idb) run
[1] when [int main(void):182 0x8049f48]
[1] stopped at [int main(void):182 0x8049f48]
    182     List<Node> nodeList;

The optional quiet specifier tells the debugger to omit this information.

(idb) when quiet in main { stop }
[#11: when quiet in int main(void) { stop } ]
(idb) run
(idb) list $curline: 1
>   182     List<Node> nodeList;
Detectors

The debugger uses several kinds of detectors, each corresponding to a particular kind of event:

A place detector specifies a place or location in your program. It can refer to the beginning of a function, a particular line in one of your source files, a specific value of the PC (program counter), or certain sets of these.

A watch detector specifies a variable or other memory locations that should be monitored to detect certain kinds of access (read, write, and so on).

A signal detector specifies a set of signals to be monitored.

An unaligned access detector specifies any kind of memory access using an unaligned access.

This section describes each type of detector.

Place Detectors

You can use place detectors to determine when execution reaches a particular place or location in your program:

The in function_name detector specifies the event where execution reaches the entry of the named function. For example:

If the function name is ambiguous (more than one function can match the name in some languages, including C++), the debugger prompts you with a list of alternatives from which to choose.

(idb) stop in foo
Select from
----------------------------------------------------
     1 int C::foo(double*)
     2 void C::foo(float)
     3 void C::foo(int)
     4 void C::foo(void)
     5 None of the above
----------------------------------------------------
2
[#4: stop in void C::foo(float) ]

If you choose the last option ("None of the above"), then no function is selected and no breakpoint is defined.

The in all function_name detector is the same as in function_name except that it specifies all of the functions that match the given name, whether one or more:

(idb) stop in all foo
[#3: stop in all foo ]

The pc address_expression detector specifies the event where execution reaches the given machine address:

(idb) stop pc $pc + 8
[#7: stop PC == 0x804a539 ]

The at line_specifier detector specifies the event where code associated with a particular line of the source is reached:

(idb) stop at 190
[#8: stop at "x_list.cxx":190 ]

If no code is associated with the given line number, the debugger finds and substitutes the closest higher line number that has associated code.

The every procedure entry detector specifies that a breakpoint should be established for every function entry point in the program.

(idb) stop every procedure entry
[#9: stop every procedure entry ]

Note: This command can be very time consuming because it searches your entire program — including all shared libraries that it references — and establishes breakpoints for every entry point in every executable image. This can also considerably slow execution of your program as it runs.

A disadvantage of this command is that it establishes breakpoints for hundreds or even thousands of entry points about which you have little or no information. For example, if you use stop every proc entry immediately after loading a program and then run it, the debugger will stop or trace over 100 entry points before reaching your main entry point. About the only thing that you can do if execution stops at most such unknown places is continue until some function relevant to your debugging is reached.

The every instruction detector specifies a breakpoint for every instruction in your entire program:

(idb) stop every instruction
[#10: stop every instruction ]

When used with the stop disposition, a subsequent continue behaves essentially the same as a step by instruction command (see stepi).

When used with the when disposition, subsequent next and step commands allow you to trace all of the instructions that are executed as a result of those stepping commands. Beware that even when next is used to step over a called routine, the trace output includes all of the instructions that are executed within the called routine (and any routines that it calls).

Note: This command will slow execution of your program considerably.

The detector expression (that is, an expression not preceded by one of the keywords in, at, or pc) specifies either a function name or line number depending on how the expression is parsed and evaluated. An expression that evaluates to the name of a function is handled just like the equivalent command that uses in in the detector; otherwise, it is handled like the equivalent command that uses at in the detector.

Watch Detectors

You can use watch detectors to determine when a variable or other memory location is read or written and/or changed. Breakpoints with watch detectors are also known as watchpoints.

watch_detector
        : basic_watch_detector watch_detector_modifiers

basic_watch_detector
        : variable expression
        | memory start_address_expression
        | memory start_address_expression , end_address_expression
        | memory start_address_expression : byte_count_expression

watch_detector_modifiers
        : [ access_modifier ] [ within_modifier ]

access_modifier
        : write
        | read
        | changed
        | any

within_modifier
        : within function_name

You can specify a variable whose memory is to be watched, or specify the memory directly. The accesses that are considered can be limited to those that write (the default), read, write and actually change the value, or can include all accesses.

If you specify a variable, the memory to be watched includes all of the memory for that variable, as determined by the variable's type. The following example watches for write access to variable _nextNode, which is allocated in the 8 bytes at the address shown in the last line of the example:

(idb) whatis _nextNode
struct Node* Node::_nextNode
(idb) print "sizeof(_nextNode) =", sizeof((_nextNode))
sizeof(_nextNode) = 4
(idb) stop variable _nextNode write
[#3: stop variable _nextNode write ]

The specified variable is watched. If "p" is a pointer, watch variable p will watch the content of the pointer, not the memory pointed to by "p". Use watch memory *p to watch the memory pointed to by "p".

If you specify memory directly in terms of its address, the memory to be watched is defined as follows:

  • By default (no last address or size given), then 8 bytes beginning at the given start address:

    (idb) when memory &_nextNode : 8 any
    [#4: when memory &_nextNode : 8 any ]
    
  • If an end address is given, then all bytes of memory from the start address to and including the end address:

    (idb) stop memory &_nextNode, ((long)&_nextNode) + 3 read
    [#5: stop memory &_nextNode, ((long)&_nextNode) + 3 read ]
    

    This watches the 4 bytes specified on the command line.

  • If you specify a byte count, then the given number of bytes starting at the given start address:

    (idb) stop memory &_nextNode : 2 changed
    [#6: stop memory &_nextNode : 2 changed ]
    

    This watches the 2 bytes specified on the command line for a change in contents.

If you specify the within modifier, then only those accesses that occur within the given function (but not any function it calls) are watched. For example:

(idb) whatis t
int t
(idb) stop variable t write within foo
[#2: stop variable t write within void C::foo(void) ]
(idb) cont
[2] Address 0x804bbfc was accessed at:
void C::foo(void): x_overload.cxx
 [line 22, 0x8048af3]	foo+0x6:              movl     $0x0, 0x804bbfc
	0x804bbfc: Old value = 0x0000000f
	0x804bbfc: New value = 0x00000000
[2] stopped at [void C::foo(void):22 0x8048afd]
     22 void C::foo()         { t = 0; state++; return; }
Signal Detectors

You can use signal detectors to determine when a particular signal is raised:

signal_detector
        : signal signal_id ,...

signal_id
        : integer_constant
        | signal_name

You can specify signals by numeric value or by their conventional operating system names, without or without the leading "SIG":

(idb) stop signal SEGV, 3, SIGINT
[#2: stop signal SEGV, 3, SIGINT ]

If the debugger catches a signal event, then a subsequent simple continue will resume execution without raising the signal again in your process. However, a signal can be specified as part of the continue command to send the signal to your process when it resumes.

Unaligned Access Detectors

You can use an unaligned access detector to determine when an unaligned memory access occurs:

unaligned_detector
        : unaligned

Unaligned accesses may be automatically handled by the operating system. By default, an unaligned access results in an information message and then is corrected so that your program can continue. (You or your system administrator can choose a different default. See uac(1) for more information.) This message looks like this:

Unaligned access pid=30231 <x_signals> va=0x11ffff791 pc=0x120001af4 ra=0x120001b84 inst=0xa0220000

You can request the debugger to detect unaligned accesses:

(idb) stop unaligned access
[#1: stop unaligned access ]
(idb) run
Thread encountered Unaligned Access
[1] stopped at [int unalignedAccess(void):27 0x120001af8]
     27     return y;
Unaligned Access Detector (Linux* Only)

Unaligned accesses are automatically handled and quietly corrected on Linux. The debugger cannot detect these events.

Thread Filter

A thread filter determines whether a detected event should be further considered for breakpoint processing.

thread_filter
        : thread thread_id ,...

The thread_id expressions are evaluated at the time the breakpoint command is entered, and each must yield an integer value.

A detected event is retained for further consideration only if the thread in which the event occurs matches one of the given threads. If not, the detection is quietly ignored.

If the thread_filter does not indicate a match, then any related logical filter is not evaluated.

Logical Filter

A logical filter determines whether a detected event should be further considered for breakpoint processing:

logical_filter
        : if expression

A detected event is retained for further consideration only if the given expression evaluates to true. If not, the detection is quietly ignored.

The expression is checked syntactically in the context of the place where the breakpoint command is given: it must be syntactically valid according to the language rules that apply there. However, the expression is not evaluated and names that occur in the expression need not be visible. After the syntax check, the expression is remembered in an internal form and is not rechecked later when it is evaluated.

If an error occurs when the expression is evaluated, for example, because a name in the expression is not defined, then the error is reported and the value of the expression is assumed to be true.

An error in the expression does not change the disposition. If continuation was specified, then that is still what occurs. For example:

(idb) when in List<Node>::append if x
[#5: when in void List<Node>::append(struct Node* const) if x ]
(idb) cont
Symbol "x" is not defined.
[Error while evaluating breakpoint condition - taken as true]
[5] when [void List<Node>::append(struct Node* const):148 0x8058026]
Symbol "x" is not defined.
[Error while evaluating breakpoint condition - taken as true]
[5] when [void List<Node>::append(struct Node* const):148 0x8058026]
[4] stopped at [int main(void):195 0x804a1ad]
    195     nodeList.append(new IntNode(3));

It is valid for a logical filter expression to contain a call to another routine in your program. Such a call is evaluated in the same way as if it occurred in a call or print command. However, execution of the called routine might result in triggering a breakpoint; this is called a recursive breakpoint.

Breakpoint Actions

The action part of a breakpoint command specifies actions to be performed when the event part has triggered (including passing any thread and/or logical filters):

breakpoint_actions
        : { action_list }

action_list
        : command
        | command ;
        | command ;...
Special Commands

The following debugger commands behave differently in some fashion when used within a breakpoint action list:

  • Simple stop

    A simple_stop_command is a stop without any detector or other parameters:

    simple_stop_command
            : stop
    

    If used within a breakpoint action list, it specifies that the disposition for the breakpoint should be to stop after completion of action list processing, even if the breakpoint was specified with the when disposition. If used outside an action list, it has no effect.

    A simple stop command does not terminate action list processing; it only affects the disposition that applies later. For example:

    (idb) when in List<Node>::print { stop ; print "*** stopped ***"}
    [#6: when in void List<Node>::print(void) { stop ; print "*** stopped ***"} ]
    (idb) cont
    [6] when [void List<Node>::print(void):162 0x80580ae]
    *** stopped ***
    [6] stopped at [void List<Node>::print(void):162 0x80580ae]
        162     Node* currentNode = _firstNode;
    
  • History

    The history command does not display commands that are performed as part of the action list of a breakpoint.

Commands to Use with Caution

You must be very careful when using some commands in breakpoint action lists. The following commands cause the debugger to resume execution of your program in the midst of action list processing:

  • call
  • continue
  • goto
  • next
  • return
  • step
  • Any command that contains an expression whose evaluation involves calling a function in your program

It is easy in such cases to lose track of just what state breakpoint processing is really in or where you really are in your program. Such confusion may mislead or misdirect your debugging effort. For further discussion, see the section on Recursive breakpoints.

Commands to Avoid

You should avoid altogether some commands in breakpoint action lists. The following are commands that directly or indirectly change the process that the debugger is controlling:

  • attach and detach
  • run and rerun
  • process with an argument

The debugger does not explicitly prohibit these commands, but their behavior within action lists is implementation-defined and subject to change from release to release. In very specialized cases, you may be able to obtain useful results by using them in action lists, but do not expect the same behavior over the long term.

When Multiple Breakpoints Trigger at Once

It is possible for multiple breakpoints to specify the same event, or possibly overlapping events. Thus, more that one breakpoint detector may trigger at the same time.

When more than one breakpoint detector triggers, the thread filters and logical filters of all the breakpoints involved are processed before the action part of any breakpoint is performed.

After the set of breakpoints that trigger is determined, the action parts of each of them are performed in an undefined order.

After all action parts are performed, execution of the program is resumed only if all of the breakpoints so specify in their disposition. If any one of them specifies a break, the debugger prompts you for further commands.

Recursive Breakpoints

The following commands cause the debugger to resume execution of your program while in the midst of action list processing:

  • call
  • continue
  • goto
  • next
  • return
  • step
  • Any command that contains an expression whose evaluation involves calling a function in your program

In all of these cases, the debugger temporarily suspends processing of the current breakpoint to start your program executing again and then waits for that execution to complete. As long as no new breakpoint is triggered during that execution, all will be fine. However, if a new breakpoint triggers, in particular one with the stop disposition, then you may be prompted for new command input for the recursive breakpoint even before the initial breakpoint has completed. Further, continuing execution may ultimately allow the original breakpoint to complete, at which time its disposition will come into play.

It is easy in such cases to lose track of just what state breakpoint processing is really in or where you really are in your program. Such confusion may mislead or misdirect your debugging effort. See the call command example, which locates suspended execution in nested function calls.

Breakpoints and C++

This section describes how to use breakpoints when debugging C++ programs.

Member Functions

Setting breakpoints in C++ member functions is illustrated using the following program:

(idb) list 3: 25
      3 class C {
      4 public:
      5     void foo();
      6     void foo(int);
      7     void foo(float);
      8     int  foo(double *);
      9 };
     10
     11 C  o;
     12 C* p = new C;
     13 int t     = 0;
     14 int state = 1;
     15
     16 main(){
     17     t++;
     18     o.foo();
     19
     20 }
     21
     22 void C::foo()         { t = 0; state++; return; }
     23 void C::foo(int i)    { state++; return;  }
     24 void C::foo(float f)  { state++; return;  }
     25 int  C::foo(double *) { return state;}

You must name member functions in a way that makes them visible at the current position, according to the normal C++ visibility rules. For example:

(idb) stop in main
[#1: stop in int main(void) ]
(idb) run
[1] stopped at [int main(void):17 0x8048ad3]
     17     t++;
(idb) stop in foo
Symbol "foo" is not defined.
foo has no valid breakpoint address
Warning: Breakpoint not set

If not positioned within a member function of a class, it is generally necessary to name the desired member function using type qualification, an object of the class type, or a pointer to an object of the class type. For example:

(idb) stop in C::foo
Select from
----------------------------------------------------
     1 int C::foo(double*)
     2 void C::foo(float)
     3 void C::foo(int)
     4 void C::foo(void)
     5 None of the above
----------------------------------------------------
3
[#5: stop in void C::foo(int) ]
(idb) stop in o.foo
Select from
----------------------------------------------------
     1 int C::foo(double*)
     2 void C::foo(float)
     3 void C::foo(int)
     4 void C::foo(void)
     5 None of the above
----------------------------------------------------
1
[#6: stop in int C::foo(double*) ]
(idb) stop in p->foo
Select from
----------------------------------------------------
     1 int C::foo(double*)
     2 void C::foo(float)
     3 void C::foo(int)
     4 void C::foo(void)
     5 None of the above
----------------------------------------------------
4
[#7: stop in void C::foo(void) ]

You can avoid the ambiguity associated with an overloaded function by specifying a complete signature for the function name. For example:

(idb) stop in C::foo(void)
[#8: stop in void C::foo(void) ]
(idb) stop in C::foo(int)
[#9: stop in void C::foo(int) ]
Templates and Instantiations

The debugger has no knowledge of templates that may occur in your program. However, you can usually debug template instantiations the same way as the equivalent non-instantiated class or function.

Debugging of template instantiations is illustrated using the following source text:

(idb) list 144: 13
    144 template <class NODETYPE>
    145 void List<NODETYPE>::append(NODETYPE* const node)
    146 {
    147
    148     if (!_firstNode)
    149         _firstNode = node;
    150     else {
    151         Node* currentNode = _firstNode;
    152         while (currentNode->getNextNode())
    153             currentNode = currentNode->getNextNode();
    154 	currentNode->setNextNode(node);
    155     }
    156 }

Normal debugging commands then apply to the instantiation (not the template as such):

(idb) whatis List<Node>::append
void List<Node>::append(class Node* const)
(idb) stop in List<Node>::append
[#1: stop in void List<Node>::append(class Node* const) ]
(idb) run
[1] stopped at [void List<Node>::append(class Node* const):148 0x120001d9c]
    148     if (!_firstNode)
(idb) where 2
>0  0x120001d9c in ((List<Node>*)0x11fffee68)->List<Node>::append(node=0x140002c00) "x_list.cxx":148
#1  0x1200024a4 in main() "x_list.cxx":187
Exception Handlers

When working with exception handlers, you can set a breakpoint at the appropriate line to determine if an exception is thrown. In addition, you can set breakpoints in these functions that are part of the C++ library support for exceptions:

terminate
Gains control when any unhandled exception occurs, which will result in program termination.
unexpected
Gains control when a function containing an exception specification tries to throw an exception that is not included in that specification.

These special library functions are illustrated using the following source:

(idb) list 30: 29
     30 // Throw an exception.  The "throw(int)" syntax tells the compiler that
     31 // only integer exceptions can escape this method.  This will result in
     32 // an unexpected exception from C++.
     33 //
     34 void throwAnException() throw(int)
     35 {
     36     throw "Bug";
     37 }
     38
     39 // Provide some depth to the stack, for demonstration purposes
     40 //
     41 void someOperation()
     42 {
     43     int z = unalignedAccess();  // Some tests ignore this exception
     44     throwAnException();
     45 }
     46
     47 main()
     48 {
     49     try {
     50         someOperation();
     51     }
     52     catch(char* str) {
     53         std::cout << "Caught exception [" << str << "]" << std::endl;
     54     }
     55     catch(...) {
     56         std::cout << "Caught something" << std::endl;
     57     }
     58 }

You can trace the flow of execution, as in the following:

(idb) stop at 52
[#1: stop at "x_signals.cxx":52 ]
(idb) stop in all terminate
Symbol "terminate" is not defined.
No value for expression terminate
Warning: Breakpoint not set
(idb) stop in all unexpected
Symbol "unexpected" is not defined.
No value for expression unexpected
Warning: Breakpoint not set
(idb) run
Caught something
Process has exited with status 0
(idb) where
The "where" command has failed because there is no running program.
(idb) cont
The "cont" command has failed because there is no running program.
(idb) where
The "where" command has failed because there is no running program.
(idb) cont
The "cont" command has failed because there is no running program.
(idb) where
The "where" command has failed because there is no running program.
(idb) cont
The "cont" command has failed because there is no running program.
Special Signal Breakpoints

Signals are operating-system-defined events that can be handled by the debugger.

The catch and ignore Commands

You can use two special breakpoint commands, catch and ignore, to handle signal events:

signal_command
        : catch_command
        | ignore_command

catch_command
        : catch [ signal_id ]

ignore_command
        : ignore [ signal_id ]

A catch command with an operand specifies that the debugger should catch and handle the given signal. You can specify the signal by integer number or by standard signal name, with or without the leading "SIG". The catch command is equivalent to the breakpoint command:

(idb) catch BUS

or

(idb) stop signal SIGBUS
[#1: stop signal SIGBUS ]

with these exceptions:

  • No entry is made in the breakpoint table for a catch command.
  • A catch for a signal that is already being caught does not create an additional breakpoint for that signal.

An ignore command with an operand specifies that the given signal should not be caught or handled by the debugger; rather, such a signal is passed to your program. The ignore command is equivalent to deleting the breakpoint created by a catch command for that signal.:

(idb) ignore BUS

A catch command without an operand lists all signals that are currently being handled. Similarly, an ignore command without an operand lists the signals that are currently being ignored. Together, the two lists show all signals known to the debugger.

You can issue these commands immediately after the debugger starts to show which signals are caught and which are ignored by default:

(idb) catch
INT, ILL, ABRT, FPE, SEGV, TERM, QUIT, TRAP, BUS, SIGSYS, PIPE, URG, STOP, TTIN, TTOU, XCPU, XFSZ, PROF, USR1, USR2, VTALRM
(idb) ignore
RTMIN, RTMIN1, RTMIN2, RTMIN3, RTMIN4, RTMIN5, RTMIN6, RTMIN7, RTMAX, RTMAX7, RTMAX6, RTMAX5, RTMAX4, RTMAX3, RTMAX2, RTMAX1, HUP, KILL, ALRM, TSTP, CONT, CLD, WINCH, POLL
Unaligned Accesses

You can request the debugger to catch unaligned accesses:

(idb) catch unaligned

This command is very much like the stop unaligned command:

Although this looks like a normal catch command, it differs in several respects:

  • unaligned is not the name of a signal.
  • There is no corresponding signal number.
  • unaligned is never listed by either the catch or ignore commands without an argument.

Like other catch commands, the following rules apply:

  • No entry is made in the breakpoint table for a catch command.
  • Repeating the command does not create an additional breakpoint.

Note: You cannot specify unaligned in a signal detector of a normal breakpoint definition.

You can request the debugger to ignore unaligned accesses when catch unaligned is in effect (the default) by using the following command:

(idb) ignore unaligned

However, if a breakpoint was defined using an unaligned access detector, then it must be disabled using a disable or delete breakpoint command.

Unaligned Accesses (Linux Only)

Unaligned accesses are automatically handled and quietly corrected on Linux. The debugger cannot catch these events.

Ctrl/C

If your program seems to be caught in a loop, you can press Ctrl/C. The debugger interprets this as a command to send a signal interrupt (SIGINT) to your program. Because the debugger itself catches signal SIGINT by default, this interrupts your program and returns control to the debugger prompt.

If you give the command ignore SIGINT, then it is no longer possible to regain control of your program using Ctrl/C. In that case, signal SIGINT is delivered directly to your program. Unless your program has explicitly arranged otherwise, SIGINT will result in program termination.

Breakpoint Interactions with exec(), fork(), dlopen() and dlclose() System Calls

A process starts with a copy of its parent's memory as the result of a fork() system call; after running for a while within that memory, the process will often make an exec() system call to start a new executable file within that process.

The debugger keeps track of the exec() calls that occur so that it can keep track of various properties associated with each executable file. In particular, the breakpoint table is one of those properties. Thus, if you run or rerun your program, the same breakpoints can be re-established, even though a new process is initiated. Similarly, if you work with more than one process, each process has a distinct breakpoint table associated with it.

When a dlopen() system call occurs, the debugger reprocesses the current breakpoint table and automatically sets up the means to detect any events that apply to the newly loaded image.

When a dlclose() system call occurs, the debugger also reprocesses the breakpoint and de-activates any events that apply to the unloaded image.

Obsolete Breakpoint Commands (DBX Mode only)

The following forms of breakpoint commands are obsolete, but are still supported for backward compatibilty with earlier versions of the debugger:

obsolete_breakpoint_definition_command
        : obsolete_watch_breakpoint_definition_command
        | obsolete_trace_breakpoint_definition_command
        | obsolete_stopi_breakpoint_definition_command
        | obsolete_wheni_breakpoint_definition_command
        | obsolete_tracei_breakpoint_definition_command
Obsolete Watchpoint Definition

An obsolete watchpoint definition is similar to a stop variable or stop memory breakpoint:

obsolete_watch_breakpoint_definition_command
        : watch obsolete_watch_detector
            [ obsolete_watch_modifiers ]
            [ breakpoint_actions ]

obsolete_watch_detector
        : variable variable_name
        | [ memory ] start_address_expression
        | [ memory ] start_address_expression , end_address_expression
        | [ memory ] start_address_expression : byte_count_expression

obsolete_watch_modifiers
        : [ access_modifier ]
          [ thread_filter ]
          [ within_modifier ]
          [ logical_filter ]

An obsolete watchpoint and a stop command differ in the following respects:

  • The obsolete watchpoint command begins with watch instead of stop.
  • The keyword memory is optional; if omitted, it is assumed.
  • The order of filters and modifiers is different.

These differences are purely syntactic; the semantics are the same.

(idb) watch variable _firstNode write
[#3: watch variable _firstNode write ]
(idb) cont
[3] Address 0xbffff0fc was accessed at:
void List<Node>::append(struct Node* const): x_list.cxx
 [line 149, 0x8058035]	append(struct Node* const)+0x15:         movl     %edx, (%eax)
	0xbffff0fc: Old value = 0x00000000
	0xbffff0fc: New value = 0x080b0f40
[3] stopped at [void List<Node>::append(struct Node* const):149 0x8058037]
    149         _firstNode = node;
Obsolete Tracepoint Definition

An obsolete tracepoint definition is similar to a when in or when at breakpoint, possibly combined with watching for a change of a variable's value:

obsolete_trace_breakpoint_definition_command
        : trace [ variable_name ]
            [ thread_filter ]
            [ where_modifier ]
            [ logical_filter ]
            [ breakpoint_actions ]
        | trace function_name [ logical_filter ] [ breakpoint_actions ]
        | trace line_specifier [ logical_filter ] [ breakpoint_actions ]

where_modifier
        : in function_name
        | at line_specifier

line_specifier
        : quoted_filename:line_number
        | line_number

quoted_filename
        : "filename"
        | 'filename'

Following are the differences between an obsolete tracepoint and a when command:

  • The obsolete tracepoint command begins with trace instead of when.

  • If you specify a variable name, a trace identification line is displayed only when the value of the variable changes (and the logical filter evaluates to true).

    The debugger implementation of trace for detecting variable changes tends to be slow — at each place where control might be stopped, as specified by the where modifier and filters, the value of the variable is compared to the value remembered at the time execution began.

  • The order of filters and modifiers is different.

For example:

(idb) trace in List<Node>::print
[#7: trace in void List<Node>::print(void) ]
(idb) trace i in List<Node>::print
[#8: trace i in void List<Node>::print(void) ]
(idb) trace List<Node>::print if i { print "Test 1"}
[#9: trace in void List<Node>::print(void) if i { print "Test 1"} ]

If the trace command is given with no arguments, the debugger prints a trace identification line when each function in your program is entered. For example:

(idb) trace
[#10: trace ]
(idb) status
#10 at procedure entry { trace-proc }

This is equivalent to the when every proc entry command (with equivalent performance degradation).

Instruction-Related Breakpoint Commands

The following commands control obsolete instruction-related breakpoints:

obsolete_stopi_breakpoint_definition_command
        : stopi [ expression ]
            [ thread_filter ] [ match_address ] [ logical_filter ]

obsolete_tracei_breakpoint_definition_command
        : tracei [ expression ]
            [ thread_filter ] [ match_address ] [ logical_filter ]

obsolete_wheni_breakpoint_definition_command
        : wheni [ expression ]
            [ thread_filter ] [ match_address ] [ logical_filter ]
            breakpoint_actions

match_address
        : at address_expression

The stopi, tracei, and wheni forms of breakpoint definition are similar to the corresponding stop, trace, and when forms, with the following differences:

  • They have a different initial keyword.

  • If you specify a variable name, then breakpoint triggers only when the value of the variable changes (and the logical and thread filters are true).

    The debugger implementation of tracei for detecting variable changes tends to be slow: at each place where control might be stopped, as specified by the where modifier and filters, the value of the variable is compared to the value remembered at the time execution began.

    Most important, the variable change and filter tests are performed after every instruction is executed, making these definitions especially demanding on program performance.

  • The order of filters and modifiers is different.

  • The at keyword is followed by an address in these commands, instead of a line number.

Breakpoint Tables

As breakpoints are defined, they are recorded in a breakpoint table associated with the current program. You can display and modify this table in certain limited ways.

breakpoint_table_command
        : show_all_breakpoints_command
        | delete_breakpoint_command
        | enable_breakpoint_command
        | disable_breakpoint_command

Each entry in the breakpoint table has the following properties:

  • A unique breakpoint number that is used to identify and refer to that breakpoint.
  • An event description that characterizes the circumstances under which the breakpoint triggers.
  • Actions (a possibly empty list of debugger commands) to be performed when the breakpoint triggers.
  • A final disposition: either continue or break (stop).
  • Enabled and disabled states.

In addition to the main effects of a breakpoint definition, as discussed in Breakpoint Definitions, a breakpoint definition also sets the debugger variable $lasteventmade to the breakpoint number of the breakpoint just defined. This value can be recalled for later use if desired. For example:

(idb) stop in List<Node>::append
[#2: stop in void List<Node>::append(struct Node* const) ]
(idb) cont
[2] stopped at [void List<Node>::append(struct Node* const):148 0x8058026]
    148     if (!_firstNode)
(idb) print $lasteventmade
2
(idb) set $my_break = $lasteventmade
(idb) print $my_break
2

If an error occurs in a breakpoint command, the variable $lasteventmade is not changed.

Showing Breakpoint Status

Use the following commands to display the current breakpoint table:

DBX Mode
show_all_breakpoints_command
        : status
GDB Mode
show_all_breakpoints_command
        : info breakpoints [ expression ]
        | info watchpoints [ expression ]
        | info break       [ expression ]
        | info b           [ expression ]
        | i    breakpoints [ expression ]
        | i    watchpoints [ expression ]
        | i    break       [ expression ]
        | i    b           [ expression ]

All these commands are synonyms.

Specify breakpoint number to print information about particular breakpoint. If you do not specify an argument, the debugger prints information about all breakpoints.

Each entry in the current breakpoint table is displayed showing all of its properties. For example:

DBX Mode
(idb) status
#1 PC==0x8049f48 in int main(void) "x_list.cxx":182 { stop }
#2 PC==0x8058026 in void List<Node>::append(struct Node* const) "x_list.cxx":148 { break }
#3 Access memory (write) 0xbffff0fc to 0xbffff0ff { stop }
GDB Mode
(idb) info breakpoints
Num Type           Disp Enb Address    What
1   breakpoint     keep y   0x804a0a8  in main at x_list.cxx:182
2   breakpoint     keep y   0x804aa52  in List<Node>::append(class Node * const) at x_list.cxx:148
3   watchpoint     keep y              _firstNode, { Access memory (changed) 0xbfffe94c to 0xbfffe94f }

When an entry in the current breakpoint table references a shared object that is not currently mapped, it's contribution to the What column indicates Not Currently Mapped.

When large or complex values are passed by value to the routine in the status line, the output can be voluminous. You can set the control variable $statusargs to 0 to suppress the output of argument type information in the status line.

Enabling, Disabling, and Deleting Breakpoints

When a breakpoint is defined, it is enabled by default. When the debugger starts or resumes process execution, it first adapts the process so that it can detect when the given events occur. A breakpoint can be disabled so it is not involved in determining when the process should next stop. A breakpoint that is no longer required can be deleted entirely.

DBX Mode
disable_breakpoint_command
        : disable all
        | disable breakpoint_number_expression ,...

enable_breakpoint_command
        : enable all
        | enable breakpoint_number_expression ,...

delete_breakpoint_command
        : delete all
        | delete breakpoint_number_expression ,...
GDB Mode
disable_breakpoint_command
        : disable [ breakpoints ] [ bpnums ]
        | dis     [ breakpoints ] [ bpnums ]
enable_breakpoint_command
        : enable [ breakpoints ] [ bpnums ]
delete_breakpoint_command
        : delete [ breakpoints ] [ bpnums ]
        | d      [ breakpoints ] [ bpnums ]
bpnums
        : bpnum ...

bpnum
        : integer

You can specify one or more breakpoint numbers to disable, enable or delete. If you do not specify any arguments, the debugger disables, enables or deletes all the breakpoints.

For example:

DBX Mode
(idb) disable 1
(idb) status
#1 PC==0x8049f48 in int main(void) "x_list.cxx":182 { stop } Disabled
#2 PC==0x8058026 in void List<Node>::append(struct Node* const) "x_list.cxx":148 { break }
#3 Access memory (write) 0xbffff0fc to 0xbffff0ff { stop }
(idb) disable 10 - 8,1 + 1 + 1
(idb) status
#1 PC==0x8049f48 in int main(void) "x_list.cxx":182 { stop } Disabled
#2 PC==0x8058026 in void List<Node>::append(struct Node* const) "x_list.cxx":148 { break } Disabled
#3 Access memory (write) 0xbffff0fc to 0xbffff0ff { stop } Disabled
(idb) delete 1
(idb) status
#2 PC==0x8058026 in void List<Node>::append(struct Node* const) "x_list.cxx":148 { break } Disabled
#3 Access memory (write) 0xbffff0fc to 0xbffff0ff { stop } Disabled
(idb) enable all
(idb) status
#2 PC==0x8058026 in void List<Node>::append(struct Node* const) "x_list.cxx":148 { break }
#3 Access memory (write) 0xbffff0fc to 0xbffff0ff { stop }
GDB Mode
(idb) disable 1
(idb) info breakpoints
Num Type           Disp Enb Address    What
1   breakpoint     keep n   0x804a0a8  in main at x_list.cxx:182
2   breakpoint     keep y   0x804aa52  in List<Node>::append(class Node * const) at x_list.cxx:148
3   watchpoint     keep y              _firstNode, { Access memory (changed) 0xbfffe94c to 0xbfffe94f }
(idb) disable 3,2
(idb) info breakpoints
Num Type           Disp Enb Address    What
1   breakpoint     keep n   0x804a0a8  in main at x_list.cxx:182
2   breakpoint     keep n   0x804aa52  in List<Node>::append(class Node * const) at x_list.cxx:148
3   watchpoint     keep n              _firstNode, { Access memory (changed) 0xbfffe94c to 0xbfffe94f }
(idb) delete 1
(idb) info breakpoints
Num Type           Disp Enb Address    What
2   breakpoint     keep n   0x804aa52  in List<Node>::append(class Node * const) at x_list.cxx:148
3   watchpoint     keep n              _firstNode, { Access memory (changed) 0xbfffe94c to 0xbfffe94f }
(idb) enable
(idb) info breakpoints
Num Type           Disp Enb Address    What
2   breakpoint     keep y   0x804aa52  in List<Node>::append(class Node * const) at x_list.cxx:148
3   watchpoint     keep y              _firstNode, { Access memory (changed) 0xbfffe94c to 0xbfffe94f }

Chapter 8 — Looking Around at the Code, the Data, and Other Process Information

This chapter describes how to look at the following components of a running process:

Looking at the Source Files

The debugger supports commands to perform the following operations with source files:

  • Determine the location of the source files
  • Select a particular file as the current file
  • List portions of the current file
  • Search through the current file for target strings
browse_source_command
        : source_directory_mapping_command
        | source_searchlist_command
        | select_source_file_command
        | list_source_file_command
        | search_source_file_command

Special debugging information that the compiler puts in the .o files correlates the machine instructions and data back to the source files and the positions they came from.

Source files are compiled and linked into executable files. During debugging, the debugger tries to find these source files to display them for you. If the source files have moved, or if the paths to them are relative, the debugger may not be able to locate them. All the information the debugger needs comes from the executable files or shared libraries, not from the source files.

How the Debugger Finds Source Files

The debugger searches for a source file (dir_name/base_name) using the following algorithm:

  1. If dir_name is mapped to another source directory (mapped_dir_name), look for mapped_dir_name/base_name.
  2. If Step 1 fails to find a readable file:
    Case 1: If dir_name is absolute, look for dir_name/base_name.
    Case 2: If dir_name is relative, for each entry use_dir in use_list, look for use_dir/dir_name/base_name. Note that the use_list entries are tried in the order they appear in the use_list.
  3. If Step 2 fails, for each entry use_dir in use_list, look for use_dir/base_name. Just as in Step 2, the use_list entries are tried in the order they appear in the use_list.
  4. If Step 3 fails, the debugger cannot find any source file.

The debugger uses the first-found readable file as the source file.

The debugger has source directory mapping commands that:

  • Inform you in which directories the debugger is looking for the source files.
  • Allow you to designate directories in which the debugger will look for the source files.

The following example shows how to use source directory mapping. Suppose you compile x_solarSystem as follows:

% pwd
/usr/users/debugger/sandbox/test/src/common/Examples
% ls -R
bin/ src/

./bin:
x_solarSystem*

./src:
solarSystemSrc/

./src/solarSystemSrc:
base_class_includes/    main/                   star.cxx
derived_class_includes/ orbit.cxx
heavenlyBody.cxx        planet.cxx

./src/solarSystemSrc/base_class_includes:
heavenlyBody.h  orbit.h

./src/solarSystemSrc/derived_class_includes:
planet.h  star.h

./src/solarSystemSrc/main:
solarSystem.cxx
% cd src
% cc -g -o ../bin/x_solarSystem \
  -IsolarSystemSrc/base_class_includes \
  -IsolarSystemSrc/derived_class_includes \
  main/solarSystem.cxx heavenlyBody.cxx orbit.cxx planet.cxx star.cxx

Then you move the directory solarSystemSrc elsewhere:

% mv solarSystemSrc movedSolarSystemSrc

Now debug x_solarSystem in /usr/users/debugger/sandbox/test/src/common/Examples/bin:

DBX Mode
(idb) list $curline - 10: 20
    ./solarSystemSrc/main/solarSystem.cxx
    ../src/solarSystemSrc/main/solarSystem.cxx
    /usr/examples/solarSystemSrc/main/solarSystem.cxx
    ./solarSystem.cxx
    ../src/solarSystem.cxx
    /usr/examples/solarSystem.cxx

The debugger cannot find the file because it has been moved to another directory.

The following command displays a summary of the source directories in a.out. The ellipsis (...) here means that solarSystemSrc contains one or more source directories.

DBX Mode
(idb) show source directory
.
solarSystemSrc
 ...

Information: You can further expand a '...' using the command

    show source directory <directory>
or
    show all source directory <directory>

where <directory> is the directory on the line above the '...'.
The first command displays only the children of <directory>, whereas
the second command displays all the descendants of <directory>.

The following command directs the debugger to look for source files originally in solarSystemSrc in movedSolarSystemSrc instead. This time, the debugger finds the source file.

DBX Mode
(idb) map source directory solarSystemSrc ../src/movedSolarSystemSrc
(idb) list $curline - 10: 20
    104
    105     // Insert the new entry appropriately
    106     //
    107     if (iAmBiggerThan < biggestCount) {
    108         biggestMoons[iAmBiggerThan] = moon;
    109     }
    110 }
    111
    112 int main()
    113 {
>   114     unsigned int j = 1;    // for scoping examples
    115     for (unsigned int i = 0; i < biggestCount; i++)
    116         biggestMoons[i] = NULL;
    117
    118     Star *sun = new Star("Sol", G, 2);
    119     buildOurSolarSystem(sun);
    120     sun->printBodyAndItsSatellites(j);
    121     printBiggestMoons();
    122
    123     return 0;

The following command gives a complete list of source directories. As you can see, solarSystemSrc is mapped to movedSolarSystemSrc. As a side effect of mapping solarSystemSrc to movedSolarSystemSrc, the subdirectories in solarSystemSrc are mapped to their counterparts under movedSolarSystemSrc.

DBX Mode
(idb) show all source directory
.
solarSystemSrc  *=>  ../src/movedSolarSystemSrc
 solarSystemSrc/base_class_includes  =>  ../src/movedSolarSystemSrc/base_class_includes
 solarSystemSrc/derived_class_includes  =>  ../src/movedSolarSystemSrc/derived_class_includes
 solarSystemSrc/main  =>  ../src/movedSolarSystemSrc/main

To summarize, the debugger provides the following four commands for checking and setting source directory mappings:

Use the show source directory(dbx) command to display the directory mapping information of directory_name and its child directories (or immediate subdirectory). If directory_name is not specified, the mapping information of all the source directories whose parent is not a source directory is displayed.

The show all source directory(dbx) command is identical to the show source directory(dbx) command except that the mapping information of all the descendants of directory_name is displayed:

DBX Mode
(idb) show source directory
.
solarSystemSrc  *=>  ../src/movedSolarSystemSrc
 ...
(idb) show all source directory
.
solarSystemSrc  *=>  ../src/movedSolarSystemSrc
 solarSystemSrc/base_class_includes  =>  ../src/movedSolarSystemSrc/base_class_includes
 solarSystemSrc/derived_class_includes  =>  ../src/movedSolarSystemSrc/derived_class_includes
 solarSystemSrc/main  =>  ../src/movedSolarSystemSrc/main

When you further expand ellipsis points (...) where directory is the directory on the line above the ellipsis points:

  • The show source directory(dbx) command displays only the children of directory_name.
  • The show all source directory(dbx) command displays all the descendants of directory_name.

Use the map source directory(dbx) command to tell the debugger that the source files in the directory from_directory_name can now be found in to_directory_name.

The unmap source directory(dbx) command maps from_directory_name back to itself; in other words, if from_directory_name has been mapped to some other directory, this command will restore its default mapping. For example:

DBX Mode
(idb) show source directory
.
solarSystemSrc  *=>  ../src/movedSolarSystemSrc
 ...
(idb) show source directory solarSystemSrc
solarSystemSrc  *=>  ../src/movedSolarSystemSrc
 solarSystemSrc/base_class_includes  =>  ../src/movedSolarSystemSrc/base_class_includes
 solarSystemSrc/derived_class_includes  =>  ../src/movedSolarSystemSrc/derived_class_includes
 solarSystemSrc/main  =>  ../src/movedSolarSystemSrc/main
(idb) unmap source directory solarSystemSrc
(idb) show source directory solarSystemSrc
solarSystemSrc
 solarSystemSrc/base_class_includes
 solarSystemSrc/derived_class_includes
 solarSystemSrc/main

Note: The symbol *=> means that you are setting the mapping explicitly using the map source directory(dbx) command, whereas => means that the mapping is derived from an existing explicit mapping.

By default, the use_list is: (1) the current directory and (2) the directory containing the executable file. (dbx) Each process has its own use_list. You can also use the idb command -I option to specify search directories.

The following commands let you view and modify the use_list.

GDB Mode
source_search_list_command
        : directory_command
        | show directories

Enter the use(dbx) without an argument or show directories(gdb) command to list the directories in which the debugger searches for source code files. Specify a directory argument to make source code files in that directory available to the debugger. You can also use the idb command -I option to specify search directories, which puts those directories in the use_list.

You can customize your debugger environment source code search paths by adding commands to your .dbxinit file that use the use command:

DBX Mode
use_command
        : use [directory_name ...]

If the directory_name is specified, it is either appended to or replaces the use_list, depending on whether the value of the $dbxuse debugger variable is zero (append) or non-zero (replace).

GDB Mode
directory_command
        : directory [directory_name ...]

Use the directory command with no argument to reset the use_list to empty value.
If the directory_name is specified, it is prepended to use_list. Several directory names may be given to the directory command, separated by ':' or whitespace. If you specify a directory, which is already in the list, it is moved forward.
To get the full list of directories in the search list, use the show directories command.

DBX Mode

The unuse command removes entries from the use_list:

unuse_command
        : unuse [directory_name ...]
        | unuse *

Enter the unuse command without the directory_name to set the search list to the default (the home directory, the current directory, and the directory containing the executable file). Include the directory names to remove them from the search list. The asterisk (*) argument removes all directories from the search list.

How the Debugger Chooses Which Source File to List

The debugger has a concept of current source file, so you do not have to explicitly specify a source file in many commands. Whenever the process stops, the current source file is set to the source file for the code currently executing. The commands up, down, class(dbx), and file(dbx) also set the current source file.

DBX Mode

You can see and modify the current source file selection:

select_source_file_command
        : file [ filename ]
        : fileexpr [ expression ]

Use the file command without a file name to display the name of the current file scope. Include the file name to change the file scope. Change the file scope to set a breakpoint in a function not in the file currently being executed.

To see source code for or set a breakpoint in a function not in the file currently being executed, use the file command to set the file scope.

If the file name is not a literal, use the fileexpr command. For example, if you have a script that calculates a file name in a debugger variable or in a routine that returns a file name as a string, you can use fileexpr to set the file.

The following example uses the file command to set the debugger file scope to a file different from the main program, and then stops at line number 26 in that file. This example also shows the fileexpr command setting the current scope back to the original file, which is solarSystem.cxx.

(idb) run
[1] stopped at [int main(void):114 0x804c440]
    114     unsigned int j = 1;    // for scoping examples
(idb) file
solarSystemSrc/main/solarSystem.cxx
(idb) set $originalFile = "solarSystem.cxx"
(idb) list 24: 10
     24     Moon   *phobos    = new Moon("Phobos",       9,   11, mars);
     25     Moon   *deimos    = new Moon("Deimos",      23,    6, mars);
     26
     27     Planet *jupiter   = new Planet("Jupiter",     778330, sun);
     28     Moon   *io        = new Moon("Io",         422, 1815, jupiter);
     29     Moon   *europa    = new Moon("Europa",     671, 1569, jupiter);
     30     Moon   *ganymede  = new Moon("Ganymede",  1070, 2631, jupiter);
     31     Moon   *callisto  = new Moon("Callisto",  1883, 2400, jupiter);
     32     Moon   *amalthea  = new Moon("Amalthea",   181,   98, jupiter);
     33
(idb) file star.cxx
(idb) list 24: 10
     24 // Stars are simple objects
     25 //
     26 Star::Star(
     27     char*           name,
     28     StellarClass    classification,
     29     StellarSubclass subclassification)
     30         : HeavenlyBody(name),
     31           _classification(classification),
     32           _subclassification(subclassification)
     33 {
(idb) stop at 26
[#2: stop at "solarSystemSrc/star.cxx":26 ]
(idb) cont
[2] stopped at [Star::Star(char*, enum StellarClass, StellarSubclass):26 0x804d1f4]
     26 Star::Star(
(idb) file
solarSystemSrc/star.cxx
(idb) fileexpr $originalFile
(idb) file
solarSystemSrc/main/solarSystem.cxx
(idb) list 24: 10
     24     Moon   *phobos    = new Moon("Phobos",       9,   11, mars);
     25     Moon   *deimos    = new Moon("Deimos",      23,    6, mars);
     26
     27     Planet *jupiter   = new Planet("Jupiter",     778330, sun);
     28     Moon   *io        = new Moon("Io",         422, 1815, jupiter);
     29     Moon   *europa    = new Moon("Europa",     671, 1569, jupiter);
     30     Moon   *ganymede  = new Moon("Ganymede",  1070, 2631, jupiter);
     31     Moon   *callisto  = new Moon("Callisto",  1883, 2400, jupiter);
     32     Moon   *amalthea  = new Moon("Amalthea",   181,   98, jupiter);
     33
GDB Mode

In the GDB mode, the current file can be changed, for example, using the list command.
See the example:

(idb) run
Starting program: /usr/examples/x_solarSystem

Breakpoint 1, main () at solarSystemSrc/main/solarSystem.cxx:114
114	    unsigned int j = 1;    // for scoping examples
(idb) info source
Current source file is solarSystemSrc/main/solarSystem.cxx
(idb)
(idb) list 24,+10
24	    Moon   *phobos    = new Moon("Phobos",       9,   11, mars);
25	    Moon   *deimos    = new Moon("Deimos",      23,    6, mars);
26
27	    Planet *jupiter   = new Planet("Jupiter",     778330, sun);
28	    Moon   *io        = new Moon("Io",         422, 1815, jupiter);
29	    Moon   *europa    = new Moon("Europa",     671, 1569, jupiter);
30	    Moon   *ganymede  = new Moon("Ganymede",  1070, 2631, jupiter);
31	    Moon   *callisto  = new Moon("Callisto",  1883, 2400, jupiter);
32	    Moon   *amalthea  = new Moon("Amalthea",   181,   98, jupiter);
33
(idb) list star.cxx:24,+10
24	// Stars are simple objects
25	//
26	Star::Star(
27	    char*           name,
28	    StellarClass    classification,
29	    StellarSubclass subclassification)
30	        : HeavenlyBody(name),
31	          _classification(classification),
32	          _subclassification(subclassification)
33	{
(idb) info source
Current source file is star.cxx
(idb) break 26
Breakpoint 2 at 0x804d1a0: file solarSystemSrc/star.cxx, line 26.
(idb) continue
Continuing.

Breakpoint 2, Star::Star (this=0x80d7f70, name=0x808ecfc "Sol", classification=G, subclassification=2 '\002') at solarSystemSrc/star.cxx:26
26	Star::Star(
(idb) info source
Current source file is star.cxx
(idb) list solarSystem.cxx:24,+10
24	    Moon   *phobos    = new Moon("Phobos",       9,   11, mars);
25	    Moon   *deimos    = new Moon("Deimos",      23,    6, mars);
26
27	    Planet *jupiter   = new Planet("Jupiter",     778330, sun);
28	    Moon   *io        = new Moon("Io",         422, 1815, jupiter);
29	    Moon   *europa    = new Moon("Europa",     671, 1569, jupiter);
30	    Moon   *ganymede  = new Moon("Ganymede",  1070, 2631, jupiter);
31	    Moon   *callisto  = new Moon("Callisto",  1883, 2400, jupiter);
32	    Moon   *amalthea  = new Moon("Amalthea",   181,   98, jupiter);
33
(idb) info source
Current source file is solarSystem.cxx
Listing Source Files
DBX Mode

The simplest way to see a source file is to use a text editor. The edit command will display an editor on the current file, using the current definition of the EDITOR environment variable, if there is one.

However, some primitive inspection capabilities are built into the debugger. The list command displays source lines, which can be defined by following way:

  • The position of the program counter
  • The last line listed, if multiple list commands are entered
  • The line number specified as the arguments to the list command
DBX Mode
list_source_file_command
        : list [ line_expression ]
        | list line_expression , line_expression
        | list line_expression : line_expression

line_expression
        : expression

If specified, the first expression must evaluate to either an integer (the line number of the first line to display within the current source file) or a function (the first line of the function).

Specify the exact range of source lines as either a comma followed by the expression for the last line, or a colon followed by the expression for the the number of lines. This second expression must evaluate to an integer value.

If a second expression is not given, the debugger shows 20 lines, fewer if the end of source file is reached.

GDB Mode
list_source_file_command
        : list [ [+ | -] line_expression ] [ , [ [+ | -] line_expression ] ]
        | list function

If a function name or the only one line_expression is specified as the single argument then lines centred around the function beginning or specified line are printed.
Commands list + and list - print some lines after and before the last printed correspondently.

The line in arguments can be specified by following way:

  • integer number - the line in the current source file with specified number
  • +offset and -offset - the line moved on 'offset' lines upper or downer from the last printed
  • *address - the line which contains code for specified program address

User can specify a source file name for arguments given in form integer number and function by following way:
filename:integer_number and filename:function.

For example, to list lines 16 through 20:

DBX Mode
(idb) list 16, 20
     16
     17 class Node {
     18 public:
     19     Node ();
     20
GDB Mode
(idb) list 16,20
16
17	class Node {
18	public:
19	    Node ();
20

For example, to list 6 lines, beginning with line 16:

DBX Mode
(idb) list 16: 6
     16
     17 class Node {
     18 public:
     19     Node ();
     20
     21     virtual void printNodeData() const = 0;
GDB Mode
(idb) list 16,+6
16
17	class Node {
18	public:
19	    Node ();
20
21	    virtual void printNodeData() const = 0;
Searching the Content of Source Files

The following search commands search through the current source file to help you find the lines to list:

DBX Mode
search_source_file_command
        : / [ string ]
        | ? [ string ]
DBX Mode

Note: The string is actually just the rest of the line, not a string literal. The rest of the line is still having alias expansion done on it.

Use a slash (/) to search forward from the most recently listed line; use a question mark (?) to search backward. Like most searches, it will stop at the end (or beginning) of the file being searched, and will wrap if the command is repeated at that point.

When the string is omitted, the previous search continues from where it found the string. When the string is present, the search starts from either the start (/) or the end (?) of the current line.

When a match is found, the debugger lists the line number and the line. That line becomes the starting point for any further searches, or for a list command. For example:

  1. To locate _firstNode:

    DBX Mode
    (idb) /_firstNode
         69     NODETYPE* _firstNode;
    
    GDB Mode
    (idb) forward-search _firstNode
    69	    NODETYPE* _firstNode;
    
  2. Then to locate append before line 69:

    DBX Mode
    (idb) ?append
         65     void      append  (NODETYPE* const node);
    
    GDB Mode
    (idb) reverse-search append
    65	    void      append  (NODETYPE* const node);
    
  3. Then to locate append after line 65:

    DBX Mode
    (idb) /append
        145 void List<NODETYPE>::append(NODETYPE* const node)
    
    GDB Mode
    (idb) forward-search append
    145	void List<NODETYPE>::append(NODETYPE* const node)
    

The debugger provides parameterized aliases and debugger variables of arbitrary types. You can use these to do list traversal (see the array navigation example).

Looking at the Threads

A thread is a single, sequential flow of control within a process. Each thread contains a single point of execution. Threads execute within (and share) a single address space; therefore, a process's threads can read and write the same memory locations.

Thread Levels

The debugger supports three levels of threads:

  • decthreads (user spece threads), also known as POSIX threads on Tru64
  • pthreads (kernel threads), also known as POSIX threads in Linux
  • Kernel threads (operating system level threads), also known as native threads

To specify the thread level, set the $threadlevel debugger variable to one of the following strings:

  • decthreads — for POSIX thread library debugging on Tru64
  • pthreads — for POSIX thread library debugging on Intel® platforms in Linux
  • native — for kernel thread debugging.

For example:

(idb) set $threadlevel = "decthreads"

For core file debugging, the $threadlevel is always set to "native".

Thread Manipulation Commands

You can use a variety of commands to manipulate the threads:

Thread Display Commands

You can use the following commands to display threads:

DBX Mode
show_thread_command
        : show thread [ thread_id_list ] [ thread-state-filter ]

thread_id_list
        : thread_id ,...
        | *

thread_id
        : expression

thread_state_filter
        : with state eq thread_state

eq
        : ==               (for C, and C++)
        | .eq.             (for Fortran)

thread_state
        : ready
        | running
        | terminated
        | blocked

Use the show thread command without parameters to list all the threads known to the debugger.

If you specify one or more thread identifiers, the debugger displays information about the threads you specify, if the thread matches what you specified in the list. If you omit a thread specification, the debugger displays information for all threads.

Use the show thread commands to list threads that have specific characteristics, such as threads that are currently blocked. For example:

(idb) print $threadlevel
"decthreads"
(idb) show thread
  Thread Name                      State           Substate    Policy       Pri
  ------ ------------------------- --------------- ----------- ------------ ---
*      1 default thread            running VP 3                SCHED_OTHER  19
      -1 manager thread            blk SCS                     SCHED_RR     19
      -2 null thread for slot 0    running VP 1                null thread  -1
      -3 null thread for slot 1    ready VP 3                  null thread  -1
      -4 null thread for slot 2    new             new         null thread  -1
      -5 null thread for slot 3    new             new         null thread  -1
>      2 threads(0x140000798)      blocked         cond 3      SCHED_OTHER  19
       3 threads+8(0x1400007a0)    blocked         cond 3      SCHED_OTHER  19
       4 threads+16(0x1400007a8)   blocked         cond 3      SCHED_OTHER  19
       5 threads+24(0x1400007b0)   blocked         cond 3      SCHED_OTHER  19
       6 threads+32(0x1400007b8)   blocked         cond 3      SCHED_OTHER  19
(idb) set $threadlevel = "native"
(idb) print $threadlevel
"native"
(idb) show thread
   Id                  State
*  0x9                 stopped
*  0x9                 unstarted
   0x3                 unstarted
   0x7                 unstarted
GDB Mode
info_threads_command
        : info threads

thread
        : expression

Use the info threads command to list all the threads known to the debugger.

(idb) info threads
  0 Thread 1024 (LWP 19513)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  1 Thread 2049 (LWP 19514)  0x805a42a in __clone from /tmp/pthread_manythreads
* 2 Thread 1026 (LWP 19515)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  3 Thread 2051 (LWP 19516)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  4 Thread 3076 (LWP 19517)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
  5 Thread 4101 (LWP 19518)  0x8048288 in prime_search at pthread_manythreads.c:79

Note: In the output, the right bracket indicator (>) marks the current thread, whereas the asterisk (*) indicator marks the thread with the event that stopped the application.

You can switch to a different thread as the current thread. The debugger variable $curthread contains the thread identifier of the current thread.

switch_thread_command
          : thread [ thread_id ]

The $curthread value is updated when program execution stops or completes. You can modify the current thread by assigning $curthread a valid thread identifier. This is equivalent to issuing the thread thread_id command. When there is no process or program, $curthread is set to 0.

Use the thread command without a thread identifier to identify the current thread. Supply a thread identifier to make another thread the current thread.

GDB Mode
(idb) thread 2
* 2 Thread 1026 (LWP 19515)  0x804f8f6 in __sigsuspend from /tmp/pthread_manythreads
Mutex Queries

A mutex (mutual exclusion) semaphore is a programming flag that allows multiple pthreads to synchronize access to shared resources, to ensure the following:

  • All threads see a clean and consistent view of the data, without allowing one thread to change something while another thread is reading it.
  • Two threads do not change different parts of the data at the same time, possibly in inconsistent ways.

Use the show mutex command to list information about currently available pthread mutexes:

DBX Mode
show_mutex_variable_command
        : show mutex [ mutex_id_list ] [ mutex_state_filter ]

mutex_id_list
        : mutex_id  ,...
        | (mutex_id ,...)

mutex_state_filter
        : with state eq mutex_state

eq
        : ==               (for C, and C++)
        | .eq.             (for Fortran)

mutex_state
        : locked

If you specify one or more mutex identifiers, the debugger displays information about only those mutexes specified, provided that the list matches the identifiers of currently available mutexes. If you omit the mutex identifier specification, the debugger displays information about all mutexes currently available.

Use the show mutex with state == locked command to display information exclusively for locked mutexes.

If $verbose is set to 1, the sequence numbers of the threads locking the mutexes are displayed.

The following example shows the output from a simple show mutex command:

DBX Mode
(idb) show mutex
Mutex  Name                      State Owner  Pri Type     Waiters (+Count)
------ ------------------------- ----- ------ --- -------- --------------------
     1 malloc heap                                Normal
     2 malloc hash                                Normal
     3 malloc cache[0]                            Normal
     4 malloc cache[1]                            Normal
     5 malloc cache[2]                            Normal
     6 malloc cache[3]                            Normal
     7 malloc cache[4]                            Normal
     8 malloc cache[5]                            Normal
     9 malloc cache[6]                            Normal
    10 malloc cache[7]                            Normal
    11 malloc cache[8]                            Normal
    12 malloc cache[9]                            Normal
    13 malloc cache[10]                           Normal
    14 malloc cache[11]                           Normal
    15 malloc cache[12]                           Normal
    16 malloc cache[13]                           Normal
    17 malloc cache[14]                           Normal
    18 malloc cache[15]                           Normal
    19 malloc cache[16]                           Normal
    20 malloc cache[17]                           Normal
    21 malloc cache[18]                           Normal
    22 malloc cache[19]                           Normal
    23 malloc cache[20]                           Normal
    24 malloc cache[21]                           Normal
    25 malloc cache[22]                           Normal
    26 malloc cache[23]                           Normal
    27 malloc cache[24]                           Normal
    28 malloc cache[25]                           Normal
    29 malloc cache[26]                           Normal
    30 malloc cache[27]                           Normal
    31 malloc cache[28]                           Normal
    32 brk                                        Normal
    33 exc cr                                     Recurs
    34 exc read rwl                               Normal
    35 known mutex queue                          Normal
    36 known cond queue                           Normal
    37 known VP queue                             Normal
    38 known rwl queue                            Normal
    39 VM 0 lookaside                             Normal
    40 VM 1 lookaside                             Normal
    41 VM 2 lookaside                             Normal
    42 VM 0 cache                                 Normal
    43 VM 1 cache                                 Normal
    44 VM 2 cache                                 Normal
    45 debugger client registry                   Normal
    46 Global lock                                Recurs
    47 ldr                                        Recurs
    48 prime_list(0x140000660)                    Normal
    49 cond_mutex(0x1400006c0)                    Normal
    50 current_mutex(0x140000690                  Normal
    51 curr_worker_mutex(0x14000 Lock             Normal

If the application being debugged has no pthreads, or if the $threadlevel is set to native, an appropriate message is issued.

Condition Variable Queries

A condition variable is a pthread synchronization object used in conjunction with a mutex. A condition variable is used when a thread has locked a mutex to gain access to data and then finds it must wait for some other thread to change some aspect of the data before it can continue.

Use the show condition command to list information about currently available condition variables. If you supply one or more condition identifiers, the debugger displays information about the condition variables you specify, provided that the list matches the identities of currently available condition variables. If you omit the condition variable specification, the debugger displays information about all the condition variables currently available.

Use the show condition with state == wait command to display information only for condition variables that have one or more threads waiting. If $verbose is set to 1, the sequence numbers of the threads waiting on the condition are displayed.

The following example shows output from a simple show condition command:

DBX Mode
(idb) show condition
Cond   Name                      Mutex  Type  Waiters (+Count)
------ ------------------------- ------ ----- ---------------------------------
     1 _exc_read_mutex+72(0x3ffc
     2 _exc_read_mutex+112(0x3ff
     3 cond_var(0x140000720)         49       2, 3, 4, 5, 6
     4 curr_worker(0x140000748)

If the application being debugged has no pthreads, or if the $threadlevel is set to native, an appropriate message is issued.

Other Thread Commmands

You can use the where command to display the stack trace of current threads. You can specify one or more threads or all threads.

The print command evaluates an optional expression in the context of the current thread and displays the result.

The call command evalutes an expression in the context of the current thread and makes the call in the context of the current thread.

The printregs command prints the registers for the current thread.

Undocumented pthread Support

You can pass an undocumented string directly into the undocumented pthread debugging support. This is an internal debugging aid, not intended for general use.

Looking at the Call Stack

Most programming languages have some concept of functions, routines, or subroutines, capturing the notion of code that is invoked from many places. A running program needs a call stack of call frames for the called functions. Each call frame contains both the information needed to return to its caller and the information needed to contain the local variables of the function.

The machine code generated for these functions maintains this call stack. Some of this maintenance is done before the call, some at the start of the called function, some at the end of the called function, and some after the call.

Non-optimized machine code is usually very easy to correlate with the source code, but optimized machine code can be tricky. See Call Frames and Optimized Code and Call Frames and Machine Code Correlation for more information.

The debugger controls the call stacks of all the threads; you can use it to examine and manipulate call stacks, and use them as a basis for further queries:

call_stack_command
        : show_stack_command
        | change_stack_frame_command
        | pop_stack_frame_command

When your process is stopped by the debugger, you can show the call stack of the thread that caused the stoppage, or the call stack of any other thread.

The following commands show the most recent call frames on the call stack of the current or specified threads:

If specified, the expression must evaluate to a nonnegative integer. You can specify the number of call frames to show. If not specified, all the call frames for the thread are shown.

DBX Mode

If specified, the thread_specifier specifies the threads whose call stacks are to be shown. If not specified, just the current thread is used.

When large and complex values are passed by value to a routine on the stack, the output of the where command can be voluminous. You can set the control variable $stackargs to 0 to suppress the output of argument values in the where command.

The stack trace provides the following information for each call level:

Call level The number used to refer to a call level on the stack. The function entered most recently is at level 0. Its caller is at level 1.
Memory address The address of the next instruction to be executed at this level.
Function name The name of the function for the memory address.
File name The source file for the memory address.
Line number The number of the next source line of the memory address.

If your call stack seems to be missing routines, you may be seeing the result of a compiler optimization known as "tail calls".

If your call stack is corrupted, you may see random numbers without any routine names. In this case, it is likely that your application has gotten lost. Typically, this type of call stack display means that your application has lost track of the real stack and real code location, and is now executing random bits of memory, interpreting them as instructions.

If you are coding in C++, one of the most common ways to get a corrupt stack is for your code to try to execute a method on an invalid object. If the object has already been deleted, has not yet been initialized, is not there, or is of a completely different type, then the virtual function table will not be correct, and the application will be treating random memory as the virtual function table and calling a random place.

GDB Mode

Note: You should change the current thread to the thread whose call stack you want to examine.

Navigating the Call Stack

You can select one of the call frames as the starting point for examining variables. This call frame provides the current scope in the program for which variables exist, and tells the debugger which instance of those variables whose values you want to see.

Use the up command or the down command without the expression to change to the call frame located one level up or down the stack. Specify an expression that evaluates to an integer to change the call frame up or down the specified number of levels. If the number of levels exceeds the number of active calls on the stack in the specified direction, the debugger issues a warning message and the call frame does not change.

When the current call frame changes, the debugger displays the source line corresponding to the last instruction executed in the function executing the selected call frame.

DBX Mode

When large and complex values are passed by value to a routine on the stack, the output of the up and down commands can be voluminous. You can set the control variable $stackargs to 0 to suppress the output of argument values in the up and down commands.

Use the func command without the loc to display the current function. To change the function scope to a function that has a call frame in the call stack, specify the loc either as the name of the function or as an integer expression evaluating to the call level. If you specify the name, the most-recently entered call frame for that function becomes the current call frame.

If no frames are available to select from, the debugger context is set to the static context of the named function. The current scope and current language are set based on that function. Types and static variables local to that function are now visible and can be evaluated.

If you enter an integer expression, the debugger moves to the frame at level n, just as if you had entered up n at the level 0 function.

GDB Mode

up-silently and down-silently commands are similar to up and down respectively. But they work without invoking a new frame.

The frame command selects frame by given number or address. If there is no argument the command displays info about current stack frame.

In the following example, the current call frame is changed to one for method Planet::print so that a variable in that instance of print() can be displayed:

DBX Mode
(idb) where 4
#0  0x804cd5e in ((Planet*)0x80ec628)->Planet::print(i=2) "solarSystemSrc/planet.cxx":19
#1  0x804c75b in ((HeavenlyBody*)0x80ec628)->HeavenlyBody::printBodyAndItsSatellites(i=2) "solarSystemSrc/heavenlyBody.cxx":62
>2  0x804c790 in ((HeavenlyBody*)0x80ec5d0)->HeavenlyBody::printBodyAndItsSatellites(i=1) "solarSystemSrc/heavenlyBody.cxx":68
#3  0x804c4fc in main() "solarSystemSrc/main/solarSystem.cxx":120
(idb) list $curline - 5: 10
     63
     64     // Recursively deal with the satellites.  Redeclare i for scoping examples.
     65     //
     66     unsigned int j = 1;
     67     for (HeavenlyBody* i = _firstSatellite; i; i = i->_outerNeighbor) {
>    68     	i->printBodyAndItsSatellites(j++);
     69     }
     70 }
(idb) whatis i
class HeavenlyBody* i
(idb) print i
0x80ec628
(idb) func Planet::print
virtual void Planet::print(unsigned int) in solarSystemSrc/planet.cxx line No. 19:
     19     std::cout << "(" << i
(idb) where 4
>0  0x804cd5e in ((Planet*)0x80ec628)->Planet::print(i=2) "solarSystemSrc/planet.cxx":19
#1  0x804c75b in ((HeavenlyBody*)0x80ec628)->HeavenlyBody::printBodyAndItsSatellites(i=2) "solarSystemSrc/heavenlyBody.cxx":62
#2  0x804c790 in ((HeavenlyBody*)0x80ec5d0)->HeavenlyBody::printBodyAndItsSatellites(i=1) "solarSystemSrc/heavenlyBody.cxx":68
#3  0x804c4fc in main() "solarSystemSrc/main/solarSystem.cxx":120
(idb) list $curline - 5: 10
     14 {
     15 }
     16
     17 void Planet::print(unsigned int i) const
     18 {
>    19     std::cout << "(" << i
     20          << ") Planet [" << HeavenlyBody::name() << "]; ";
     21     printOrbitalParameters();
     22     std::cout << std::endl;
     23 }
(idb) whatis i
unsigned int i
(idb) print i
2

In the previous example, instead of entering func Planet::print, you can enter down 2. (You would use down in this case because the current call frame at the start of the example was not the bottommost frame.) Note that the final stack trace in this example lists a call frame for function Planet::print as the current call frame (denoted by the > character).

GDB Mode

There are some commands to print info about the selected stack frame:

  • info frame
    The command prints following info about the current frame:
    • the address of the frame
    • the address of the next frame down (called by this frame)
    • the address of the next frame up (caller of this frame)
    • the language in which the source code corresponding to this frame is written
    • the address of the frame's arguments
    • the address of the frame's local variables
    • the program counter saved in it (the address of execution in the caller frame)
    • which registers were saved in the frame
  • info frame addr
    The command prints a verbose description of the frame at address addr, without selecting that frame.
  • info args
    The command prints the arguments of the selected frame, each on a separate line.
  • info locals
    The command prints the local variables of the selected frame, each on a separate line. These are all variables (declared either static or automatic) accessible at the point of execution of the selected frame.
  • info catch
    The command prints a list of all the exception handlers that are active in the current stack frame at the current point of execution.
The pop Command

The pop command removes one or more call frames from the call stack:

DBX Mode
pop_stack_frame_command
        : pop [ expression ]

The default is one call frame. The pop command undoes the work already done by the removed execution frames. It does not, however, reverse side effects, such as changes to global variables.

Note: Because it is extremely unlikely this will fix all the effects of a half-executed call, this command is not recommended for general use. Furthermore, the pop command does not provide a way to specify a return value when the frame being discarded corresponds to a function that should return a value. You may need to use the assign command to restore the values of global variables.

Instead of the pop command, you may want to use the return command, which finishes the call corresponding to the selected frame.

GDB Mode
pop_stack_frame_command
        : return [ expression ]

When you use return, the selected stack frame is discarded (and all frames within it). If you wish to specify a value to be returned, give that value as the argument to return.

The return command does not resume execution. It leaves the program stopped in the state that would exist if the function had just returned. In contrast, the finish command resumes execution until the selected stack frame returns naturally.

Call Frames and Optimized Code

When optimized machine code is generated by the compilers, the compiler generates code that maintains the call stack, but sometimes the function boundaries are changed in one of two ways:

  • Inlining is when the compiler completely eliminates the call by instead generating the instructions for the called function at the call site, usually followed by merging those instructions with the other instructions surrounding the call site.
  • Outlining is when the compiler creates a function where one did not exist explicitly in the source. For example, the compiler turns a loop body into a function, so that it can generate code that uses threads to execute the different iterations in parallel; or the compiler creates a single shared function to replace several sections of the source that are similar.

Depending on the information the compiler makes available to the debugger, inlined calls may or may not show up in the call stack display. Outlined calls will show up, and will be correlated to the code they came from. The compiler will probably have supplied the debugger with some invented name for the function.

Call Frames and Machine Code Correlation

On a RISC processor the following is the machine code typically generated for a call to a function:

  • The machine code before the call performs the following operations:
    • Sets some context registers
    • Puts the parameters either in registers or memory
    • Loads the address of the function into a register
    • Loads the address to return to into a register
    • Branches to the function
  • The machine code at the start of the called function performs the following operations:
    • Sets some context registers
    • Allocates stack space
    • Saves some registers in the stack space
    • Performs some setup of the local variables
  • The machine code at the end of the called function performs the following operations:
    • Restores the saved registers from the stack space
    • Deallocates the stack space
    • Branches to the address to return to
  • The machine code at the return address of the call frame sets some context registers.

When the thread is partway through the call frame creation or tear-down, the debugger will still show the call frame, but will not be able to show correct values for the variables or parameters.

Special C++ Issues

For nonstatic member functions, the implicit this pointer is displayed as the address on the stack trace along with the class type of the object, as shown in the following example:

DBX Mode
(idb) stop in List<Node>::print
[#3: stop in void List<Node>::print(void) ]
(idb) cont
[3] stopped at [void List<Node>::print(void):162 0x804aad2]
    162     Node* currentNode = _firstNode;
(idb) where 2
>0  0x804aad2 in ((List<Node>*)0xbfffecfc)->List<Node>::print() "x_list.cxx":162
#1  0x804a53a in main() "x_list.cxx":203
GDB Mode
(idb) break List<Node>::print
Breakpoint 3 at 0x804aad2: file x_list.cxx, line 162.
(idb) continue
Continuing.

Breakpoint 3, List<Node>::print (this=0xbfffecec) at x_list.cxx:162
162	    Node* currentNode = _firstNode;
(idb) backtrace 2
>0  0x804aad2   in ((List<Node>*)(const class List<Node> *) 0xbfffecec)->List<Node>::print(this=(const class List<Node> *) 0xbfffecec) "x_list.cxx":162
#1  0x804a53a   in main() "x_list.cxx":203

Looking at the Data

After you have seen the call stack (show_stack_command), selected the call frame containing the variables you wish to examine (change_stack_frame_command), and looked at the source this function is executing (looking at the source), you usually want to examine some of the variables or even evaluate some expressions. You can use the print command and the call command to do this. You can also use the following commands to help you determine what to look at and what you are seeing:

The print Command

You can print the values of one or more expressions or all local variables. You can also use the print command to evaluate complex expressions involving typecasts, pointer dereferences, multiple variables, constants, and any legal operators allowed by the language of the program you are debugging:

DBX Mode
print_command
        : print [ expression ,... ]
        | print rescoped_expression
        | print printable-type
        | printb [ expression ,... ]
        | printd [ expression ,... ]
        | printo [ expression ,... ]
        | printx [ expression ,... ]

rescoped_expression
        : filename ` qual_symbol
        | ` qual_symbol

qual_symbol
        : expression
        | qual_symbol ` expression

For an array, the debugger prints every cell in the array if you do not specify a specific cell.

Use the $hexints, $decints, or $octints variables to select a radix for the output of the print command. If you do not want to change the radix permanently, use the printx, printd, printo , and printb commands to print expressions in hexadecimal, decimal, octal, or binary base format, respectively.

GDB Mode
print_command
        : print [/format specifier][expression]

format specifier
        : x | d | u | o | t | a | c | f

By default, GDB prints a value according to its data type. But user might want to print a number in hex, or a pointer in decimal. Or user might want to view data in memory at a certain address as a character string or as an instruction. In this case user should specify an output format.
To specify how to print a value already computed a user should print after the print command a slash and a format letter. The format letters supported are:

  • x
    Regard the bits of the value as an integer, and print the integer in hexadecimal
  • d
    Print as integer in signed decimal.
  • u
    Print as integer in unsigned decimal.
  • o
    Print as integer in octal.
  • t
    Print as integer in binary. The letter `t' stands for "two".
  • a
    Print as an address, both absolute in hexadecimal and as an offset from the nearest preceding symbol.
  • c
    Regard as an integer and print it as a character constant.
  • f
    Regard the bits of the value as a floating point number and print using typical floating point syntax.

For an array, the debugger prints every cell in the array if you do not specify a specific cell.

Consider the following declarations in a C++ program:

DBX Mode
(idb) list 59: 2
     59 const unsigned int biggestCount = 10;
     60 static Moon *biggestMoons[biggestCount];
GDB Mode
(idb) list 59,+2
59	const unsigned int biggestCount = 10;
60	static Moon *biggestMoons[biggestCount];

The following example uses the print command to display a nonstring array:

DBX Mode
(idb) print biggestMoons
[0] = 0x80eca28,[1] = 0x80ecd88,[2] = 0x80eca88,[3] = 0x80ec968,[4] = 0x80ec788,[5] = 0x80ec9c8,[6] = 0x80ed148,[7] = 0x80ed028,[8] = 0x80ecd28,[9] = 0x80ed088
GDB Mode
(idb) print biggestMoons
$4 = {(class Moon *) 0x80eca28, (class Moon *) 0x80ecd88, (class Moon *) 0x80eca88, (class Moon *) 0x80ec968, (class Moon *) 0x80ec788, (class Moon *) 0x80ec9c8, (class Moon *) 0x80ed148, (class Moon *) 0x80ed028, (class Moon *) 0x80ecd28, (class Moon *) 0x80ed088}

The following example shows how to print individual values of an array:

DBX Mode
(idb) print biggestMoons[3]
0x80ec968
(idb) print *biggestMoons[3]
class Moon {
  _radius = 1815;
  _name = 0x808d71c="Io";         // class Planet::HeavenlyBody
  _innerNeighbor = 0x0;           // class Planet::HeavenlyBody
  _outerNeighbor = 0x80ec9c8;     // class Planet::HeavenlyBody
  _firstSatellite = 0x0;          // class Planet::HeavenlyBody
  _lastSatellite = 0x0;           // class Planet::HeavenlyBody
  _primary = 0x80ec908;           // class Planet::Orbit
  _distance = 422;                // class Planet::Orbit
  _name = 0x80eca00="Jupiter 1";  // class Planet::Orbit
}
GDB Mode
(idb) print biggestMoons[3]
$7 = (class Moon *) 0x80ec968
(idb) print *biggestMoons[3]
$8 = {<Planet> = {<HeavenlyBody> = {_name = 0x808d71c "Io", _innerNeighbor = 0x0, _outerNeighbor = 0x80ec9c8, _firstSatellite = 0x0, _lastSatellite = 0x0}, <Orbit> = {_primary = 0x80ec908, _distance = 422, _name = 0x80eca00 "Jupiter 1"}}, _radius = 1815}
Dereferencing Pointers

Pointers are variables that contain addresses. By dereferencing a pointer in the command interface, you can print the value at the address pointed to by the pointer. In C and C++ programs, variables containing a pointer are dereferenced using the * operator. The following example shows how to dereference a pointer in C++ programs:

(idb) whatis newNode
class IntNode* newNode
(idb) print newNode
0x80b9740
(idb) print *newNode
class IntNode {
  _data = 1;
  _nextNode = 0x0;                // class Node
}
Printing C Strings

The debugger does not print more than the first $maxstrlen characters of a null-terminated string. Change this debugger variable if it is showing either more or less than you wish to see.

Printing Floating Point Numbers

Floating point numbers are represented inside the computer in binary floating point. They are converted to decimal floating point when printed. The two formats are not the same, and some numbers are easily represented in decimal but not in binary (for example the number 1.1). The internal binary form for such numbers is an approximation, the closest that can be made given the number of bits available.

Normally, when a binary floating point number is printed, the shortest decimal number which would be represented by that binary number is used as the number to print, as it is a legitimate representation of the internal binary number. However, to see a more exact (extended form) representation of a binary floating point number, you can set the $floatshrinking debugger variable to 0 (zero).

The following example shows the result of converting 1.1 (shortened form) to the closest long double binary floating point number (extended form).

(idb) p $floatshrinking
1
(idb) p 1.1
1.1
(idb) set $floatshrinking = 0
(idb) p 1.1
1.10000000000000000000000000000000008

Currently, the extended forms are only available for long double variables and expressions.

For more detail on floating point representation, see ANSI IEEE standard 754-1985.

Restrictions on the print Command

Expressions containing labels are not supported. Variables involving static anonymous unions and enumerated types may not be able to be printed. Printing a structure that is declared but not defined in a compilation unit may generate an error message indicating that the structure is opaque.

The printf Command

Use the printf command to format and display a complex structure. The first argument is a string expression of characters and conversion specifications using the same format specifiers as the printf C function. The printf command requires a running target program because it uses libc.

printf_command
        : printf [ format_string [ , expression  ,... ] ]

For example:

(idb) printf "The PC is 0x%x", $pc
The PC is 0x804a53a
The printi Command

The printi command takes one or more numerical expressions and interprets each one as an assembly instruction, printing out the instruction, and its arguments when applicable. This command is typically used by engineers performing machine-level debugging.

printi_command
	: printi [ expression  ,... ]

For example:

(idb) $curpc/1i
int main(void): x_list.cxx
*[line 182, 0x8049973]	main+0x1b:            pushl    %edi
(idb) $curpc/1dd
0x8049973:    746949975
(idb) printi $pc
main+0x1b:            pushl    %edi
The printregs Command

Use the printregs command to display the values of all the hardware registers. The list of registers displayed by the debugger is machine-dependent. By default, most values are displayed in decimal radix. To display the register values in hexadecimal radix, set the $hexints variable to 1.

print_registers_command
        : printregs

For example:

(idb) printregs
$eax           0x80b0c14	134941716
$ecx           0x40018000	1073840128
$edx           0x0	0
$ebx           0x401ae9e4	1075505636
$esp [$sp]     0xbffff58c	-1073744500
$ebp           0xbffff668	-1073744280
$esi           0x40016b64	1073834852
$edi           0xbffff6dc	-1073744164
$eip [$pc]     0x804a3e2	134521826
$eflags        0x286	646
$cs            0x23	35
$ss            0x2b	43
$ds            0x2b	43
$es            0x2b	43
$fs            0x0	0
$gs            0x0	0
$orig_eax      0xffffffff	-1
$fctrl         0x37f	895
$fstat         0x0	0
$ftag          0x0	0
$fiseg         0x23	35
$fioff         0x805badf	134593247
$foseg         0x2b	43
$fooff         0xbffff368	-1073745048
$fop           0x1bd	445
$f0            0x000000000000ffff0000000000000000 -inf
$f1            0x000000000000ffff0000000000000000 -inf
$f2            0x000000000000ffff0000000000000000 -inf
$f3            0x000000000000ffff0000000000000000 -inf
$f4            0x000000000000ffff0000000000000000 -inf
$f5            0x000000000000ffff0000000000000000 -inf
$f6            0x000000000000ffff0000000000000000 -inf
$f7            0x0000000000004002a1f7cf0000000000 10.123
$xmm0          0x00000000000000000000000000000000
$xmm1          0x00000000000000000000000000000000
$xmm2          0x00000000000000000000000000000000
$xmm3          0x00000000000000000000000000000000
$xmm4          0x00000000000000000000000000000000
$xmm5          0x00000000000000000000000000000000
$xmm6          0x00000000000000000000000000000000
$xmm7          0x00000000000000000000000000000000
$mxcsr         0x1f80	8064
$vfp           0xbffff668	0xbffff668
The printt Command

The printt command takes one or more numerical expressions and interprets each one as the number of seconds since the Epoch (00:00:00 UTC 1 Jan 1970; see ctime(3) for more information).

printt_command
        : printt [ expression  ,... ]

For example:

(idb) printt 0
(UTC) Thu Jan  1 00:00:00 1970
(idb) printt 978325200
(UTC) Mon Jan  1 05:00:00 2001
The dump Command

Use the dump command without an argument to list the parameters and local variables in the current function. To list the parameters and local variables in an active function, specify it as an argument.

Use the dump . command (include the dot) to list the parameters and local variables for all functions active on the stack:

dump_command
        : dump qual_symbol
        | dump .

For example:

(idb) dump
>0  0x804a53a in main() "x_list.cxx":203
cNode=0x80b9750
cNode1=0x80b9768
cNode2=0x80b97a0
newNode=0x80b9740
newNode2=0x80b9790
nodeList=class List<Node> { ... }

When large and complex values are passed by value to a routine on the stack, the output of the dump command can be voluminous. You can set the control variable $stackargs to 0 to suppress the output of argument values in the dump command.

The call Command

After a breakpoint or a signal suspends program execution, you can execute a single function in your program by using the call command, or by including a function call in the expression argument of a debugger command. Calling a function lets you test the function's operation with a specific set of parameters.

call_command
        : call call-expression

Specify the function as if you were calling it from within the program. If the function has no parameters, specify empty parentheses (()). For multithreaded applications, the call is made in the context of the current thread. For C++ applications, when you set the $overloadmenu debugger variable to 1 and call an overloaded function, the debugger lists the overloaded functions and calls the function you specify. When the function you call completes normally, the debugger restores the stack and the current context that existed before the function was called.

While the program counter is saved and restored, calling a function does not shield the program state from alteration if the function you call allocates memory or alters global variables. If the function affects global program variables, for instance, those variables will be changed permanently.

Functions compiled without the debugger option to include debugging information may lack important parameter information and are less likely to yield consistent results when called.

The call command executes the specified function with the parameters you supply and then returns control to you (at the debugger prompt) when the function returns. The call command discards the return value of the function. If you embed the function call in the expression argument of a print command, the debugger prints the return value after the function returns. The following example shows both methods of calling a function:

(idb) call earth->distance()
(idb) print earth->distance()
149600

In the previous example, the call command results in the return value being discarded while the embedded call passes the return value of the function to the print command, which in turn prints the value. You can also embed the call within a more involved expression, as shown in the following example:

(idb) print earth->distance() - 100000
49600
(idb) print mars->distance() - earth->distance()
78340
(idb) call io->print(3)
(3) Moon [Io], radius [1815] km; <Jupiter 1> orbits at 422 Megameters

All breakpoints or tracepoints defined and enabled during the session are active when a called function is executing. When program execution halts during function execution, you can examine program information, execute one line or instruction, continue execution of the function, or call another function.

When you call a function when execution is suspended in a called function, you are nesting function calls, as shown in the following example:

(idb) where 2
>0  0x804bb6b in buildOurSolarSystem(sun=0x80cf4d0) "solarSystemSrc/main/solarSystem.cxx":55
#1  0x804c4f2 in main() "solarSystemSrc/main/solarSystem.cxx":119
(idb) stop in Planet::print
[#2: stop in virtual void Planet::print(unsigned int) ]
(idb) call mars->print(1)
[2] stopped at [virtual void Planet::print(unsigned int):19 0x804cd72]
     19     std::cout << "(" << i
(idb) where
>0  0x804cd72 in ((Planet*)0x80cf6e8)->Planet::print(i=1) "solarSystemSrc/planet.cxx":19
#1  0x4004b1a4 in __do_global_ctors_aux(...) in /lib/i686/libm.so.6
#2  0x80cf6e8
(idb) next
stopped at [virtual void Planet::print(unsigned int):20 0x804cdc6]
     20          << ") Planet [" << HeavenlyBody::name() << "]; ";
(idb) stop in Orbit::distance
[#3: stop in Megameters Orbit::distance(void) ]
(idb) print distance()
[3] stopped at [Megameters Orbit::distance(void):41 0x804cac7]
     41     return _distance;
(idb) where
>0  0x804cac7 in ((Orbit*)0x80cf700)->Orbit::distance() "solarSystemSrc/orbit.cxx":41
#1  0x4004b1a4 in __do_global_ctors_aux(...) in /lib/i686/libm.so.6
#2  0x80cf700
(idb) disable 3
(idb) cont
Called Procedure Returned
stopped at [virtual void Planet::print(unsigned int):20 0x804cdc6]
     20          << ") Planet [" << HeavenlyBody::name() << "]; ";
(idb) where
>0  0x804cdc6 in ((Planet*)0x80cf6e8)->Planet::print(i=1) "solarSystemSrc/planet.cxx":20
#1  0x4004b1a4 in __do_global_ctors_aux(...) in /lib/i686/libm.so.6
#2  0x80cf6e8
(idb) cont
(1) Planet [Mars]; <Sol 4> orbits at 227940 Megameters
Called Procedure Returned
stopped at [void buildOurSolarSystem(struct Star*):55 0x804bb6b]
     55     Planet *pluto     = new Planet("Pluto",      5913520, sun);
Restrictions on the call Command

The debugger supports function calls and expression evaluations that call functions, with the following limitations:

  • The debugger does not support passing and returning structures by value.
  • The debugger does not implicitly construct temporary objects for call parameters.
  • Optimization can prevent the debugger from knowing the type of a function return. Therefore, the debugger assumes returns are of the type int if the functions are optimized. If the returns are a different type, it may be necessary to cast the result when calling the optimized functions.
The whatis Command

You can print information about the basic nature of a whatis_expression. The expression can be a normal language expression or the name of a type, function, or other language entity. The debugger shows you information about the entity rather than evaluating it. However, it will evaluate any contained expressions, such as pointers, needed to determine the entity to which you are referring.

whatis_command
        : whatis whatis_expression

The following example uses the whatis command to determine the storage representation for the data member _classification:

(idb) whatis sun->_classification
const enum StellarClass Star::_classification
(idb) whatis StellarClass
enum StellarClass {O, B, A, F, G, K, M, R, N, S}
(idb) print sun->_classification
G
The whereis Command

The whereis command lists all declarations of a variable and each declaration's fully qualified scope information.

The scope information of a variable usually consists of the name of the source file that contains the function in which the variable is declared, the name of that function, and the name of the variable. The components of the scope information are separated by back-quotes (`).

whereis_command
        : whereis whereis_name
        | whereis whereis_string

whereis_name
        : identifier_or_typedef_name
        | ( identifier_or_typedef_name )

whereis_string
	: string

You can use the whereis command with the whereis_name to obtain information needed to differentiate overloaded identifiers that are in different units, or within different routines in the same unit. The following example shows how to set breakpoints in two C++ methods, both named print:

(idb) whereis print
"solarSystemSrc/heavenlyBody.cxx"`HeavenlyBody::print(unsigned int)
"solarSystemSrc/main/solarSystem.cxx"`HeavenlyBody::print(unsigned int)
"solarSystemSrc/main/solarSystem.cxx"`Moon::print(unsigned int)
"solarSystemSrc/main/solarSystem.cxx"`Planet::print(unsigned int)
"solarSystemSrc/main/solarSystem.cxx"`Star::print(unsigned int)
"solarSystemSrc/orbit.cxx"`HeavenlyBody::print(unsigned int)
"solarSystemSrc/planet.cxx"`HeavenlyBody::print(unsigned int)
"solarSystemSrc/planet.cxx"`Moon::print(unsigned int)
"solarSystemSrc/planet.cxx"`Planet::print(unsigned int)
"solarSystemSrc/star.cxx"`HeavenlyBody::print(unsigned int)
"solarSystemSrc/star.cxx"`Star::print(unsigned int)
print
(idb) stop in "solarSystemSrc/derived_class_includes/planet.h"`Planet::print
Select from
----------------------------------------------------
     1 planet.h containing Kilometers
     2 planet.h containing Kilometers
     3 None of the above
----------------------------------------------------
1
[#2: stop in virtual void Planet::print(unsigned int) ]
(idb) stop in "solarSystemSrc/derived_class_includes/star.h"`Star::print
Select from
----------------------------------------------------
     1 star.h containing StellarClass
     2 star.h containing StellarClass
     3 None of the above
----------------------------------------------------
1
[#3: stop in virtual void Star::print(unsigned int) ]

See also the which command for another example of the whereis command.

If you are not sure how to spell a symbol, you can use the whereis command with the whereis_string to search the symbol table for the regular expression represented by the quoted string. All symbols that match the rules of the regular expression are displayed in ascending order. For example:

(idb) whereis planet
Symbol not found
(idb) whereis "[Pp]lanet"
"solarSystemSrc/derived_class_includes/planet.h"`Moon::Moon(char*, Megameters, Kilometers, class Planet*)
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet::Planet(char*, Megameters, class HeavenlyBody*)
"solarSystemSrc/derived_class_includes/planet.h"`Planet::Planet(char*, Megameters, class HeavenlyBody*)
"solarSystemSrc/derived_class_includes/planet.h"`Planet::print(unsigned int)
"solarSystemSrc/derived_class_includes/planet.h"`__INTER__Moon_Moon_Orbit_Planet_Xv
"solarSystemSrc/derived_class_includes/planet.h"`__INTER__Planet_Planet_Orbit_Xv
"solarSystemSrc/derived_class_includes/planet.h"`__dt__6PlanetXv
__T_6Planet
__cxxexsig6Planet
__vtbl_5Orbit6Planet
__vtbl_5Orbit6Planet4Moon
__vtbl_6Planet
solarSystemSrc/derived_class_includes/planet.h
solarSystemSrc/derived_class_includes/planet.h
solarSystemSrc/derived_class_includes/planet.h
solarSystemSrc/planet.cxx
(idb) whereis "^Planet$"
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet::Planet(char*, Megameters, class HeavenlyBody*)
(idb) whereis Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet::Planet(char*, Megameters, class HeavenlyBody*)
(idb) which Planet
"solarSystemSrc/derived_class_includes/planet.h"`Planet
(idb) whatis Planet
class Planet : HeavenlyBody, Orbit {
  Planet(char*, Megameters, class HeavenlyBody*);
  virtual void print(unsigned int);
}

You can use the $symbolsearchlimit debugger variable to specify the maximum number of symbols that will be returned by the whereis command for a regular expression search. The default value for the $symbolsearchlimit variable is 100; a value of 0 indicates no limit.

The which Command

Use the which command to determine which declaration an identifier resolves to. The which command shows the fully qualified scope information for the instance of the specified expression visible from the current scope.

The scope information of a variable usually consists of the name of the source file that contains the function in which the variable is declared, the name of that function, and the name of the variable. The components of the scope information are separated by back-quotes (`).

which_command
        : which which_name

which_name
        : identifier_or_typedef_name
        | ( identifier_or_typedef_name )

The following example shows how to use the whereis and which commands to determine a variable's scope:

(idb) where 4
>0  0x804cd5e in ((Planet*)0x80ec628)->Planet::print(i=2) "solarSystemSrc/planet.cxx":19
#1  0x804c75b in ((HeavenlyBody*)0x80ec628)->HeavenlyBody::printBodyAndItsSatellites(i=2) "solarSystemSrc/heavenlyBody.cxx":62
#2  0x804c790 in ((HeavenlyBody*)0x80ec5d0)->HeavenlyBody::printBodyAndItsSatellites(i=1) "solarSystemSrc/heavenlyBody.cxx":68
#3  0x804c4fc in main() "solarSystemSrc/main/solarSystem.cxx":120
(idb) which i
"solarSystemSrc/planet.cxx"`Planet::print(unsigned int)`i
(idb) assign i = 10
(idb) print i
10
(idb) whereis i
"solarSystemSrc/heavenlyBody.cxx"`HeavenlyBody::printBodyAndItsSatellites(unsigned int)`i
"solarSystemSrc/heavenlyBody.cxx"`HeavenlyBody::printBodyAndItsSatellites(unsigned int)`i
"solarSystemSrc/heavenlyBody.cxx"`HeavenlyBody::satelliteNumber(class HeavenlyBody*)`i
"solarSystemSrc/main/solarSystem.cxx"`main`i
"solarSystemSrc/main/solarSystem.cxx"`printBiggestMoons`i
"solarSystemSrc/main/solarSystem.cxx"`trackBiggestMoons(class Moon*)`i
"solarSystemSrc/planet.cxx"`Moon::print(unsigned int)`i
"solarSystemSrc/planet.cxx"`Planet::print(unsigned int)`i
"solarSystemSrc/star.cxx"`Star::print(unsigned int)`i
(idb) func HeavenlyBody::printBodyAndItsSatellites
void HeavenlyBody::printBodyAndItsSatellites(unsigned int) in solarSystemSrc/heavenlyBody.cxx line No. 62:
     62     print(i);
(idb) which i
"solarSystemSrc/heavenlyBody.cxx"`HeavenlyBody::printBodyAndItsSatellites(unsigned int)`i
(idb) print i
2
Notes on C++ Debugging

The following sections describe the debugger commands specific to debugging C++ programs.

Setting the Class Scope Using the class Command

The debugger maintains the concept of a current context in which to perform lookup of program variable names. The current context includes a file scope and either a function scope or a class scope. The debugger automatically updates the current context when program execution suspends.

The class command lets you set the scope to a class in the program you are debugging:

c++_look_around_command
        : class [ class_name ]

If class_name is not specified, the class command displays the current class context.

Setting the class scope nullifies the function scope and vice versa. To return to the default (current function) scope, use the command func 0.

Explicitly setting the debugger's current context to a class enables you to view a class to:

  • Set a breakpoint in a member function
  • Print static data members
  • Examine any data member's type

After the class scope is set, you can set breakpoints in the class's member functions and examine data without explicitly mentioning the class name. If you do not want to affect the current context, you can use the scope resolution operator (::) to access a class whose members are not currently visible. Use the class command without an argument to display the current class scope. Specify an argument to change the class scope. After the class scope is set, refer to members of the class by omitting the classname:: prefix.

The following example shows the use of the class command to set the class scope to List<Node> in order to make member function append visible so a breakpoint can be set in append:

(idb) stop in append
Symbol "append" is not defined.
append has no valid breakpoint address
Warning: Breakpoint not set
(idb) class List<Node>
struct List<Node> {
  struct Node* _firstNode;
  List(void);
  void append(struct Node* const);
  void print(void);
  ~List(void);
}
(idb) stop in append
[#1: stop in void List<Node>::append(struct Node* const) ]
Displaying Class Information

The whatis and print commands display information on a class. Use the whatis command to display static information about the classes. Use the print command to view dynamic information about class objects.

The whatis command displays the class type declaration, including the following:

  • Data members
  • Member functions
  • Constructors
  • Destructors
  • Static data members
  • Static member functions

For classes that are derived from other classes, the data members and member functions inherited from the base class are not displayed. Any member functions that are redefined from the base class are displayed.

The print command lets you display the value of data members and static members. Information regarding the public, private, or protected status of class members is not provided, because the debugger relaxes the related access rules to be more helpful to users.

The type signatures of member functions, constructors, and destructors are displayed in a form that is appropriate for later use in resolving references to overloaded functions.

The following example shows the whatis and print commands in conjunction with a class:

(idb) list 43: 12
     43 // Compound Node - contains integer and float data items
     44 //
     45 class CompoundNode : public IntNode {
     46 public:
     47     CompoundNode (float fdata, int idata);
     48
     49     void printNodeData() const;
     50
     51 private:
     52     float _fdata;
     53 };
     54
(idb) whatis CompoundNode
struct CompoundNode : IntNode {
  float _fdata;
  CompoundNode(float, int);
  virtual void printNodeData(void);
}
(idb) whatis CompoundNode::CompoundNode
CompoundNode::CompoundNode(float, int)
(idb) stop in CompoundNode::printNodeData
[#1: stop in virtual void CompoundNode::printNodeData(void) ]
(idb) run
The list is:
Node 1 type is integer, value is 1
[1] stopped at [virtual void CompoundNode::printNodeData(void):109 0x804a654]
    109     cout << " type is compound, value is ";
(idb) print _fdata
12.3450003
Displaying Object Information

The whatis and print commands display information on instances of classes (objects). Use the whatis command to display the class type of an object. Use the print command to display the current value of an object.

You can also display individual object members using the member access operators, period (.) and right arrow (->), in a print command.

You can use the scope resolution operator (::) to refer to global variables, to refer to hidden members in base classes, to explicitly refer to a member that is inherited, or to name a member hidden by the current context.

When you are in the context of a nested class, you must use the scope resolution operator to access members of the enclosing class.

The following example shows how to use the whatis and print commands to display object information:

(idb) whatis this
const struct CompoundNode* const this
(idb) whatis *this
struct CompoundNode : IntNode {
  float _fdata;
  CompoundNode(float, int);
  virtual void printNodeData(void);
}
(idb) print *this
struct CompoundNode {
  _fdata = 12.3450003;
  _data = 2;                      // class IntNode
  _nextNode = 0x80b0f68;          // class IntNode::Node
}
(idb) print _fdata, _data
12.3450003 2
(idb) print this->_fdata, this->_data
12.3450003 2
Displaying Static and Dynamic Type Information

When displaying object information for C++ class pointers or references, you have the option of viewing either static type information or dynamic type information.

The static type of a class pointer or reference is its type as defined in the source code, and thus cannot change. The dynamic type is the type of the object being referenced, before any casts were made to that object, and thus may change during program execution.

The debugger provides a debugger variable, $usedynamictypes, which allows you to control which form of the type information is displayed. The default value for this variable is true (1), which indicates that the dynamic type information is displayed. Setting this variable to false (0) instructs the debugger to display static type information. The output of the print, trace, tracei, and whatis commands are affected.

The display of dynamic type information is supported for C++ class pointers and references. All other types display static type information. In addition, if the dynamic type of an object cannot be determined, the debugger defaults to the use of static type information.

This debugger functionality does not relax the C++ visibility rules regarding object member access through a pointer/reference (only members of the static type are accessible). For more information about the C++ visibility rules, see The Annotated C++ Reference Manual (by Margaret E. Ellis and Bjarne Stroustrup, 1990, Addison-Wesley Publishing Company).

In order for dynamic type information to be displayed, the object's static type must have at least one virtual function defined as part of its interface (either one it introduced or one it inherited from a base class). If no virtual functions are present for an object, only the static type information for that object is available for display.

The following example shows debugger output with $usedynamictypes set to 0 (false):

(idb) print $usedynamictypes
0
(idb) whatis *this
class HeavenlyBody {
  const char* const _name;
  class HeavenlyBody* _innerNeighbor;
  class HeavenlyBody* _outerNeighbor;
  class HeavenlyBody* _firstSatellite;
  class HeavenlyBody* _lastSatellite;
  HeavenlyBody(char*);
  void addSatellite(class HeavenlyBody*);
  const char* name(void);
  virtual void print(unsigned int);
  void printBodyAndItsSatellites(unsigned int);
  unsigned int satelliteNumber(class HeavenlyBody*);
}
(idb) print *this
class HeavenlyBody {
  _name = 0x8090968="Moon";
  _innerNeighbor = 0x0;
  _outerNeighbor = 0x0;
  _firstSatellite = 0x0;
  _lastSatellite = 0x0;
}

The following example displays debugger output with $usedynamictypes set to 1 (true). The output is for the same object as the previous example, at the same point in program execution:

(idb) print $usedynamictypes
1
(idb) whatis *this
class Moon : Planet {
  const Kilometers _radius;
  Moon(char*, Megameters, Kilometers, class Planet*);
  virtual void print(unsigned int);
  Kilometers radius(void);
}
(idb) print *this
class Moon {
  _radius = 1738;
  _name = 0x8090968="Moon";       // class Planet::HeavenlyBody
  _innerNeighbor = 0x0;           // class Planet::HeavenlyBody
  _outerNeighbor = 0x0;           // class Planet::HeavenlyBody
  _firstSatellite = 0x0;          // class Planet::HeavenlyBody
  _lastSatellite = 0x0;           // class Planet::HeavenlyBody
  _primary = 0x80e2b48;           // class Planet::Orbit
  _distance = 384;                // class Planet::Orbit
  _name = 0x80e2c40="Earth 1";    // class Planet::Orbit
}
Displaying Virtual and Inherited Class Information

When you use the print command to display information on an instance of a derived class, the debugger displays both the new class members as well as the members inherited from a base class. Pointers to members of a class are not supported.

When you use the print command to display the format of C++ classes, the class name (or structure/union name) is displayed at the top of the output. Data members of a class that are inherited from another class are commented using a double slash (//). Only those data members that are inherited within the current class being printed are commented.

The following example shows how the debugger uses C++ style comments to identify inherited class members. In the example, class CompoundNode inherits from class IntNode, which inherits from class Node. When printing a class CompoundNode object, the data member _data is commented with "// class IntNode", signifying that it is inherited from class IntNode. The member _nextNode is commented with "// class IntNode::Node", showing that it is inherited from class IntNode, which inherits it from class Node. This commenting is also provided for C++ structs.

(idb) where 3
>0  0x804a519 in ((Node*)0x80b7c70)->Node::Node() "x_list.cxx":79
#1  0x804a568 in ((IntNode*)0x80b7c70)->IntNode::IntNode(data=2) "x_list.cxx":88
#2  0x804a61f in ((CompoundNode*)0x80b7c70)->CompoundNode::CompoundNode(fdata=12.3450003, idata=2) "x_list.cxx":103
(idb) whatis *this
class Node {
  class Node* _nextNode;
  Node(void);
  class Node* getNextNode(void);
  virtual void printNodeData(void);
  void setNextNode(class Node*);
}
(idb) print *this
class Node {
  _nextNode = 0x0;
}
(idb) up 1
>1  0x804a568 in ((IntNode*)0x80b7c70)->IntNode::IntNode(data=2) "x_list.cxx":88
     88 IntNode::IntNode(int data) : _data(data)
(idb) whatis *this
class IntNode : Node {
  int _data;
  IntNode(int);
  virtual void printNodeData(void);
}
(idb) print *this
class IntNode {
  _data = 0;
  _nextNode = 0x0;                // class Node
}
(idb) up 1
>2  0x804a61f in ((CompoundNode*)0x80b7c70)->CompoundNode::CompoundNode(fdata=12.3450003, idata=2) "x_list.cxx":103
    103     IntNode(idata),
(idb) whatis *this
class CompoundNode : IntNode {
  float _fdata;
  CompoundNode(float, int);
  virtual void printNodeData(void);
}
(idb) print *this
class CompoundNode {
  _fdata = 0;
  _data = 0;                      // class IntNode
  _nextNode = 0x0;                // class IntNode::Node
}

If two members in an object have the same name but different base class types (multiple inheritance), you can refer to the members using the following syntax:

object.class::member

or

object->class::member

This syntax is more effective than using the object.member and object->member syntaxes, which can be ambiguous. In all cases, the debugger uses the C++ language rules as defined in The Annotated C++ Reference Manual to determine which member you are specifying.

The following example shows a case in which the expanded syntax can be used:

(idb) print *jupiter
class Planet {
  _name = 0x808eca0="Jupiter";    // class HeavenlyBody
  _innerNeighbor = 0x80d8188;     // class HeavenlyBody
  _outerNeighbor = 0x80d84e8;     // class HeavenlyBody
  _firstSatellite = 0x80d8308;    // class HeavenlyBody
  _lastSatellite = 0x80d8488;     // class HeavenlyBody
  _primary = 0x80d7f70;           // class Orbit
  _distance = 778330;             // class Orbit
  _name = 0x80d8340="Sol 5";      // class Orbit
}
(idb) print jupiter->_name
Overloaded Values
0x80d8340="Sol 5"
0x808eca0="Jupiter"
(idb) print jupiter->HeavenlyBody::_name
0x808eca0="Jupiter"
(idb) print jupiter->Orbit::_name
0x80d8340="Sol 5"
Member Functions on the Stack Trace

The implicit this pointer, which is a part of all nonstatic member functions, is displayed as the address on the stack trace. The class type of the object is also given.

Sometimes the debugger does not see class type names with internal linkage. When this happens, the debugger issues the following error message:

Name is overloaded.

Trying to examine an inlined member function that is not called results in the following error:

Member function has been inlined.

The debugger will report this error regardless of the setting of the -noinline_auto compilation flag. As a workaround, include a call to the given member function somewhere in your program. (The call does not need to be executed.)

If a program is not compiled with the -g flag, a breakpoint set on an inlined member function may confuse the debugger.

Resolving Ambiguous References to Overloaded Functions

In most cases, the debugger works with one specific function at a time. In the case of overloaded function names, you must specify the desired overloaded function. Following are two ways to resolve references to overloaded function names, both under the control of the $overloadmenu debugger variable (the default setting of this debugger variable is 1):

  • Choose the correct reference from a selection menu.

    If the $overloadmenu variable is set to 1 (the default), whenever you specify a function name that is overloaded, a menu appears with all the possible functions; you must select from this menu. In this example, a breakpoint is set in foo, which is overloaded:

    (idb) set $overloadmenu = 1
    (idb) stop in C::foo
    Select from
    ----------------------------------------------------
         1 int C::foo(double*)
         2 void C::foo(float)
         3 void C::foo(int)
         4 void C::foo(void)
         5 None of the above
    ----------------------------------------------------
    1
    [#10: stop in int C::foo(double*) ]
    
  • Enter the function name with its full type signature.

    If you prefer this method, set the $overloadmenu variable to 0. To see the possible type signatures for the overloaded function, first display all the declarations of an overloaded function by using the whatis command.

    You cannot select a version of an overloaded function that has a type signature containing ellipsis points (...). Pointers to functions with type signatures that contain the list parameter or ellipsis points are not supported.

    Use the specific function type signature to refer to the desired version of the overloaded function. If a function has no parameter, include the void parameter as the function's type signature. In the following example, the function context is set to foo(double *), as foo is overloaded:

    (idb) func foo
    Error: foo is overloaded
    (idb) func foo(double *)
    int C::foo(double*) in x_overload.cxx line No. 25:
         25 int  C::foo(double *) { return state;}
    
Advanced Program Information — Verbose Mode

By default, the debugger gives no information on virtual base class pointers for the following:

  • Derived classes
  • Virtual pointer tables for virtual functions
  • Compiler-generated function members
  • Compiler-generated temporary variables
  • Implicit parameters in member functions

By setting the $verbose debugger variable to 1, you can request that this information be printed in subsequent debugger responses. When the $verbose debugger variable is set to 1 and you display the contents of a class using the whatis command, several of the class members listed are not in the source code of the original class definition. The following line shows specific output from the whatis command for one of the additional members:

(idb) whatis CompoundNode::__vptr
(array [subrange 0 ... 0 of int] of union {
  void <member function>(void)* fptr;
  int ioffset;
})* Node::__vptr

The __vptr variable contains the addresses of all virtual functions associated with the class. The compiler generates several other class members for internal use.

The compiler generates additional parameters for nonstatic member functions. When the $verbose debugger variable is set to 1, these extra parameters are displayed as part of each member function's type signature. If you specify a version of an overloaded function by entering its type signature and the variable is set to 1, you must include these parameters. Do not include these parameters if the variable is set to 0.

When the $verbose variable is set to 1, the output of the dump command includes not only standard program variables but also compiler-generated temporary variables.

The following example prints class information using the whatis command under different settings of the $verbose variable:

(idb) set $verbose = 0
(idb) whatis CompoundNode
class CompoundNode : IntNode {
  float _fdata;
  CompoundNode(float, int);
  virtual void printNodeData(void);
}
(idb) set $verbose = 1
(idb) whatis CompoundNode
class CompoundNode : IntNode {
  float _fdata;
  CompoundNode(class CompoundNode* const, float, int);
  virtual void printNodeData(const class CompoundNode* const);
}

Looking at the Generated Code

This section discusses the following topics:

Memory Display and Search Commands

You can use the following commands to read arbitrary memory locations in your program:

machinecode_level_command
     : examine_command
     : search_command

examine_command
     : address_expression / [ count ] [ mode ]
     | address_expression ? [ count ] [ mode ]
     | address_expression , address_expression / [ mode ]

search_command
     : address_expression / [ count ] search_mode   value  mask
     | address_expression ? [ count ]  search_mode  value  mask
     | address_expression , address_expression /  search_mode  value  mask

count
     : integer_constant

mode
     : d 	Print a short word in decimal
     | dd      	Print a 32-bit (4-byte) decimal display
     | D      	Print a long word in decimal
     | u	Print a short word in unsigned decimal
     | uu	Print a 32-bit (4-byte) unsigned decimal display
     | U	Print a long word in unsigned decimal
     | o	Print a short word in octal
     | oo	Print a 32-bit (4-byte) octal display
     | O	Print a long word in octal
     | x	Print a short word in hexadecimal
     | xx	Print a 32-bit (4-byte) hexadecimal display
     | X       	Print a long word in hexadecimal
     | b       	Print a byte in hex
     | c       	Print a byte as a character
     | s       	Print a string of characters (a C-style string ending in null)
     | C       	Print a wide character as a character
     | S       	Print a null terminated string of wide characters
     | f       	Print a single precision real number
     | g       	Print a double precision real number
     | L       	Print a long double precision real number
     | i       	Disassemble machine instructions

search_mode
     : m   	32-bit search mode
     | M       	64-bit search mode

value
     : integer_constant

mask
     : integer_constant

The first examine_command displays the count number of memory values in the requested format, starting at address_expression. If count is not specified, 1 is assumed. The count value must be a positive value.

If you wish to see memory values leading up to the address_expression, use the second examine_command. The second examine_command displays count number of memory values in the requested format ending at the address_expression. If count is not specified, 1 is assumed. The count value must be a positive value.

The third examine_command displays memory values in the requested format starting at the smaller of the two address_expressions and ending at the larger address_expression.

You can display stored values in the following formats by specifying mode. If mode is not specified, the mode used in the previous / command is assumed. If no previous / command exists, X is assumed.

When disassembling machine instructions, use the $regstyle variable to customize how the registers are displayed.

The search_commands allow you to search memory. Use the address_expression and count to determine the range of memory to search. Use the search_mode to specify whether you want to search 32 or 64-bit chunks. The debugger will start at the specified starting address and read a chunk of memory (either 32 or 64 bits in size) and apply the mask and comparison on that chunk of memory. For example, if you want to search memory for a particular instruction or search an array of either integer or floating-point values, the 32-bit search would be efficient because machine instructions and integer and floating-point data types are 32 bits in length. Use the value to specify the memory value to seek. Use the mask to specify those bits that must match the same bits in the specified value. To ensure that a possible match will be found, the debugger applies the mask to the input value prior to starting the search, to remove any bits that could prevent a match from occurring. Then, for each memory location searched, the debugger applies the mask to the memory value and then compares it with this new input value. If a match is found, then the address and memory value are displayed.

For example, suppose the user wishes to check an array of 100 integers in memory to see if any values are NULL (0):

(idb) array,&(array[99])/m 0x0 0xfffffff
0x1400005d0:  0x00000000

Suppose the user wishes to search for the trapb instruction:

(idb) printi 0x63ff0000
trapb
(idb) main/1000m 0x63ff0000 0xfffffff
0x12561314: 0x63ff0000

Use the debugger variable $memorymatchall to cause the debugger to output all matches in the specified range. Suppose you want to search a long integer array of 100 values for the first value over 80, and then want to find all values in the array over 80:

(idb) printx 80
0x50
(idb) longarray/100M 0x50 0x50
0x140002680: 0x0000000000000050
(idb) set $memorymatchall
(idb) longarray/100M 0x50 0x50
0x140002680: 0x0000000000000050
0x140002688: 0x0000000000000051
0x140002690: 0x0000000000000052
0x140002698: 0x0000000000000053
0x1400026a0: 0x0000000000000054
0x1400026a8: 0x0000000000000055
0x1400026b0: 0x0000000000000056
0x1400026b8: 0x0000000000000057
0x1400026c0: 0x0000000000000058
0x1400026c8: 0x0000000000000059
0x1400026d0: 0x000000000000005a
0x1400026d8: 0x000000000000005b
0x1400026e0: 0x000000000000005c
0x1400026e8: 0x000000000000005d
0x1400026f0: 0x000000000000005e
0x1400026f8: 0x000000000000005f
(idb)
Machine-Level Debugging

The debugger lets you debug your programs at the machine-code level as well as at the source-code level. Using debugger commands, you can examine and edit values in memory, print the values of all machine registers, and step through program execution one machine instruction at a time.

Only those users familiar with machine-language programming and executable file code structure will find low-level debugging useful.

For more information on machine-level debugging, see Machine-Level Debugging in Part III.

Looking at Shared Libraries

shared_library_command
        : listobj
        | readsharedobj filename
        | delsharedobj  filename

Use the listobj command to list all loaded objects, including the main image and the shared libraries. For each object, the information listed consists of the full object name (with pathname) and the starting and ending addresses for the .text, .data, and .bss sections.

Use the readsharedobj command to read in the symbol table information for the specified shared object. This object must be a shared library. You can use the command only when you specify the debuggee; that is, either the debugger has been invoked with it, or the debuggee was loaded by the load command.

Conversely, use the delsharedobj command to remove the symbol table information for the shared object from the debugger.

Chapter 9 — Modifying the Process

In addition to the normal side effects of evaluating expressions, including calls, you can explicitly modify the memory of the current process and also modify the actual loadable file (either executable file or shared library) that has been mapped into memory.

The following sections discuss these commands.

The assign and the set variable Commands

Use the assign(dbx) and the set variable(gdb) commands to change the value associated with a variable, memory address, or expression that is accessible according to the scope and visibility rules of the language. The expression can be any expression that is valid in the current context.

DBX Mode
modifying_command
        : assign target = expression
        | patch target = expression

target
        : unary_expression
GDB Mode
modifying_command
        : set [ variable ] expression

The set variable command evaluates the specified expression. If expression includes assignment operator, it executes like all other operators. This is the way to change memory.

The only difference between the set variable and the print commands is printing the value — the set variable does not print anything.

Note: The variable can be omitted if the beginning of expression does not confuse the debugger, for example, does not look like a valid subcommand for the set command.

The following example shows how to deposit the value 5 into the data member _data of a C++ object:

DBX Mode
(idb) print node->_data
2
(idb) assign node->_data = 5
(idb) print node->_data
5
GDB Mode
(idb) print node->_data
$2 = 2
(idb) set variable node->_data = 5
(idb) print node->_data
$3 = 5

The following example shows how to change the value associated with a variable and the value associated with an expression:

DBX Mode
(idb) print *node
struct CompoundNode {
  _fdata = 12.3450003;
  _data = 5;                      // class IntNode
  _nextNode = 0x0;                // class IntNode::Node
}
(idb) assign node->_data = -32
(idb) assign node->_fdata = 3.14159 * 4.2 * 4.2
(idb) assign node->_nextNode = _firstNode
(idb) print *node
struct CompoundNode {
  _fdata = 55.4176483;
  _data = -32;                    // class IntNode
  _nextNode = 0x80c0fe0;          // class IntNode::Node
}
GDB Mode
(idb) print *node
$6 = {<IntNode> = {<Node> = {_nextNode = 0x0}, _data = 5}, _fdata = 12.345}
(idb) set variable node->_data = -32
(idb) set variable node->_fdata = 3.14159 * 4.2 * 4.2
(idb) set variable node->_nextNode = _firstNode
(idb) print *node
$7 = {<IntNode> = {<Node> = {_nextNode = 0x80b98e0}, _data = -32}, _fdata = 55.41765}

For C++, use the assign(dbx) and the set variable(gdb) commands to modify static and object data members in a class, and variables declared as reference types, type const, or type static. You cannot change the address referred to by a reference type, but you can change the value at that address.

assign [classname::]member = ["filename"] `expression
assign [object.]member = ["filename"] `expression

Note: Do not use the assign(dbx) and the set variable(gdb) commands to change the PC. When you change the PC, no adjustment to the contents of registers or memory is made. This means that if you adjust the PC forward, the skipped instructions are not executed and any changes they would have made will not have happened. It means that if you adjust the PC backward, the instructions you backed up over are not undone, and any changes they made will be in effect when execution continues again.

Because most instructions change registers or memory in ways that can impact the meaning of the application, changing the PC is very likely to cause your application to do incorrect calculations and arrive at the wrong answer. Access violations and other errors and signals may result from changing the value in the PC.

The assign and the set variable Commands in Machine-Level Debugging

You can use the assign(dbx) and the set variable(gdb) commands to alter the contents of memory specified by an address as shown in the following example:

DBX Mode
(idb) set $address = &(node->_data)
(idb) print $address
0x80c0ff8
(idb) print *(int *)($address)
-32
(idb) assign *(int *)($address) = 1024
(idb) print *(int *)($address)
1024
GDB Mode
(idb) set variable $address = &(node->_data)
(idb) print $address
$11 = (int *) 0x80b98f8
(idb) print *(int *)($address)
$12 = 32
(idb) set variable *(int *)($address) = 1024
(idb) print *(int *)($address)
$13 = 1024

See Machine-Level Debugging for more information.

The patch Command (DBX Mode only)

Use the patch command to correct bad data or instructions in executable disk files. You can patch the text, initialized data, or read-only data areas. You cannot patch the bss segment, or stack and register locations, because they do not exist on disk files.

Use this command exclusively when you need to change the on-disk binary. Use the assign command when you need only to modify debuggee memory. If the image is executing when you issue the patch command, the corresponding location in the debuggee address space is updated as well. (The debuggee is updated regardless of whether the patch to disk succeeded, as long as the source and destination expressions can be processed by the assign command.) If your program is loaded but not yet started, the patch to disk is performed without the corresponding assign to memory.

(idb) run
[1] stopped at [int main(void):24 0x120001324]
24     return 0;
(idb) patch i = 10
0x1400000d0 = 10
(idb) patch j = i + 12
0x1400000d8 = 22
(idb)

Note: When you use the patch command, the original binary is not overwritten, but is saved with the string ~backup appended to the file name. This allows you to revert to the original binary if necessary. A file with the string ~temp appended to the file name may also be created. It may be deleted after the debugging session is over.

Chapter 10 — Continuing Execution of the Process

Before continuing the process, you should decide whether or not to make a snapshot, in case you want to revert to that snapshot state and try a different set of steps. After creating the snapshot, use the following commands to continue executing the program:

continue_command
        : step_into_command
        | step_over_command
        | step_out_of_command
        | cont_command
        | cont_from_place_command
        | finish_command

The step and stepi Commands

Use the step command to execute a line of source code. When the line being stepped contains a function call, the step command steps into the function and stops at the first executable statement.

Use the stepi command to step into the next machine instruction. When the instruction contains a function call, the stepi command steps into the function being called.

For multithreaded applications, use these commands to continue execution of the current thread while putting all other threads on hold.

If you supply the optional expression argument, the debugger evaluates the expression as a positive integer that specifies the number of times to execute the command. The expression can be any expression that is valid in the current context.

step_into_command
        : step  [ step_number ]
        | stepi [ step_number ]
step_number
        :  expression

In the following example, two step commands continue executing a C++ program:

DBX Mode
(idb) step
stopped at [void List<Node>::append(class Node* const):151 0x80583f9]
    151         Node* currentNode = _firstNode;
(idb) step
stopped at [void List<Node>::append(class Node* const):152 0x8058401]
    152         while (currentNode->getNextNode())
GDB Mode
(idb) step
151	        Node* currentNode = _firstNode;
(idb) step
152	        while (currentNode->getNextNode())

The following example shows stepping by instruction (stepi). To see stepping into calls, see the next example.

DBX Mode
(idb) $curpc/4i
void List<Node>::append(class Node* const): x_list.cxx
*[line 156, 0x8058463]	append(class Node* const)+0x83:         leave
 [line 156, 0x8058464]	append(class Node* const)+0x84:         ret
 [line 156, 0x8058465]	append(class Node* const)+0x85:         nop
 [0x8058466]	append(class Node* const)+0x86:         nop
(idb) stepi
stopped at [void List<Node>::append(class Node* const):156 0x8058464]	append(class Node* const)+0x84:         ret
(idb) return
stopped at [int main(void):190 0x804a0d8]
    190     nodeList.append(cNode);
GDB Mode
(idb) x /4i $pc
0x804aacf <append+131>:         leave
0x804aad0 <append+132>:         ret
0x804aad1 <append+133>:         nop
0x804aad2 <append+134>:         nop
(idb) stepi
156	}
(idb) finish
main () at x_list.cxx:190
190	    nodeList.append(cNode);

The next and nexti Commands

Use the next command to execute a line of source code. When the next line to be executed contains a function call, the next command executes the function being called and stops the process at the line immediately after the function call.

Use the nexti command to execute a machine instruction. When the instruction contains a function call, the nexti command executes the function being called and stops the process at the instruction immediately after the call instruction.

For multithreaded applications, use these commands to step the current thread while putting all other threads on hold.

If you supply the optional expression argument, the debugger evaluates the expression as a positive integer that specifies the number of times to execute the command. The expression can be any expression that is valid in the current context.

step_over_command
        : next  [ step_number ]
        | nexti [ step_number ]

step_number
        :  expression

For example:

DBX Mode
(idb) next
stopped at [int main(void):192 0x804a0db]
    192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
(idb) next
stopped at [int main(void):193 0x804a182]
    193     nodeList.append(cNode1);
GDB Mode
(idb) next
192	    CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
(idb) next
193	    nodeList.append(cNode1);

The following example shows the difference between stepi and nexti over the same call:

DBX Mode
(idb) cont
[2] stopped at [void List<Node>::append(class Node* const):152 0x8058401]
    152         while (currentNode->getNextNode())
(idb) $curpc/4i
void List<Node>::append(class Node* const): x_list.cxx
*[line 152, 0x8058401]	append(class Node* const)+0x21:         pushl    %edi
 [line 152, 0x8058402]	append(class Node* const)+0x22:         movl     -4(%ebp), %eax
 [line 152, 0x8058405]	append(class Node* const)+0x25:         movl     %eax, (%esp)
 [line 152, 0x8058408]	append(class Node* const)+0x28:         call     getNextNode
(idb) stepi 3
stopped at [void List<Node>::append(class Node* const):152 0x8058408]	append(class Node* const)+0x28:         call     getNextNode
(idb) stepi
stopped at [class Node* Node::getNextNode(void):81 0x804a51c]	getNextNode:          pushl    %ebp
(idb) cont
[2] stopped at [void List<Node>::append(class Node* const):152 0x8058432]
    152         while (currentNode->getNextNode())
(idb) nexti
stopped at [void List<Node>::append(class Node* const):152 0x8058433]	append(class Node* const)+0x53:         movl     -4(%ebp), %eax
(idb) nexti
stopped at [void List<Node>::append(class Node* const):152 0x8058436]	append(class Node* const)+0x56:         movl     %eax, (%esp)
GDB Mode
(idb) continue
Continuing.

Breakpoint 2, List<Node>::append (this=0xbfffe95c, node=0x80b4f08) at x_list.cxx:152
152	        while (currentNode->getNextNode())
(idb) x /4i $pc
0x804aa6d <append+33>:         pushl    %edi
0x804aa6e <append+34>:         movl     -4(%ebp), %eax
0x804aa71 <append+37>:         movl     %eax, (%esp)
0x804aa74 <append+40>:         call     0x804a68c <getNextNode>
(idb) stepi 3
152	        while (currentNode->getNextNode())
(idb) stepi
Node::getNextNode () at x_list.cxx:81
81	Node* Node::getNextNode()            {return _nextNode; }
(idb) continue
Continuing.

Breakpoint 2, List<Node>::append (this=0xbfffe95c, node=0x80b4f08) at x_list.cxx:152
152	        while (currentNode->getNextNode())
(idb) nexti
152	        while (currentNode->getNextNode())
(idb) nexti
152	        while (currentNode->getNextNode())

The return Command

Use the return(dbx) and finish(gdb) command without an argument to continue execution of the current function until it returns to its caller.
Use return function_name(dbx) to continue the execution until control is returned to the specified function. The function must be active on the call stack.

In the following example, the next command is used to step through process execution in the append method. The return(dbx) / finish(gdb) command finishes the append method and returns control to the caller.

DBX Mode
(idb) next
stopped at [void List<Node>::append(class Node* const):154 0x805844b]
    154 	currentNode->setNextNode(node);
(idb) return
stopped at [int main(void):193 0x804a19a]
    193     nodeList.append(cNode1);
GDB Mode
(idb) next
154		currentNode->setNextNode(node);
(idb) finish
main () at x_list.cxx:193
193	    nodeList.append(cNode1);

The return(dbx) / finish(gdb) command is sensitive to the user's location in the call stack. Suppose function A calls function B, which calls function C. Execution has stopped in function C, and you entered the up command, so you were now in function B, at the point where it called function C. Using the return (dbx) / finish(gdb) command here would return you to function A, at the point where function A called function B. Functions B and C will have completed execution.

The cont Command

Use the cont(dbx) and continue(gdb) command without a parameter value to resume process execution until a breakpoint, a signal , an error, or normal process termination is encountered. Specify a signal parameter value to send an operating system signal to the process.

When you use the cont(dbx) and continue(gdb) command, the debugger resumes execution of the entire process.

In the following example, a cont(dbx) / continue(gdb) command resumes process execution after it was suspended by a breakpoint.

DBX Mode
(idb) list $curline - 5: 10
    188
    189     CompoundNode* cNode = new CompoundNode(12.345, 2);
    190     nodeList.append(cNode);
    191
    192     CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
>   193     nodeList.append(cNode1);
    194
    195     nodeList.append(new IntNode(3));
    196
    197     IntNode* newNode2 = new IntNode(4);
(idb) stop at 198
[#3: stop at "x_list.cxx":198 ]
(idb) cont
[3] stopped at [int main(void):198 0x804a2e6]
    198     nodeList.append(newNode2);
GDB Mode
(idb) list
183
184	    // add entries to list
185	    //
186	    IntNode* newNode = new IntNode(1);
187	    nodeList.append(newNode);
188
189	    CompoundNode* cNode = new CompoundNode(12.345, 2);
190	    nodeList.append(cNode);
191
192	    CompoundNode* cNode1 = new CompoundNode(3.1415, 7);
193	    nodeList.append(cNode1);
194
195	    nodeList.append(new IntNode(3));
196
197	    IntNode* newNode2 = new IntNode(4);
198	    nodeList.append(newNode2);
199
200	    CompoundNode* cNode2 = new CompoundNode(10.123, 5);
201	    nodeList.append(cNode2);
202
(idb) break 198
Breakpoint 3 at 0x804a456: file x_list.cxx, line 198.
(idb) continue
Continuing.

Breakpoint 3, main () at x_list.cxx:198
198	    nodeList.append(newNode2);
DBX Mode

The signal parameter value can be either a signal number or a string name (for example, SIGSEGV). The default is 0, which allows the process to continue execution without specifying a signal. If you specify a signal parameter value, the process continues execution with that signal.

Use the in argument to continue until the named function is reached. The function name must be valid. If the function name is overloaded and you do not resolve the scope of the function in the command line, the debugger prompts you with the list of overloaded functions bearing that name from which to choose.

Use the to parameter value to resume execution and then halt when the specified source line is reached. The form of the optional to parameter must be either:

  • quoted_filename:line_number, which explicitly identifies both the source file and the line number where execution is to be halted.
  • line_number, a positive numeric, which indicates the line number of the current source file where execution is to be halted.

You can repeat the cont command (n +1) times by entering n cont.

You can set a one-time breakpoint on an instruction address before continuing by entering conti to address_expression.

GDB Mode

Use the optional argument number_expression of the continue command to specify a further number of times to ignore a breakpoint at this location.

The until command continues running until a source line past the current line, in the current stack frame, is reached. This command is used to avoid single stepping through a loop more than once.
If place_detector is specified the command continues running until either the specified location is reached, or the current stack frame returns.

The goto Command

DBX Mode

The goto(dbx) command is intended for the users who want to 'skip over' the execution of a portion of source code.

cont_from_place_command
        : goto line_expression

line_expression
        : expression

Chapter 11 — Using Snapshots as an Undo Mechanism

DBX Mode

You can save the current state of the debuggee process in a snapshot, and later revert to that state and try a different set of steps. Conceptually speaking, this feature is similar to the undo function in text editors, except that with snapshots you have control of the granularity of each undo. See the Introduction for a quick overview.

snapshot_command
        : save_snapshot_command
        | clone_snapshot_command
        | show_snapshot_command
        | delete_snapshot_command

The following sections discuss these commands and address the limitations of snapshots.

The save snapshot Command

Use the save snapshot command to save the state of the current process in a snapshot. Snapshots are numbered sequentially starting from 1.

save_snapshot_command
        : save snapshot

In the following example, the first line of the save snapshot message shows the snapshot_number (1), the time it is saved, and the ID number of the process that implements the snapshot. The next two lines show the status of the snapshot.

(idb) save snapshot
# 1 saved at 13:27:54 (PID: 29077).
    stopped at [int main(void):182 0x1200023f8]
    182     List<Node> nodeList;

The clone snapshot Command

Use the clone snapshot command to revert the state of the debuggee process to that of a previously saved snapshot. By doing this, you can conveniently return to the state saved in the snapshot as opposed to rerunning the process and re-entering the debugger command sequence that brought you to that state.

Note that rerun and clone snapshot are different in that rerun always executes the process from the beginning, whereas clone snapshot does not execute the process at all; it simply duplicates the saved snapshot (using a mechanism similar to the fork system call) and behaves as though the process execution has stopped at the point when the snapshot was saved.

The clone snapshot command clones the snapshot specified by snapshot_id. If no snapshot_id is specified, the most-recently saved existing snapshot is cloned.

clone_snapshot_command
        : clone snapshot  [ snapshot_id ]

snapshot_id
        : expression

Cloning a snapshot has two side effects:

  • The snapshots created after the cloned snapshot are deleted. For example, suppose four snapshots are saved from a process. Cloning the second snapshot results in the deletion of the third and fourth snapshots.

  • The current process is killed and replaced by the clone process. Thus, if you enter show process after cloning a snapshot, you will see that the process ID of the current process has changed to that of the cloned process. For example:

    (idb) show process
    >localhost:29013 (/usr/examples/x_list) paused.
    (idb) clone snapshot
    Process has exited
    Process 29089 cloned from Snapshot 1.
    # 1 saved at 13:27:54 (PID: 29077).
        stopped at [int main(void):182 0x1200023f8]
        182     List<Node> nodeList;
    (idb) show process
    >localhost:29089 (/usr/examples/x_list) paused.
    

The show snapshot Command

Use the show snapshot * and show snapshot all commands to display all the snapshots that have been saved from the current process. Use show snapshot snapshot_id_list to display the snapshots specified. If no snapshots are specified, the most-recently saved existing snapshot is displayed.

show_snapshot_command
        : show snapshot [ snapshot_id_list ]

snapshot_id_list
        : snapshot_id ,...
        | all
        | *

snapshot_id
        : expression

For example:

(idb) show snapshot all
# 1 saved at 13:27:54 (PID: 29077).
    stopped at [int main(void):182 0x1200023f8]
    182     List<Node> nodeList;

The delete snapshot Command

Use the delete snapshot * and delete snapshot all commands to delete all the snapshots that have been saved from the current process. Use delete snapshot snapshot_id_list to delete the specified snapshots. If no snapshots are specified, the most-recently saved existing snapshot is deleted.

delete_snapshot_command
        : delete snapshot [ snapshot_id_list ]

snapshot_id_list
        : snapshot_id ,...
        | all
        | *

snapshot_id
        : expression

For example:

(idb) show snapshot all
# 1 saved at 13:27:54 (PID: 29077).
    stopped at [int main(void):182 0x1200023f8]
    182     List<Node> nodeList;
(idb) delete snapshot
(idb) show snapshot all
No snapshots have been saved.

Snapshot Limitations

Snapshots have the following limitations:

  • Snapshots are implemented by causing the process to fork. The state saved in a snapshot does not include I/O and forks. In other words, when you clone a snapshot, the I/O that has been done since the snapshot was saved is not undone; likewise, the child processes that have been spawned since the snapshot was saved are not killed.
  • The save snapshot command saves the state of the current process only. If you are doing multiprocess debugging, you might want to save a snapshot for each process.
  • Snapshots on a multithreaded process are not supported.
  • Snapshots are not supported for core file debugging.

Chapter 12 — Debugging Optimized Code

TBD

Chapter 13 — Support Limitations

This chapter contains sections which describe the limitations on support for the following languages:

Limitations on Support for C++

The debugger interprets C++ names and expressions using the ANSI standard rules. C++ is a distinct language, rather than a superset of C. Where the semantics of C and C++ differ, the debugger provides the interpretation appropriate for the language of the program being debugged.

To make the debugger more useful, it relaxes some standard C++ name visibility rules. For example, you can reference public, protected, and private class members.

The following limitations apply to debugging a C++ program:

  • If a program is not compiled to include debug information, do not set a breakpoint on an inline member function; it may confuse the debugger.
  • When you use the debugger to display virtual and inherited class information, the debugger does not support pointers to members of a class.
  • The debugger does not support calling the C++ constructs new and delete. As alternatives, use the malloc() and free() routines from C.
  • Sometimes the debugger does not see class type names with internal linkage, and it issues an error message stating that the name is overloaded.
  • You cannot select a version of an overloaded function that has a type signature containing ellipsis points (...).
  • Pointers to functions with type signatures that contain parameter list or ellipsis arguments are not supported.
  • The debugger does not support consideration of namespace using directives when performing name resolution.

Limitations for debugging templates include the following:

  • You cannot specify a template by name in a debugger command. You must use the name of an instantiation of the template.
  • Setting a breakpoint at a line number that is inside a template function will not necessarily stop at all instantiations of the function within the given file, but only at a randomly chosen few.

Limitations on Support for Fortran

The debugger and the operating system support the Fortran language with certain limitations, which are described in the following sections.

Be aware of the following data-type limitations when you debug a Fortran program:

  • The debugger does not allow setting a breakpoint on a program routine named MAIN.
  • Substring notation is not supported.

The following limitations apply only to Fortran 90:

  • Fortran 90 array constructors, structure constructors, adjustable arrays, and vector subscripts are not supported.
  • Fortran 90 user-defined (derived) operators are not supported.
  • The debugger does not handle variables of 16-bit character data types.
Limitations on Procedure Invocations

Following are the limitations on Fortran procedure invocations:

  • The debugger does not support invocations of user-defined procedures unless they have been compiled with debug information.
  • The debugger does not support complex or real*16 procedure return values.
  • The debugger does not support real*16 or complex*32 procedure arguments.

Part III
Advanced Topics

Part III provides information to make advanced use of the debugger by expanding on topics presented in Part II.

It is intended to be used as a reference and therefore contains a wide variation in the level of information provided in each section.

Chapter 14 — Preparing Your Program for Debugging

This chapter describes how you can modify your program to wait for the debugger.

Modifying Your Program to Wait for the Debugger

To modify your program to wait for the debugger, complete the following steps:

  1. Add some code similar to the following :

    void loopForDebugger()
    {
        if (!getenv("loopForDebugger")) return;
        static volatile int debuggerPresent = 0;
        while (!debuggerPresent);
    }
    
  2. Call this function at some convenient point.

  3. Use the debugger's attach capability to get the process under control.

When you have the process under debugger control, you can assign a non-zero value to debuggerPresent, and continue your program. For example:

% setenv loopForDebugger ""
% a.out arg1 arg2 &
% idb -pid 1709 a.out
^C
(idb) assign debuggerPresent = 1
...
(idb) cont

Chapter 15 — Debugger's Command Processing Structure

This chapter is divided into the following sections:

Lexical Elements of Commands

After the debugger assembles complete lines of input from the characters it reads from stdin or from the file specified in the source command (as described in Part II), each line is then converted into a sequence of lexical elements known as lexemes. For example, print(a++); is converted into the following lexemes:

  1. print
  2. (
  3. a
  4. ++
  5. )
  6. ;

The sequence of lexemes is then recognized as debugger commands containing language-specific expressions by a process known as parsing. The recognition is based on a set of rules known as a grammar. The debugger uses a single grammar for commands regardless of the current program's language, but it has language-specific subgrammars for some of the pieces of commands, such as names, expressions, and so on.

Lexical States

The debugger starts processing each line according to the rules of the LKEYWORD lexical state. It recognizes the longest lexeme it can, according to these rules. After recognizing the lexeme, it may stay in the same state, or it may change to a different lexical state with different rules.

The debugger moves between the following lexical states as it recognizes the lexemes:

Lexical State Description
LBPT Breakpoint commands have lexemes that change the lexical state to LBPT.
LEXPORT The command export changes the lexical state to LEXPORT. This state recognizes an evironment variable identifier.
LFILE Commands that require file names have lexemes that change the lexical state to LFILE so that things like mysrc/foo.txt are recognized as a file name, and not as a variable mysrc being divided by a structure data member foo.txt.
LFINT Commands that require either a file name or a process ID have lexemes to change the lexical state to LFINT.
LKEYWORD All but one of the debugger commands begin with one or more keywords. The exception is the examine command. The debugger recognizes the '{' and ';' lexemes that must precede a command, and uses those to reset to the LKEYWORD state for the beginning of the next command.
LLINE Commands that require arbitrary character input have lexemes that change the lexical state to LLINE.
LNORM Language expressions involve language-specific lexemes. The lexemes that precede an expression change the lexical state to LNORM, and then LNORM recognizes the language-specific lexemes.
LSETENV The command setenv changes the lexical state to LSETENV. This state recognizes an evironment variable identifier.
LSIGNAL Commands that require signal names have lexemes that change the lexical state to LSIGNAL.
LWORD Commands that require shell words have lexemes that change the lexical state to LWORD.

The rules that each lexical state uses are described in the following sections, in a format known as a lex regular expression. Rather than repeating some of descriptions, a set of common subrules is described first. After the common subrules, we describe the initial state, the rules, and for each recognized token, the next lexical state.

Identifiers

All languages have a concept of an identifier, composed of letters, digits, and other special characters. The debugger also uses keywords composed of letters; therefore, rules are required to determine which identifiers are actually debugger keywords.

All debugger commands, except examine, begin with leading keywords. Because the examine command begins with an expression, all identifiers must be recognized as such from both the LKEYWORD state that starts commands and the LNORM state that the debugger uses for processing expressions.

Some debugger commands have keywords embedded in them following expressions, and the ends of expressions are hard to recognize. You can use identifiers that have the same spelling as an embedded keyword simply by enclosing the whole expression in parentheses (()). For more information on using keywords within commands, see Section 4.4.4.

Furthermore, the C and C++ grammars need to know whether an identifier is a typedef or struct/class identifier. The debugger currently makes the determination at the time the whole command is parsed, rather than deferring the determination to when the expression itself is being evaluated. This can result in a misidentification of the identifier.

When in the following four lexical states, the debugger can recognize identifiers:

  • LKEYWORD, LNORM, LBPT
    Language Regular Expression
    C, C++, Fortran {LT}({LT}|{DG})*
    The state is changed to LNORM to process the rest of the expression.
  • LSIGNAL
    Language Regular Expression
    All{LT}({LT}|{DG})*
    The state is left as LSIGNAL to process the next signal.
If your operating system supports internationalization (I18N), {LT} equates to {LTI18N}.
Embedded Keywords

The complete set of embedded keywords follows:
Lexeme Representation Initial Lexical State Changed Lexical State Language Specific?
ANY any LNORM LNORM Shared by all
AT at LBPT, LNORM LNORM Shared by all
CHANGED changed LNORM LNORM Shared by all
IF if LBPT LNORM Shared by all
IN in LBPT, LNORM LNORM Shared by all
IN_ALL in{Whitespace}all {Whitespace} LBPT, LNORM LNORM Shared by all
POLICY policy LNORM LNORM Shared by all
PRIORITY priority LNORM LNORM Shared by all
READ read LNORM LNORM Shared by all
THREAD thread LNORM LNORM Shared by all
THREAD_ALL thread{Whitespace}all
thread{Whitespace} *
LNORM LNORM Shared by all
TO to LNORM LNORM Shared by all
WITH_STATE with{Whitespace}state LNORM LNORM Shared by all
WITHIN within LNORM LNORM Shared by all
WRITE write LNORM LNORM Shared by all

NOTE: THREAD is both a leading and an embedded keyword.

Leading Keywords
Leading keywords are recognized only at the beginning of commands. You do not need to use parentheses (()) to use them as a normal identifier, unless they occur at the start of an examine command.

Leading keywords may differ between languages. The complete set follows:
Lexeme Representation
(Some May Be Language Specific)
Initial Lexical State Changed Lexical State Language Specific?
ALIAS alias LKEYWORD LNORM Shared by all
ASSIGN assign LKEYWORD LNORM Shared by all
ATTACH attach LKEYWORD LNORM Shared by all
CALL call LKEYWORD LNORM Shared by all
CATCH catch LKEYWORD LSIGNAL Shared by all
CATCH_UNALIGN catch{Whitespace}unaligned LKEYWORD LNORM Shared by all
CLASS class LKEYWORD LNORM C++ Special Case
CLONE_SNAPSHOT clone{Whitespace}snapshot LKEYWORD LNORM Shared by all
CONDITION condition LKEYWORD LNORM Shared by all
CONT cont LKEYWORD LNORM Shared by all
CONTI conti LKEYWORD LNORM Shared by all
CONT_THREAD cont{Whitespace}thread LKEYWORD LNORM Shared by all
DELETE delete LKEYWORD LNORM Shared by all, also used for C++ special case
DELETE_ALL delete{Whitespace}*
delete{Whitespace}all
LKEYWORD LNORM Shared by all
DELSHAREDOBJ delsharedobj LKEYWORD LFILE Shared by all
DETACH detach LKEYWORD LNORM Shared by all
DISABLE disable LKEYWORD LNORM Shared by all
DISABLE_ALL disable{Whitespace}*
disable{Whitespace}all
LKEYWORD LNORM Shared by all
DOWN down LKEYWORD LNORM Shared by all
DUMP dump LKEYWORD LNORM Shared by all
EDIT edit LKEYWORD LFILE Shared by all
ELSE else LKEYWORD LKEYWORD Shared by all
ENABLE enable LKEYWORD LNORM Shared by all
ENABLE_ALL enable{Whitespace}*
enable{Whitespace}all
LKEYWORD LNORM Shared by all
EXPAND_AGGREGATED_MESSAGE expand{Whitespace}aggregated{Whitespace}message LKEYWORD LNORM Shared by all
EXPORT export LKEYWORD LNORM Shared by all
FILECMD file LKEYWORD LFILE Shared by all
FILEEXPR fileexpr LKEYWORD LFILE Shared by all
FOCUS focus LKEYWORD LNORM Shared by all
FOCUS_ALL focus{Whitespace}*
focus{Whitespace}all
LKEYWORD LNORM Shared by all
FUNC func LKEYWORD LNORM Shared by all
GOTO goto LKEYWORD LNORM Shared by all
HELP help LKEYWORD LLINE Shared by all
HISTORY history LKEYWORD LNORM Shared by all
HPFGET hpfget LKEYWORD LNORM Fortran
IF if LKEYWORD LNORM Shared by all
IGNORE ignore LKEYWORD LSIGNAL Shared by all
IGNORE_UNALIGN ignore{Whitespace}unaligned LKEYWORD LNORM Shared by all
INFO info LKEYWORD LKEYWORD Shared by all
INPUT input LKEYWORD LFILE Shared by all
IO io LKEYWORD LFILE Shared by all
KILL kill LKEYWORD LNORM Shared by all
KPS kps LKEYWORD LNORM Shared by all
LIST list LKEYWORD LNORM Shared by all
LISTOBJ listobj LKEYWORD LNORM Shared by all
LOAD load LKEYWORD LFILE Shared by all
MAP_SOURCE_DIRECTORY map{Whitespace}source{Whitespace}directory LKEYWORD LNORM Shared by all
MUTEX mutex LKEYWORD LNORM Shared by all
NEXT next LKEYWORD LNORM Shared by all
NEXTI nexti LKEYWORD LNORM Shared by all
OUTPUT output LKEYWORD LFILE Shared by all
PATCH patch LKEYWORD LNORM Shared by all
PLAYBACK playback LKEYWORD LKEYWORD Shared by all
POP pop LKEYWORD LNORM Shared by all
PRINT print LKEYWORD LNORM Shared by all
PRINTB printb LKEYWORD LNORM Shared by all
PRINTD printd LKEYWORD LNORM Shared by all
PRINTENV printenv LKEYWORD LNORM Shared by all
PRINTF printf LKEYWORD LNORM Shared by all
PRINTI printi LKEYWORD LNORM Shared by all
PRINTO printo LKEYWORD LNORM Shared by all
PRINTT printt LKEYWORD LNORM Shared by all
PRINTX printx LKEYWORD LNORM Shared by all
PRINTREGS printregs LKEYWORD LNORM Shared by all
PROCESS process LKEYWORD LNORM Shared by all
PROCESS_ALL process{Whitespace}*
process{Whitespace}all
LKEYWORD LNORM Shared by all
PTHREAD pthread LKEYWORD LNORM Shared by all
QUESTION ? LKEYWORD LNORM Shared by all
QUIT quit LKEYWORD LNORM Shared by all
READSHAREDOBJ readsharedobj LKEYWORD LFILE Shared by all
RECORD record LKEYWORD LKEYWORD Shared by all
RERUN rerun LKEYWORD LWORD Shared by all
RETURN return LKEYWORD LNORM Shared by all
RUN run LKEYWORD LWORD Shared by all
SAVE_SNAPSHOT save{Whitespace}snapshot LKEYWORD LNORM Shared by all
SET set LKEYWORD LNORM Shared by all
SETENV setenv LKEYWORD LNORM Shared by all
SH sh LKEYWORD LNORM Shared by all
SHOW show LKEYWORD LKEYWORD Shared by all
SHOW_AGGREGATED_MESSAGE show{Whitespace}aggregated{Whitespace}message LKEYWORD LNORM Shared by all
SHOW_AGGREGATED_MESSAGE_ALL show{Whitespace}aggregated{Whitespace}message{Whitespace}*
show{Whitespace}aggregated{Whitespace}message{Whitespace}all
LKEYWORD LNORM Shared by all
SHOW_PROCESS_SET show{Whitespace}process{Whitespace}set LKEYWORD LNORM Shared by all
SHOW_PROCESS_SET_ALL show{Whitespace}process{Whitespace}set{Whitespace}*
show{Whitespace}process{Whitespace}set{Whitespace}all
LKEYWORD LNORM Shared by all
SHOW_SOURCE_DIRECTORY show{Whitespace}source{Whitespace}directory LKEYWORD LNORM Shared by all
SHOW_ALL_SOURCE_DIRECTORY show{Whitespace}all{Whitespace}source{Whitespace}directory LKEYWORD LNORM Shared by all
SLASH / LKEYWORD LNORM Shared by all
SNAPSHOT snapshot LKEYWORD LNORM Shared by all
SNAPSHOT_ALL snapshot all LKEYWORD LNORM Shared by all
SNAPSHOT_* snapshot * LKEYWORD LNORM Shared by all
SOURCE source LKEYWORD LFILE Shared by all
STATUS status LKEYWORD LNORM Shared by all
STEP step LKEYWORD LNORM Shared by all
STEPI stepi LKEYWORD LNORM Shared by all
STOP stop LKEYWORD LBPT Shared by all
STOPI stopi LKEYWORD LNORM Shared by all
THREAD thread LKEYWORD LNORM Shared by all
TRACE trace LKEYWORD LNORM Shared by all
TRACEI tracei LKEYWORD LNORM Shared by all
UNALIAS unalias LKEYWORD LNORM Shared by all
UNLOAD unload LKEYWORD LNORM Shared by all
UNMAP_SOURCE_DIRECTORY unmap{Whitespace}source{Whitespace}directory LKEYWORD LNORM Shared by all
UNRECORD unrecord LKEYWORD LNORM Shared by all
UNSET unset LKEYWORD LNORM Shared by all
UNSETENV unsetenv LKEYWORD LNORM Shared by all
UNSETENV_ALL unsetenv{Whitespace}* LKEYWORD LNORM Shared by all
UNUSE unuse LKEYWORD LFILE Shared by all
UP up LKEYWORD LNORM Shared by all
USE use LKEYWORD LFILE Shared by all
VERSION version LKEYWORD LNORM Shared by all
WATCH watch LKEYWORD LNORM Shared by all
WATCH_MEMORY watch{Whitespace}memory LKEYWORD LNORM Shared by all
WATCH_VARIABLE watch{Whitespace}variable LKEYWORD LNORM Shared by all
WHATIS whatis LKEYWORD LNORM Shared by all
WHEN when LKEYWORD LBPTChapter Shared by all
WHENI wheni LKEYWORD LNORM Shared by all
WHERE where LKEYWORD LNORM Shared by all
WHEREIS whereis LKEYWORD LNORM Shared by all
WHERE_THREAD where{Whitespace}thread LKEYWORD LNORM Shared by all
WHERE_THREAD_ALL where{Whitespace}thread{Whitespace}*
where{Whitespace}thread{Whitespace}all
LKEYWORD LNORM Shared by all
WHICH which LKEYWORD LNORM Shared by all

Reserved Identifiers

Some identifiers are recognized as reserved words, regardless of whether they are inside parentheses (()).

The reserved words may differ between languages. The complete set follows:

Lexeme Representation
(Some May Be Language Specific)
Initial Lexical State Changed Lexical Language Specific?
CHAR char LNORM LNORM C, C++
CLASS class LNORM LNORM C++
CONST const LNORM LNORM C, C++
DELETE delete LNORM LNORM C++
DOUBLE double LNORM LNORM C, C++
ENUM enum LNORM LNORM C, C++
FLOAT float LNORM LNORM C, C++
INT int LNORM LNORM C, C++
LONG long LNORM LNORM C, C++
NEW new LNORM LNORM C++
OPERATOR operator LNORM LNORM C++
SHORT short LNORM LNORM C, C++
SIGNED signed LNORM LNORM C, C++
SIZEOF sizeof LNORM LNORM C, C++, Fortran
STRUCT struct LNORM LNORM C, C++
UNION union LNORM LNORM C, C++
UNSIGNED unsigned LNORM LNORM C, C++
VOID void LNORM LNORM C, C++
VOLATILE volatile LNORM LNORM C, C++
Lexemes Shared by All Languages
Because the debugger supports multiple languages, some of the rules must be language specific. To distinguish between the characters used for a particular language to represent a lexeme and the lexeme itself, the debugger names the lexemes, rather than using any one language's representation. For example, the lexeme GE corresponds to Fortran's '.GE.', and to C's '>='.

Some lexemes have the same representation in all languages, especially those that form part of the debugger commands apart from the language-specific expressions.

Common Elements of Lexemes
The following tables list common elements of lexemes.
ConceptRule RepresentationDescription
Decimal digit DG [0-9] One character from '0'..'9'.
Octal digit OC [0-7] One character from '0'..'7'.
Hexadecimal digit HX [0-9a-fA-F] Any of the characters '0'..'9' and any of the letters 'A'..'F' and 'a'..'f'.
Single letter LT [A-Za-z_$] Any of the characters 'A'..'Z', 'a'..'z', and the underscore (_) and dollar sign ($) characters.
Single letter
from the International Character Set
LT18N [A-Za-z_$\200-\377] Any of the characters 'A'..'Z', 'a'..'z', the underscore (_) and dollar sign ($) characters, and any character in the top half of the 8-bit character set.
Shell 'word' WD [^ \t;\n<>'"] Any character except space, tab, semicolon (;), linefeed, less than (<), greater than (>), and quotes (' or ").
File name FL [^ \t\n\}\;\>\<] Any character except space, tab, semicolon (;), linefeed, right brace (}), less than (<), greater than (>), and tick (`).
Optional exponent Exponent [eE][+-]?{DG}+ Numbers often allow an optional exponent. It is represented as an 'e' or 'E' followed by an optional plus (+) or minus (-), and then one or more decimal digits.
Whitespace Whitespace [ \t]+ Whitespace is often used to separate two lexemes that would otherwise be misconstrued as a single lexeme. For example, stop in is two keywords, but stopin is an identifier. Apart from this separating property, Whitespace is usually ignored. Whitespace is a sequence of one or more tabs or spaces.
String literal stringChar ([^"\\\n]|([\\]({simpleEscape}|
{octalEscape}|{hexEscape})))
Any character except the terminating quote character ("), or a newline (\n). If the character is a backslash (\), it is followed by an escaped sequence of characters.
Character literal charChar ([^'\\\n]|([\\]({simpleEscape}|
{octalEscape}|{hexEscape})))
Any character except the terminating quote (') character, or a newline (\n). If the character is a backslash (\), it is followed by an escaped sequence of characters.
Environment variable identifier EID [^ \t\n<>;='"&\|] Any character except space, tab, linefeed, less-than (<), greater-than (>), semicolon (;), equal sign (=), quotes (' or "), ampersand (&), backslash (\), and bar (|).
Universal character name UCN \\u{HX}{4}|\\U{HX}{8} A universal character name is a backslash (\) followed by either a lowercase 'u' and 4 hexadecimal digits, or an uppercase 'U' and 8 hexadecimal digits.

The escaped sequence of characters can be one of following three forms:
ConceptRule RepresentationDescription
Simple escape simpleEscape ([A-Za-z'"?#*\\]) One of 'A'-'Z' or 'a'-'z'. Some of these have special meanings, the most common being 'n' for newline and 't' for tab. Can be a quote (' or ") character that does not finish the literal, a question mark (?), a pound sign (#), an asterisk (*), or a backslash (\), which then becomes part of the string literal rather than causing a further escape sequence.
Octal escape octalEscape (OC{1,3}) 1 to 3 octal digits, the combined numeric value of which is the character that becomes part of the string literal.
Hexadecimal escape hexEscape ([xX]HX{1,8}) An 'x' or an 'X' followed by 1 to 8 hexadecimal digits, the combined numeric value of which is the character that becomes part of the string literal.

Whitespace and Command-Separating Lexemes Shared by All Languages
In all lexical states, unescaped newlines produce the NEWLINE token and change the lexical state to be LKEYWORD.
Initial State: LKEYWORD, LNORM, LFILE, LLINE, LWORD, LSIGNAL, LBPT
Regular Expression:[\n]
Lexeme:NEWLINE
Change to State:LKEYWORD

In all lexical states except LLINE, a semicolon also changes the lexical state to be LKEYWORD.
This is because SEMICOLON is the command separator.
Initial State: LKEYWORD, LNORM, LFILE, LSIGNAL, LBPT, LWORD
Regular Expression:";"
Lexeme:SEMICOLON
Change to State:LKEYWORD

Commands can be nested, and the following transitions support this:
Initial State: LNORM
Regular Expression:"{"
Lexeme:LBRACE
Change to State:LKEYWORD
Initial State: LKEYWORD, LNORM, LFILE, LSIGNAL, LBPT
Regular Expression:"}"
Lexeme:RBRACE
Change to State:LKEYWORD

In most lexical states, the spaces, tabs, and escaped newlines are ignored. In the LLINE state, the spaces and tabs are part of the line, but escaped newlines are still ignored. In the LWORD state, the spaces and tabs are ignored, but escaped newlines are not.
Initial State: LKEYWORD, LNORM, LFILE, LSIGNAL, LBPT
Regular Expression:[ \t]
\\\n
Lexeme:Ignored
Change to State:Unchanged
Initial State: LLINE
Regular Expression:\\\n
Lexeme:Ignored
Change to State:Unchanged
Initial State: LWORD
Regular Expression:[ \t]
Lexeme:Ignored
Change to State:Unchanged

LNORM Lexemes Shared by All Languages
The state stays in LNORM.
LexemeRegular Expression
ANY any
AT at
ATSIGN "@"
CHANGED changed
CHARACTERconstant [lL][']{charChar}+[']
COLON ":"
COMMA ","
DOLLAR "$"
DOT "."
GE ">="
GREATER ">"
HASH unknown
IF if
IN in
IN_ALL in{Whitespace}all{Whitespace}
LE "<="
LESS "<"
LPAREN "("
POLICY policy
PRIORITY priority
RPAREN ")"
READ read
SLASH "/"
STAR "*"
STATE state
STRINGliteral ["]{stringChar}*["]
THREAD thread
THREAD_ALL thread{Whitespace}all
thread{Whitespace}"*"
TICK "`"
TO to
WIDECHARACTERconstant [lL][']{charChar}+[']
WIDESTRINGliteral [lL]["]{stringChar}*["]
WITH with
WITHIN within
WRITE write
LBPT Lexemes Shared by All Languages
LexemeRegular Expression Initial Lexical StateChanged Lexical State
IN in LBPTLNORM
IN_ALL in{Whitespace}all LBPTLNORM
AT at LBPTLNORM
PC_IS pc LBPTLNORM
SIGNAL signal LBPTLNORM
UNALIGNED unaligned LBPTLNORM
VARIABLE variable LBPTLNORM
MEMORY memory LBPTLNORM
EVERY_INSTRUCTION every{Whitespace}instructionLBPTLNORM
EVERY_PROC_ENTRY every{Whitespace}proc[edure]{Whitespace}entry LBPTLNORM
QUIET quiet LBPTLBPT
LFILE Lexemes Shared by All Languages
Files are one or more characters that can appear in a file name.

The state is left as LFILE, so that commands such as use and unuse can have lists of files.
LexemeRegular Expression
FILENAME {FL}+

LKEYWORD Lexemes Shared by All Languages
The state remains in LKEYWORD.
LexemeRegular Expression
INTEGERconstant "0"{OC}+
"0"[xX]{HX}+
{DG}+
LLINE Lexemes Shared by All Languages
All characters up to the next newline are assembled into a STRINGliteral.
LWORD Lexemes Shared by All Languages
Once the lexical state has been set to LWORD, it will stay there until a NEWLINE or a SEMICOLON is found. Both of these cause the lexical state to become LKEYWORD again. The individual words recognized can be any of the following, but in each case, the state stays LWORD.
LexemeRegular Expression
GREATER ">"
LESS "<"
GREATERAMPERSAND ">&"
ONEGREATER "1>"
TWOGREATER "2>"
STRINGliteral [']{charChar}*[']
["]{stringChar}*["]
STRINGliteral {WD}* that does not end in a backslash
WIDECHARACTERconstant [lL][']{charChar}+[']
WIDESTRINGliteral [lL]["]{stringChar}*["]
LSIGNAL Lexemes Shared by All Languages
The state stays in LSIGNAL.
LexemeRegular Expression
INTEGERconstant {DG}+
IDENTIFIER {LT}({LT}|{DG})*
Lexemes That Are Represented Differently in Each Language
Lexeme Representation
(Some May Be Language Specific)
Initial Lexical State Changed Lexical State Language Specific?
AMPERSAND "&" LNORMUnchanged C, C++, Fortran
ANDAND "&&" LNORMUnchanged C, C++
ANDassign "&=" LNORMUnchanged C, C++
ARROW "->" LNORM Unchanged C, C++
ARROWstar "->*" LNORM Unchanged C++
ASSIGNOP "=" LNORM Unchanged C, C++, Fortran
BRACKETS "[]" LNORM Unchanged C, C++
CLCL "::" LNORM Unchanged C++
DECR "--" LNORM Unchanged C, C++
DIVassign "/=" LNORM Unchanged C, C++
DOTstar ".*" LNORM Unchanged C++
ELLIPSIS "..." LNORM Unchanged C++
EQ "=="
".EQ."
(IS[ \t]+)?
  ("="|("EQUAL"([ \t]+"TO")?))
LNORM Unchanged C, C++, Fortran
Fortran
 
ERassign "^=" LNORM Unchanged C, C++
GE ".GE."
(IS[ \t]+)?
  "NOT"[ \t]+
  ("<"|("LESS"([ \t]+"THAN")?))
(IS[ \t]+)?
  (">="|("GREATER"([ \t]+"THAN")?[ \t]
  +"OR"[ \t]+"EQUAL"([ \t]+"TO")?))
LNORM Unchanged Fortran
 
 
 
 
GREATER ".GT."
(IS[ \t]+)?
  (">"|("GREATER"([ \t]+"THAN")?))
LNORM Unchanged Fortran
 
HAT "^" LNORM Unchanged C, C++
INCR "++" LNORM Unchanged C, C++
LBRACKET "[" LNORM Unchanged C, C++, Fortran
LE ".LE."
(IS[ \t]+)?"NOT"[ \t]+
  (">"|("GREATER"([ \t]+"THAN")?))
(IS[ \t]+)?
  ("<="|("LESS"([ \t]+"THAN")?[ \t]+
  "OR"[ \t]+"EQUAL"([ \t]+"TO")?))
LNORM Unchanged Fortran
 
 
 
LESS ".LT."
(IS[ \t]+)?
  ("<"|("LESS"([ \t]+"THAN")?))
LNORM Unchanged Fortran
 
LOGAND ".AND." LNORM Unchanged Fortran
LOGEQV ".EQV." LNORM Unchanged Fortran
LOGNEQV ".NEQV." LNORM Unchanged Fortran
LOGNOT ".NOT." LNORM Unchanged Fortran
LOGOR ".OR." LNORM Unchanged Fortran
LOGXOR ".XOR." LNORM Unchanged Fortran
LS "<<" LNORM Unchanged C, C++
LSassign "<<=" LNORM Unchanged C, C++
MINUS "-" LNORM Unchanged C, C++, Fortran
MINUSassign "-=" LNORM Unchanged C, C++
MOD "%"
MOD
LNORM Unchanged C, C++
MODassign "%=" LNORM Unchanged C, C++
MULTassign "*=" LNORM Unchanged C, C++
NE "!="
".NE."
"/="
(IS[ \t]+)?
  "NOT"[ \t]+("="|("EQUAL"([ \t]+"TO")?))
LNORM Unchanged C, C++
Fortran
NOT "!"
NOT
LNORM Unchanged C, C++
OPENSLASH "(/" LNORM Unchanged Fortran
OR "|"
OR
LNORM Unchanged C, C++
OROR "||" LNORM Unchanged C, C++
ORassign "|=" LNORM Unchanged C, C++
PARENS "()" LNORM Unchanged C++
PERCENT "%" LNORM Unchanged Fortran
PLUS "+" LNORM Unchanged C, C++, Fortran
PLUSassign "+=" LNORM Unchanged C, C++
QUESTION "?" LNORM Unchanged C, C++
RBRACKET "]" LNORM Unchanged C, C++, Fortran
RS ">>" LNORM Unchanged C, C++
RSassign ">>=" LNORM Unchanged C, C++
SLASHCLOSE "/)" LNORM Unchanged Fortran
SLASHSLASH "//" LNORM Unchanged Fortran
STARSTAR "**" LNORM Unchanged Fortran
TWIDDLE "~" LNORM Unchanged C, C++
LKEYWORD Lexemes Specific to C++
If a C++ identifier is followed by a "::", it is assumed to be a class or namespace identifier.

If a C++ identifier is followed by a "<", complex and dubious checks are made to try to match a complete template instance specifier.

LNORM Lexemes Specific to C and C++
The lexemes in the following table are specific to C and C++. The state stays in LNORM.
Lexeme Representation Language
AMPERSAND "&" C, C++
ANDAND "&&" C, C++
ANDassign "&=" C, C++
ARROW "->" C, C++
ASSIGNOP "=" C, C++
BRACKETS "[]" C, C++
CHAR char C, C++
CONST const C, C++
DECR "--" C, C++
DIVassign "/=" C, C++
DOUBLE double C, C++
ENUM enum C, C++
EQ "==" C, C++
ERassign "^=" C, C++
FLOAT float C, C++
FLOATINGconstant {DG}*"."{DG}*
{DG}*"."{DG}* {Whitespace}?{Exponent}
{DG}+{Whitespace}?{Exponent }
C, C++
HAT "^" C, C++
IDENTIFIER, TYPEDEFname {LT}|{UCN})({LT}|{UCN}|{DG})* C, C++
INCR "++" C, C++
INT int C, C++
INTEGERconstant "0"{OC}+
"0"[xX]{HX}+
{DG}+
C, C++
LBRACKET "[" C, C++
LONG long C, C++
LS "<<" C, C++
LSassign "<<=" C, C++
MINUS "-" C, C++
MINUSassign "-=" C, C++
MOD "%" C, C++
MODassign "%=" C, C++
MULTassign "*=" C, C++
NE "!=" C, C++
NOT "!" C, C++
OR "|" C, C++
ORassign "|=" C, C++
OROR "||" C, C++
PLUS "+" C, C++
PLUSassign "+=" C, C++
QUESTION "?" C, C++
RBRACKET "]" C, C++
RS ">>" C, C++
RSassign ">>=" C, C++
SHORT short C, C++
SIGNED signed C, C++
SIZEOF sizeof C, C++
STRUCT struct C, C++
TWIDDLE "~" C, C++
UNION union C, C++
UNSIGNED unsigned C, C++
VOID void C, C++
VOLATILE volatile C, C++

The lexemes in the following table are specific to C++. The state stays in LNORM.
Lexeme Representation Language
ARROWstar "->*" C++
CLASS class C++
CLCL "::" C++
DELETE delete C++
DOTstar ".*" C++
ELLIPSIS "..." C++
NEW new C++
OPERATOR operator C++
PARENS "()" C++
THIS this C++

LNORM Lexemes Specific to Fortran
The lexemes in the following table are specific to Fortran. The state stays in LNORM.
Lexeme Representation
AMPERSAND "&"
ASSIGNOP "="
CHARACTERconstantWithKind {FortranString}
{FortranCharacterKind}{FortranString}
{FortranCharacterKind}?{FortranCString}
CharDoubleDelim [^"\\\n]|("")
CharSingleDelim [^'\\\n]|('')
CStringDoubleDelim ["]({CharDoubleDelim}|{FortranEscapeChar})*["]
CStringSingleDelim [']({CharSingleDelim}|{FortranEscapeChar})*[']
ExponentVal [+-]?{DG}+
EQ ".EQ."
EQ "=="
FortranBinaryValue [Bb]((['][01]+['])|(["][01]+["]))
FortranCharacterKind {FortranCharacterNamedKind}
{FortranCharacterNumericKind}
FortranCharacterNamedKind {FortranName}"_"
FortranCharacterNumericKind {DG}+"_"
FortranCString ({CStringSingleDelim}|{CStringDoubleDelim})[Cc]
FortranEscapeChar [\\]([AaBbFfNnRrTtVv]|{FortranOctalEscape}
                         |{FortranHexEscape}|0|[\\])
FortranHexEscape [Xx]{HX}{1,2}
FortranHexValue [Zz](([']{HX}+['])|(["]{HX}+["]))
FortranHexValueAlternative (([']{HX}+['])|(["]{HX}+["]))[Xx]
FortranKind {FortranNamedKind}
{FortranNumericKind}
FortranName [A-Za-z$]({LT}|{DG})*
FortranNamedKind "_"{FortranName}
FortranNumericKind "_"{DG}+
FortranOctalEscape {OC}{1,3}
FortranOctalValue [Oo](([']{OC}+['])|(["]{OC}+["]))
FortranOctalValueAlternative (([']{OC}+['])|(["]{OC}+["]))[Oo]
FortranString {StringSingleDelim}
{StringDoubleDelim}
GE ".GE."
GREATER ".GT."
IDENTIFIER, TYPEDEFname {LT}({LT}|{DG})*
INTEGERconstant ".TRUE."
".FALSE."
"0"{OC}+
"0X"{HX}+
{DG}+
INTEGERconstantWithKind {DG}+{FortranKind}
{DG}*"#"[0-9A-Za-z]+
{FortranBinaryValue}
{FortranOctalValue}
{FortranHexValue}
{FortranOctalValueAlternative}
{FortranHexValueAlternative}
LBRACKET "["
LE ".LE."
LESS ".LT."
LOGAND ".AND."
LOGEQV ".EQV."
LOGICALconstantWithKind ".TRUE."{FortranKind}?
".FALSE."{FortranKind}?
LOGNEQV ".NEQV."
LOGNOT ".NOT."
LOGOR ".OR."
LOGXOR ".XOR."
MINUS "-"
NE ".NE."
NE "/="
OPENSLASH "(/"
PERCENT "%"
PLUS "+"
RBRACKET "]"
RealConstant {RealSingleConstant}
{RealDoubleConstant}
{RealQuadConstant}
REALconstantWithKind RealConstant
RealDExponent [Dd]{ExponentVal}
RealDoubleConstant ({DG}+|{RealWithDecimal}){RealDExponent}
RealEExponent [Ee]{ExponentVal}
RealQExponent [Qq]{ExponentVal}
RealQuadConstant ({DG}+|{RealWithDecimal}){RealQExponent}
RealSingleConstant (({DG}+{RealEExponent})
({RealWithDecimal}{RealEExponent}?)){FortranKind}?
RealWithDecimal ({DG}+"."DG}*)
({DG}*"."{DG}+)
SIZEOF sizeof
SLASHCLOSE "/)"
SLASHSLASH "//"
STARSTAR "**"
StringDoubleDelim ["]({CharDoubleDelim}|[\\])*["]
StringSingleDelim [']({CharSingleDelim}|[\\])*[']

Grammar of Commands

Most of the grammar for commands has already been given in previous sections. This section concentrates on the grammar for expressions.
Names and Expressions Within Commands
The exact syntax of expressions is specific to the current language.
expression
	: expression for C
        | expression for C++
        | expression for Fortran
Often you can omit an expression from a command or use a convenient default instead, to change the meaning of a command.
expression-opt
        : [ expression ]
Identifiers, Keyword, and Typedef Names
The debugger uses the normal language lookup rules for identifiers, (obeying scopes, and so on,) but also extends those rules as follows:
  • All global variables are visible.
  • If the debugger cannot find the identifier within the current lexical scopes, it will successively search the lexical scopes of each of the first $framesearchlimit (default is 0) callers.
These rules can be subverted by rescoping the name.

NOTE: The debugger does not know where in the scope a declaration occurred, so all lookups consider all identifiers in the scope, whether or not they occurred before the current line.

The lexical tokens for identifiers are specific to the current language, and also to the current lexical state.

IDENTIFIER
        : identifier for LSIGNAL lexical state
        | identifier for C
        | identifier for C++
        | identifier for Fortran
TYPEDEFnames are lexically just identifiers, but when looking them up in the current scope, the debugger determines that they refer to types, such as TYPEDEFs, classes, or structs. This information is needed to correctly parse C and C++ expressions.
TYPEDEFname
        : IDENTIFIER
A few lexical tokens act as embedded keywords in some positions within expressions, but the debugger generally tries to accept them as though they were normal identifiers.
identifier-or-key-word
        : IDENTIFIER
        | embedded-key-word

embedded-key-word
        : ANY
        | CHANGED
        | READ
        | WRITE
In other contexts, the debugger is also prepared to accept TYPEDEFnames (for example, int or the name of a class).
identifier-or-typedef-name
        : identifier-or-typedef-name for C
        | identifier-or-typedef-name for C++
        | identifier-or-typedef-name for Fortran
Integer Constants
The lexical tokens for integer constants are specific to the current language.
integer_constant
    	: INTEGERconstant for C and C++
    	| INTEGERconstant for Fortran
Macros
The debugger does not currently understand usages of macros, for example, uses of C and C++ preprocessor #define macros, and so on.
Calls
You can call any function whose address can be taken, provided that the parameters can also be passed, and the result returned.
call-expression
	: call-expression for C
	| call-expression for C++
	| call-expression for Fortran
Parameters
Each language may impose its own restrictions on exactly what can be passed as a parameter.

Any expression can be passed 'by value', but C++ constructors and destructors will not be invoked. Evaluating parameters can involve evaluating nested calls.

Anything whose address can be taken can be passed 'by reference'.

The debugger has very limited understanding of array descriptors.

Comma is both the argument separator and a valid operator in C and C++. Hence, argument lists are comma-separated assignment-expressions rather than full expressions.

argument-expression-list
        : assignment-expression
        | assignment-expression COMMA argument-expression-list

arg-expression-list-opt
        : [ argument-expression-list ]

assignment-expression
        : assignment-expression for C
        | assignment-expression for C++
        | assignment-expression for Fortran
Return Results
Any scalar or structure type can be the return result of a called function. Some simple array types are also supported, but the general cases are not.

The C++ constructors and destructors are not invoked, which may cause problems.

Addresses
You can take the addresses of variables and other data that are in memory, and functions that have had code generated for them. You can also take the address of a line of source code.

Some variables may be in registers; you cannot take their addresses.

The optimizing compilers may move variables from one memory location to another, in which case you will obtain the address of the current memory location of the variable.

The optimizing compilers may eliminate unused functions, as well as functions that have had all calls inlined. Static functions in header files may result in multiple copies of the code, and the address will be of only one of those copies.

The optimizing compilers and linkers may skip some instructions on the way in during a call, so a breakpoint on the first few instructions may not be hit. When you set a breakpoint on a function, the debugger sets it deeper in the function, at the end of the entry sequence, to try to avoid this.

The address of a line of source code is the address of the first instruction in memory that came from this line, but this instruction may be branched around, so it might not be executed before any other instruction from the same line.

Address of a Source Line
The debugger has extended the syntax of most languages to allow you to get the address of the first instruction that a source line generates. If you do not specify a file via the string, then the current file is used. If you specify a DOLLAR as the line-number, then the last line in the file that generated any instructions is used.
line-address
        : ATSIGN string COLON line-number
        | ATSIGN line-number

line-number
        : INTEGERconstant
        | DOLLAR
Other Modified Forms of Expressions
The whatis_command supports supersets of the normal expression syntax of the language.
whatis-expressions
        : whatis-expressions for C
        | whatis-expressions for C++
        | whatis-expressions for Fortran
Some commands (notably the examine command and the cont command) have a syntax that inhibits the use of a full expression. In this case, a more limited form of expression is still allowed.
address-exp
        : address-exp for C
        | address-exp for C++
        | address-exp for Fortran
The cont command and the change_stack_frame_commands have a form that specifies where to continue to, or where to cut the stack back to.
loc
        : loc for C
        | loc for C++
        | loc for Fortran
The target of a modifying_command can only be a subset of the possible expressions, known as a unary-expression.
unary-expression
        : unary-expression for C
        | unary-expression for C++
        | unary-expression for Fortran
Strings
The syntax of strings is sensitive to the current lexical state and language.
string
        : LNORM string
        | LLINE string
        | LWORD string
Most of the languages have places where they allow a series of string literals to be equivalent to a single string formed of their concatenated characters.
string-literal-list
        : string-literal-list for C
        | string-literal-list for C++
Rescoped Expressions
Sometimes the normal language visibility rules are not sufficient for specifying the variable, function, and so on, to which you may want to refer. The debugger extends the language's idea of an expression with an additional possibility called a rescoped expression.

Rescoped expressions cause the debugger to look up the identifiers and so on in the qual-symbol-opt, as though it were in the source file specified by the preceding filename-tick or qual-symbol-opt.

rescoped-expression
        : filename-tick qual-symbol-opt
        | TICK qual-symbol-opt

rescoped-typedef
        : filename-tick qual-typedef-opt
        | TICK qual-typedef-opt

filename-tick
        : string-tick

string-tick
        : string TICK

qual-symbol-opt
        : expression                        /* Base (global) name */
        | qual-symbol-opt TICK expression   /* Qualified name */

qual-typedef-opt
        : qual-typedef-opt for C
        | qual-typedef-opt for C++
        | qual-typedef-opt for Fortran
In the following example, rescoped expressions are used to distinguish which x the user is querying, because there are two variables named x (one local to main and one global):
(idb) list $curline - 10: 20
      1 long x = 5;             // global x
      2
      3 int main()
      4 {
      5     int x = 7;          // local x
      6     int y = x - ::x;
>     7     return (y);
      8 }
By default, a local variable is found before a global one, so that the plain x refers to the local variable.
(idb) whatis x
int x
(idb) which x
"x_rescoped.cxx"`main`x
(idb) whatis "x_rescoped.cxx"`main`x
int x
You may use the C++ :: operator to specify the global x in C++ code or rescoped expressions in any language.
(idb) whatis ::x
long x
(idb) whatis "x_rescoped.cxx"`x
long x
(idb) print "x_rescoped.cxx"`x
5

In the following example, the x variable is used in the following places to demonstrate how rescoping expressions can find the correct variable:

  • As a variable local to main
  • As a member variable of the class Foo
  • As a global variable
  • As a local variable to Foo's member function SetandGet
  • As a local variable to the CastAndAdd function, but visible as a parameter
(idb) list $curline - 10: 20
     10 double x = 3.1415;
     11
     12 int CastAndAdd(char x) {
     13     int result = ((int)::x) + x;
     14     return result;
     15 }
     16
     17 float Foo::SetandGet() {    // multiple scopes!
     18     long x = (long)::x;     // local x = global x
     19     Foo::x = (float)x;      // member x = local x
>    20     return Foo::x;          // return member x
     21 }
     22
     23 int main () {
     24     int x = 7;
     25     x -= CastAndAdd((char)1);
     26
     27     Foo thefoo;
     28     x -= (int)thefoo.SetandGet();
     29     return x;
(idb) whatis x
long x
(idb) which x
"x_rescoped2.cxx"`Foo::SetandGet`x
(idb) whatis ::x
double x
(idb) whatis Foo::x
float Foo::x
(idb) whatis main`x
int x
(idb) whatis CastAndAdd`x
char x
Printable Types
The lexical tokens for printable types are specific to the current language.
printable-type
    	: printable-type for C
    	| printable-type for C++
    	| printable-type for Fortran
Expressions Specific to C
The debugger has an almost complete understanding of C expressions, given the general restrictions.
expression
        : assignment-expression

constant-expression
        : conditional-expression
C Identifiers
The lookup rules are almost always correct for C.
identifier-or-typedef-name
        : identifier-or-key-word
        | TYPEDEFname
C Constants
The numeric constants are treated exactly the same as in C. The enumeration constant identifiers go though the same grammar paths as variable identifiers, which has basically the same effect as the C semantics.
primary-expression
        : identifier-or-key-word
        | constant
        | string-literal-list
        | LPAREN expression RPAREN
        | process_set
        | LPAREN process_range RPAREN

string-literal-list
        : string
        | string-literal-list string

constant
        : FLOATINGconstant
        | INTEGERconstant
        | CHARACTERconstant
        | WIDECHARACTERconstant
        | WIDESTRINGliteral
C Rescoped Expressions
The C implementation of rescoped expressions is the following:
qual-typedef-opt
        : TYPEDEFname
        | qual-typedef-opt TICK TYPEDEFname

whatis-expressions
        : expression
        | rescoped-expression
        | printable-type
C Calls
Following is the C implementation of calls.
call-expression
        : expression

function-call
        : postfix-expression LPAREN [arg-expression-list] RPAREN
Restrictions and limits are documented here.
C Addresses
Following is the C implementation of addresses.
address
        : AMPERSAND postfix-expression
        | line-address
        | postfix-expression

address-exp
        : address
        | address-exp PLUS  address
        | address-exp MINUS address
        | address-exp STAR  address
Restrictions and limits are documented here.
C Loc Specifications
The C implementation of loc is the following:
loc
        : expression
        | rescoped-expression
C Types
The debugger understands the full C type specification grammar.
type-specifier
        : basic-type-specifier
        | struct-union-enum-type-specifier
        | typedef-type-specifier

basic-type-specifier
        : basic-type-name
        | type-qualifier-list basic-type-name
        | basic-type-specifier type-qualifier
        | basic-type-specifier basic-type-name

type-qualifier-list
        : type-qualifier
        | type-qualifier-list type-qualifier

type-qualifier
        : CONST
        | VOLATILE

basic-type-name
        : VOID
        | CHAR
        | SHORT
        | INT
        | LONG
        | FLOAT
        | DOUBLE
        | SIGNED
        | UNSIGNED

printable-type
        : rescoped_typedef
        | type_name

struct-union-enum-type-specifier
        : elaborated-type-name
        | type-qualifier-list elaborated-type-name
        | struct-union-enum-type-specifier type-qualifier

typedef-type-specifier
        : TYPEDEFname
        | type-qualifier-list TYPEDEFname
        | typedef-type-specifier type-qualifier

elaborated-type-name
        : struct-or-union-specifier
        | enum-specifier

struct-or-union-specifier
        : struct-or-union opt-parenthesized-identifier-or-typedef-name

opt-parenthesized-identifier-or-typedef-name
        : identifier-or-typedef-name
        | LPAREN opt-parenthesized-identifier-or-typedef-name RPAREN

struct-or-union
        : STRUCT
        | UNION

enum-specifier
        : ENUM identifier-or-typedef-name

type-name
        : type-specifier
        | type-specifier abstract-declarator
        | type-qualifier-list                       // Implicit "int"
        | type-qualifier-list abstract-declarator   // Implicit "int"

type-name-list
        : type-name
        | type-name COMMA type-name-list

abstract-declarator
        : unary-abstract-declarator
        | postfix-abstract-declarator
        | postfixing-abstract-declarator

postfixing-abstract-declarator
        : array-abstract-declarator
        | LPAREN RPAREN

array-abstract-declarator
        : BRACKETS
        | LBRACKET constant-expression RBRACKET
        | array-abstract-declarator LBRACKET constant-expression RBRACKET

unary-abstract-declarator
        : STAR
        | STAR type-qualifier-list
        | STAR abstract-declarator
        | STAR type-qualifier-list abstract-declarator

postfix-abstract-declarator
        : LPAREN unary-abstract-declarator RPAREN
        | LPAREN postfix-abstract-declarator RPAREN
        | LPAREN postfixing-abstract-declarator RPAREN
        | LPAREN unary-abstract-declarator RPAREN postfixing-abstract-declarator
C Other Forms of Expressions
The following expressions all have their usual C semantics:
assignment-expression
        : conditional-expression
        | unary-expression ASSIGNOP    assignment-expression
        | unary-expression MULTassign  assignment-expression
        | unary-expression DIVassign   assignment-expression
        | unary-expression MODassign   assignment-expression
        | unary-expression PLUSassign  assignment-expression
        | unary-expression MINUSassign assignment-expression
        | unary-expression LSassign    assignment-expression
        | unary-expression RSassign    assignment-expression
        | unary-expression ANDassign   assignment-expression
        | unary-expression ERassign    assignment-expression
        | unary-expression ORassign    assignment-expression

conditional-expression
        : logical-OR-expression
        | logical-OR-expression QUESTION expression COLON conditional-expression

logical-OR-expression
        : logical-AND-expression
        | logical-OR-expression OROR logical-AND-expression

logical-AND-expression
        : inclusive-OR-expression
        | logical-AND-expression ANDAND inclusive-OR-expression

inclusive-OR-expression
        : exclusive-OR-expression
        | inclusive-OR-expression OR exclusive-OR-expression

exclusive-OR-expression
        : AND-expression
        | exclusive-OR-expression HAT AND-expression

AND-expression
        : equality-expression
        | AND-expression AMPERSAND equality-expression

equality-expression
        : relational-expression
        | equality-expression EQ relational-expression
        | equality-expression NE relational-expression

relational-expression
        : shift-expression
        | relational-expression LESS    shift-expression
        | relational-expression GREATER shift-expression
        | relational-expression LE      shift-expression
        | relational-expression GE      shift-expression

shift-expression
        : additive-expression
        | shift-expression LS additive-expression
        | shift-expression RS additive-expression

additive-expression
        : multiplicative-expression
        | additive-expression PLUS  multiplicative-expression
        | additive-expression MINUS multiplicative-expression

multiplicative-expression
        : cast-expression
        | multiplicative-expression STAR  cast-expression
        | multiplicative-expression SLASH cast-expression
        | multiplicative-expression MOD   cast-expression

cast-expression
        : unary-expression
        | LPAREN type-name RPAREN cast-expression

unary-expression
        : postfix-expression
        | INCR unary-expression
        | DECR unary-expression
        | AMPERSAND cast-expression
        | STAR cast-expression
        | PLUS cast-expression
        | MINUS cast-expression
        | TWIDDLE cast-expression
        | NOT cast-expression
        | SIZEOF unary-expression
        | SIZEOF LPAREN type-name RPAREN
        | line-address

postfix-expression
        : primary-expression
        | postfix-expression LBRACKET expression RBRACKET
        | function-call
	| postfix-expression LPAREN type-name-list RPAREN
        | postfix-expression DOT   identifier-or-typedef-name
        | postfix-expression ARROW identifier-or-typedef-name
        | postfix-expression INCR
        | postfix-expression DECR
Expressions Specific to C++
C++ is a complex language, with a rich expression system. The debugger understands much of the system, but it does not understand how to evaluate some complex aspects of a C++ expression. It can correctly debug these when they occur in the source code.

The aspects of the expression system not processed properly during debugger expression evaluation include the following:

  • Many of the implicit conversions
  • Program-defined operators
  • Calling constructors and destructors during the debugger's own evaluation of expressions

There are also some minor restrictions in the following grammar, compared with the full C++ expression grammar, to make it unambiguous:

expression
        : assignment-expression

constant-expression
        : conditional-expression
C++ Identifiers
The debugger correctly augments the general lookup rules when inside class member functions, to look up the members correctly.

The debugger has only a limited understanding of namespaces. It correctly processes names such as UserNameSpace::NestedNamespace::userIdentifier, as well as C++ use-declarations, which introduce a new identifier into a scope.

The debugger does not currently understand C++ using-directives.

The debugger understands the relationship between struct and class identifiers and typedef identifiers.

id-or-keyword-or-typedef-name
        : identifier-or-key-word
        | TYPEDEFname
C++ Constants
The debugger treats numeric constants the same as C++ does. The enumeration constant identifiers go though the same grammar paths as variable identifiers, which has basically the same effect as the C++ semantics.
constant
        : FLOATINGconstant
        | INTEGERconstant
        | CHARACTERconstant
        | WIDECHARACTERconstant
        | WIDESTRINGliteral
C++ Calls
The debugger does not understand the following aspects of C++ calls:
  • Invoking C++ constructors and destructors to create and destroy temporaries containing the value of parameters and results.
  • Default parameters.
  • Many of the implicit conversions that may be needed for the parameters.
  • Overloading resolution. Instead, the debugger queries the user.
call-expression
        : expression
Restrictions and limits are documented here.
C++ Addresses
Following is the C++ implementation of addresses:
address
        : AMPERSAND postfix-expression /* Address of */
        | line-address
        | postfix-expression

address-exp
        : address
        | address-exp PLUS  address
        | address-exp MINUS address
        | address-exp STAR  address
Restrictions and limits are documented here.
C++ Loc
Following is the C++ implementation of loc:
loc
        : expression
        | rescoped-expression
C++ Other Modified Forms of Expressions
whatis-expressions
        : expression
        | printable-type
C++ Rescoped Expressions
The C++ implementation of rescoped expressions is as follows:
qual-typedef-opt
        : type-name
        | qual-typedef-opt TICK type-name
C++ Strings
The C++ implementation of strings is as follows:
string-literal-list
        : string
        | string-literal-list string
C++ Identifier Expressions
The debugger understands nested names. Namespaces go through the same paths as classes, hence the unusual use of TYPEDEFname.
id-expression
	: id-expression-internals

id-expression-internals
	: qualified-id
	| id-or-keyword-or-typedef-name
	| operator-function-name
	| TWIDDLE id-or-keyword-or-typedef-name

qualified-id
	: nested-name-specifier qualified-id-follower

qualified-type
	: nested-name-specifier TYPEDEFname

nested-name-specifier
	: CLCL
	| TYPEDEFname CLCL
	| nested-name-specifier TYPEDEFname CLCL

qualified-id-follower
	: identifier-or-key-word
	| operator-function-name
	| TWIDDLE id-or-keyword-or-typedef-name
C++ Types
The debugger understands the full C++ type specification grammar.
type-specifier
        : basic-type-specifier
        | struct-union-enum-type-specifier
        | typedef-type-specifier

type-qualifier-list
        : type-qualifier
        | type-qualifier-list type-qualifier

type-qualifier
        : CONST
        | VOLATILE

basic-type-specifier
        : basic-type-name basic-type-name
        | basic-type-name type-qualifier
        | type-qualifier-list basic-type-name
        | basic-type-specifier type-qualifier
        | basic-type-specifier basic-type-name

struct-union-enum-type-specifier
        : elaborated-type-name
        | type-qualifier-list elaborated-type-name
        | struct-union-enum-type-specifier type-qualifier

typedef-type-specifier
        : TYPEDEFname type-qualifier
        | type-qualifier-list TYPEDEFname
        | typedef-type-specifier type-qualifier

basic-type-name
        : VOID
        | CHAR
        | SHORT
        | INT
        | LONG
        | FLOAT
        | DOUBLE
        | SIGNED
        | UNSIGNED

elaborated-type-name
        : aggregate-name
        | enum-name

printable-type
        : rescoped-typedef
        | type-name

aggregate-name
        : aggregate-key opt-parenthesized-identifier-or-typedef-name
        | aggregate-key qualified-type

opt-parenthesized-identifier-or-typedef-name
        : id-or-keyword-or-typedef-name
        | LPAREN opt-parenthesized-identifier-or-typedef-name RPAREN

aggregate-key
        : STRUCT
        | UNION
        | CLASS

enum-name
        : ENUM id-or-keyword-or-typedef-name

parameter-type-list
        : PARENS type-qualifier-list-opt

type-name
        : type-specifier
        | qualified-type
        | basic-type-name
        | TYPEDEFname
        | type-qualifier-list
        | type-specifier abstract-declarator
        | basic-type-name abstract-declarator
        | qualified-type abstract-declarator
        | TYPEDEFname abstract-declarator
        | type-qualifier-list abstract-declarator

abstract-declarator
        : unary-abstract-declarator
        | postfix-abstract-declarator
        | postfixing-abstract-declarator

postfixing-abstract-declarator
        : array-abstract-declarator
        | parameter-type-list

array-abstract-declarator
        : BRACKETS
        | LBRACKET constant-expression RBRACKET
        | array-abstract-declarator LBRACKET constant-expression RBRACKET

unary-abstract-declarator
        : STAR
        | AMPERSAND
        | pointer-operator-type
        | STAR                  abstract-declarator
        | AMPERSAND             abstract-declarator
        | pointer-operator-type abstract-declarator

postfix-abstract-declarator
        : LPAREN unary-abstract-declarator RPAREN
        | LPAREN postfix-abstract-declarator RPAREN
        | LPAREN postfixing-abstract-declarator RPAREN
        | LPAREN unary-abstract-declarator RPAREN postfixing-abstract-declarator

pointer-operator-type
        : TYPEDEFname CLCL STAR type-qualifier-list-opt
        | STAR                  type-qualifier-list
        | AMPERSAND             type-qualifier-list
C++ Other Forms of Expressions
The following expressions all have the usual C++ semantics:
primary-expression
        : constant
        | string-literal-list
        | THIS
        | LPAREN expression RPAREN
        | operator-function-name
        | identifier-or-key-word
        | qualified-id
        | process_set
        | LPAREN process_range RPAREN

operator-function-name
        : OPERATOR operator-predefined
        | OPERATOR basic-type-name
        | OPERATOR TYPEDEFname
        | OPERATOR LPAREN type-name RPAREN
        | OPERATOR type-qualifier
        | OPERATOR qualified-type

operator-predefined
        : PLUS
        | MINUS
        | STAR
        | ...
        | DELETE
        | COMMA

type-qualifier-list-opt
        : [ type-qualifier-list ]

postfix-expression
        : primary-expression
        | postfix-expression LBRACKET expression RBRACKET
        | postfix-expression PARENS
        | postfix-expression LPAREN  argument-expression-list RPAREN
        | postfix-expression LPAREN  type-name-list RPAREN
        | postfix-expression DOT  id-expression
        | postfix-expression ARROW id-expression
        | postfix-expression INCR
        | postfix-expression DECR
        | TYPEDEFname LPAREN argument-expression-list RPAREN
        | TYPEDEFname LPAREN type-name-list RPAREN
        | basic-type-name LPAREN assignment-expression RPAREN

type-name-list
        : type-name
        | type-name COMMA type-name-list
        | type-name comma-opt-ellipsis
        | ELLIPSIS

comma-opt-ellipsis
        : ELLIPSIS
        | COMMA ELLIPSIS

unary-expression
        : postfix-expression
        | INCR unary-expression
        | DECR unary-expression
        | line-address
        | AMPERSAND cast-expression
        | STAR cast-expression
        | MINUS cast-expression
        | PLUS cast-expression
        | TWIDDLE LPAREN cast-expression RPAREN
        | NOT cast-expression
        | SIZEOF unary-expression
        | SIZEOF LPAREN type-name RPAREN
        | allocation-expression

allocation-expression
        : operator-new LPAREN type-name RPAREN operator-new-initializer
        | operator-new LPAREN argument-expression-list RPAREN LPAREN type-name RPAREN operator-new-initializer

operator-new
        : NEW
        | CLCL NEW

operator-new-initializer
        : [ PARENS ]
        | [ LPAREN argument-expression-list RPAREN ]

cast-expression
        : unary-expression
        | LPAREN type-name RPAREN cast-expression

deallocation-expression
        : cast-expression
        | DELETE               deallocation-expression
        | CLCL DELETE          deallocation-expression
        | DELETE BRACKETS      deallocation-expression
        | CLCL DELETE BRACKETS deallocation-expression

point-member-expression
        : deallocation-expression
        | point-member-expression DOTstar    deallocation-expression
        | point-member-expression ARROWstar  deallocation-expression

multiplicative-expression
        : point-member-expression
        | multiplicative-expression STAR  point-member-expression
        | multiplicative-expression SLASH point-member-expression
        | multiplicative-expression MOD   point-member-expression

additive-expression
        : multiplicative-expression
        | additive-expression PLUS  multiplicative-expression
        | additive-expression MINUS multiplicative-expression

shift-expression
        : additive-expression
        | shift-expression LS additive-expression
        | shift-expression RS additive-expression

relational-expression
        : shift-expression
        | relational-expression LESS    shift-expression
        | relational-expression GREATER shift-expression
        | relational-expression LE      shift-expression
        | relational-expression GE     shift-expression

equality-expression
        : relational-expression
        | equality-expression EQ relational-expression
        | equality-expression NE relational-expression

AND-expression
        : equality-expression
        | AND-expression AMPERSAND equality-expression

exclusive-OR-expression
        : AND-expression
        | exclusive-OR-expression HAT AND-expression

inclusive-OR-expression
        : exclusive-OR-expression
        | inclusive-OR-expression OR exclusive-OR-expression

logical-AND-expression
        : inclusive-OR-expression
        | logical-AND-expression ANDAND inclusive-OR-expression

logical-OR-expression
        : logical-AND-expression
        | logical-OR-expression OROR logical-AND-expression

conditional-expression
        : logical-OR-expression
        | logical-OR-expression QUESTION expression COLON conditional-expression

assignment-expression
        : conditional-expression
        | unary-expression ASSIGNOP    assignment-expression
        | unary-expression MULTassign  assignment-expression
        | unary-expression DIVassign   assignment-expression
        | unary-expression MODassign   assignment-expression
        | unary-expression PLUSassign  assignment-expression
        | unary-expression MINUSassign assignment-expression
        | unary-expression LSassign    assignment-expression
        | unary-expression RSassign    assignment-expression
        | unary-expression ANDassign   assignment-expression
        | unary-expression ERassign    assignment-expression
        | unary-expression ORassign    assignment-expression
Expressions Specific to Fortran
This section contains expressions specific to Fortran.
Fortran Identifiers
The Fortran implementation of identifiers is as follows:
identifier-or-typedef-name
        : identifier-or-key-word
        | TYPEDEFname
        | PROCEDUREname
Fortran Constants
real-or-imag-part
        : real_constant
        | PLUS real_constant
        | MINUS real_constant
        | integer_constant
        | PLUS integer_constant
        | MINUS integer_constant
constant
        : real_constant
        | integer_constant
        | complex-constant
        | character_constant
        | LOGICALconstantWithKind
character_constant
        : CHARACTERconstantWithKind
        | string
complex-constant
        : LPAREN real-or-imag-part COMMA real-or-imag-part RPAREN
Fortran Rescoped Expressions
The Fortran implementation of rescoped expressions is as follows:
qual-typedef-opt
        : TYPEDEFname /* Base (global) name */
        | qual-typedef-opt TICK TYPEDEFname /* Qualified name */

whatis-expressions
        : expression
        | rescoped-expression
        | printable_type
Fortran Calls
The Fortran implementation of calls is as follows:
call-expression
        : call-stmt

call-stmt
        : named-subroutine
        | named-subroutine LPAREN RPAREN
        | named-subroutine LPAREN actual-arg-spec-list RPAREN
Fortran Addresses
The Fortran implementation of addresses is as follows:
address
        : line-address
        | primary

address-exp
        : address
        | address-exp PLUS  address
        | address-exp MINUS address
        | address-exp STAR  address
Restrictions and limits are documented here.
Fortran Loc
The Fortran implementation of loc is as follows:
loc
        : expression
        | rescoped-expression
Fortran Types
The Fortran implementation of types is as follows:
type-name
        : TYPEDEFname

printable-type
        : rescoped-typedef
        | type-name
Other Forms of Fortran Expressions
expression
        : expr
        | named-procedure

assignment-expression
        : expr

constant-expression
        : constant

unary-expression
        : variable

expr
        : level-5-expr
        | expr defined-binary-op level-5-expr

level-5-expr
        : equiv-operand
        | level-5-expr LOGEQV equiv-operand
        | level-5-expr LOGNEQV equiv-operand
        | level-5-expr LOGXOR equiv-operand

equiv-operand
        : or-operand
        | equiv-operand LOGOR or-operand

or-operand
        : and-operand
        | or-operand LOGAND and-operand

and-operand
        : level-4-expr
        | LOGNOT and-operand

level-4-expr
        : level-3-expr
        | level-3-expr LESS level-3-expr
        | level-3-expr GREATER level-3-expr
        | level-3-expr LE level-3-expr
        | level-3-expr GE level-3-expr
        | level-3-expr EQ level-3-expr
        | level-3-expr NE level-3-expr

level-3-expr
        : level-2-expr
        | level-3-expr SLASHSLASH level-2-expr

level-2-expr
        : add-operand
        | level-2-expr PLUS  add-operand
        | level-2-expr MINUS add-operand

add-operand
        : add-operand-f90
        | add-operand-dec
        | unary-expr-dec

add-operand-f90
        : mult-operand-f90
        | add-operand-f90 STAR  mult-operand-f90
        | add-operand-f90 SLASH mult-operand-f90

mult-operand-f90
        : level-1-expr
        | level-1-expr STARSTAR mult-operand-f90

add-operand-dec
        : mult-operand-dec
        | add-operand-f90 STAR  mult-operand-dec
        | add-operand-f90 SLASH mult-operand-dec
        | add-operand-f90 STAR  unary-expr-dec
        | add-operand-f90 SLASH unary-expr-dec

mult-operand-dec
        : level-1-expr STARSTAR mult-operand-dec
        | level-1-expr STARSTAR unary-expr-dec

unary-expr-dec
        : PLUS  add-operand
        | MINUS add-operand

level-1-expr
        : primary
        | defined-unary-op primary

defined-unary-op
        : DOT_LETTERS_DOT

primary
        : constant
        | variable
        | function-reference
        | LPAREN expr RPAREN
        | AMPERSAND variable
        | process_set
        | LPAREN process_range RPAREN

defined-binary-op
        : DOT_LETTERS_DOT

int-expr
        : expr

scalar-int-expr
        : int-expr

variable
        : named-variable
        | subobject

named-variable
        : variable-name

subobject
        : array-elt-or-sect
        | structure-component
        | known-substring

known-substring
        : disabled-array-elt-or-sect LPAREN substring-range RPAREN
        | hf-array-abomination

substring-range
        : scalar-int-expr COLON scalar-int-expr
        | scalar-int-expr COLON
        |                 COLON scalar-int-expr
        |                 COLON

hf-array-abomination
        : named-variable
	    LPAREN section-subscript-list RPAREN
            LPAREN section-subscript RPAREN
        | structure PERCENT any-identifier
            LPAREN section-subscript-list RPAREN
            LPAREN section-subscript RPAREN
        | structure DOT any-identifier
            LPAREN section-subscript-list RPAREN
            LPAREN section-subscript RPAREN

disabled-array-elt-or-sect
        : DISABLER array-elt-or-sect

array-elt-or-sect
        : named-variable LPAREN section-subscript-list RPAREN
        | structure PERCENT any-identifier LPAREN section-subscript-list RPAREN
        | structure DOT any-identifier LPAREN section-subscript-list RPAREN

section-subscript-list
        : section-subscript
        | section-subscript COMMA section-subscript-list

subscript
        : scalar-int-expr

section-subscript
        : subscript
        | subscript-triplet

subscript-triplet
        : subscript COLON subscript COLON stride
        | subscript COLON           COLON stride
        |           COLON subscript COLON stride
        |           COLON           COLON stride
        | subscript COLON subscript
        | subscript COLON
        |           COLON subscript
        |           COLON

stride
        : scalar-int-expr

structure-component
        : structure PERCENT any-identifier
        | structure DOT any-identifier

structure
        : named-variable
        | structure-component
        | array-elt-or-sect

function-reference
        : SIZEOF LPAREN expr RPAREN
        | named-function LPAREN RPAREN
        | named-function LPAREN actual-arg-spec-list RPAREN

named-procedure
        :  PROCEDUREname

named-function
        :  PROCEDUREname

named-subroutine
        :  PROCEDUREname

actual-arg-spec-list
        : actual-arg-spec
        | actual-arg-spec COMMA actual-arg-spec-list

actual-arg-spec
        : actual-arg

actual-arg
        : expr

any-identifier
        : variable-name
        | PROCEDUREname

variable-name
        : identifier-or-key-word

PROCEDUREname
        : IDENTIFIER

Chapter 16 — Debugging Core Files

When the operating system encounters an unrecoverable error while running a process, for example a segmentation violation (SEGV), the system creates a file named core and places it in the current directory. The core file is not an executable file; it is a snaphot of the state of your process at the time the error occurred. It allows you to analyze the process at the point it crashed.

This chapter discusses the following topics:

It also contains a core file debugging example and a quick reference for transporting a core file.

Invoking the Debugger on a Core File

You can use the debugger to examine the process information in a core file. Use the following debugger command syntax to invoke the debugger on a core file:

% idb executable_file core_file

or

(idb) load executable_file core_file

The executable file is that which was being executed at the time the core file was generated.

Debugging a Core File

When debugging a core file, you can use the debugger to obtain a stack trace and the values of some variables just as you would for a stopped process.

The stack trace lists the functions in your program that were active when the dump occurred. By examining the values of a few variables along with the stack trace, you may be able to pinpoint the process state and the cause of the core dump. Core files cannot be executed; therefore the rerun, step, cont and so on commands will not work until you create a process using the run command.

In addition, if the program is multithreaded, you can examine the thread information with the show thread and thread commands. You can examine the stack trace for a particular thread or for all threads with the where thread command.

The following example uses a null pointer reference in the factorial function. This reference causes the process to abort and dump the core when it is executed. The dump command prints the value of the x variable as a null, and the print *x command reveals that you cannot dereference a null pointer.

% cat testProgram.c

#include <stdio.h>
int factorial(int i)

main() {
        int i,f;
        for (i=1 ; i<3 ; i++) {
                f = factorial(i);
                printf("%d! = %d\en",i,f);
        }
}

int factorial(int i)
int i;
{
int *x;
         x = 0;
         printf("%d",*x);
         if (i<=1)
                 return (1);
         else
                 return (i * factorial(i-1) );
}

% cc -o testProgram -g testProgram.c
% testProgram
Memory fault - core dumped.
% idb testProgram core
Welcome to the debugger Version n
------------------
object file name: testProgram
core file name: core
Reading symbolic information ...done
Core file produced from executable testProgram
Thread terminated at PC 0x120000dc4 by signal SEGV
(idb) where
>0  0x120000dc4 in factorial(i=1) testProgram.c:13
#1  0x120000d44 in main() testProgram.c:4
(idb) dump
>0  0x120000dc4 in factorial(i=1) testProgram.c:13
printf("%d",*x);
(idb) print *x
Cannot dereference 0x0
Error: no value for *x
(idb)

Transporting a Core File

Transporting core files is usually necessary to debug a core file on a system other than that which produced it. It is sometimes possible to debug a core file on a system other than that which produced it if the current system is sufficiently similar to the original system, but it will not work correctly in general.

Procedure for Transporting Core Files

The following procedure (see also quick reference) shows how to transport the core files. In this example, a.out is the name of the executable and core is the name of the core file.

You need to collect a variety of files from the original system. These include the executable, the core file, shared libraries used by the executable, and /usr/shlib/libpthreaddebug.so if the POSIX Threads Library is involved.

Do the following steps (1 through 4) on the original system:

  1. Determine the shared objects in use:

    % idb a.out core
    (idb) listobj
    (idb) quit
    
  2. Cut, paste and edit the result into a list of file names. Most will probably begin with /usr/shlib/.

  3. If /usr/shlib/libpthread.so is one of the files, add /usr/shlib/libpthreaddebug.so to the list. (If you have a privately delivered libpthread.so, there should be a privately delivered corresponding libpthreaddebug.so; use the privately delivered one.)

  4. Package the a.out, core and shared objects, for example, into a tar file. Be sure to use the tar h option to force tar to follow symbolic links as if they were normal files or directories.

    % tar cfvh mybug.tar
    

Then do the following steps (5 through 14) on the current system:

On the current system, the executable and core file are generally put in the current working directory, the shared objects are put in an "application" subdirectory, and libpthreaddebug.so is put in a "debugger" subdirectory.

  1. Create a directory for debugging the transported core files:

    % mkdir mybug
    
  2. Move to that directory:

    % cd mybug
    
  3. Get the package:

    % mv <wherever>/mybug.tar .
    
  4. Create the subdirectories applibs and dbglibs:

    % mkdir applibs dbglibs
    
  5. Unpackage the tar files. Be sure to use the tar s option to strip off any leading slashes from pathnames during extraction.

    % tar xfvs mybug.tar
    
  6. Move the shared objects (that were originally in /usr/shlib and are now in usr/shlib) into applibs:

    % mv usr/shlib/* applibs
    

    If the tar xfvs output in step 9 moved shared objects into other directories, move them into applibs as well.

  7. Make libpthreaddebug.so exist in the dbglibs directory, for example, by linking it to the file in the applibs directory.

    % ln -s ../applibs/libpthreaddebug.so dbglibs/libpthreaddebug.so
    
  8. Set the IDB_COREFILE_LIBRARY_PATH environment variable to the application subdirectory. This directs the debugger to look for shared objects (by their base names) in the application subdirectory before trying the system directories. If the POSIX Threads Library is involved, set the LD_LIBRARY_PATH environment variable to the debugger subdirectory so that the debugger will use the correct libpthreaddebug.so.

    % env IDB_COREFILE_LIBRARY_PATH=applibs \
    LD_LIBRARY_PATH=dbglibs \
    idb a.out core
    
  9. Determine that the shared objects are in the applibs subdirectory rather than in /usr/shlib/:

    (idb) listobj
    

    For an alternative method when the debugger cannot be run on the original system, see the corefile_listobj.c example.

  10. Debug as usual:

    (idb)
    

Core File Debugging Example

The following is a complete example, from core creation, through transporting and core file debugging:

  1. Create the core file:

    % a.out -segv
    Segmentation fault (core dumped)
    
  2. Determine the shared objects using the debugger on the original system:

    % idb a.out core
    Welcome to the debugger Version n
    ------------------
    object file name: a.out
    core file name: core
    Reading symbolic information ...done
    Core file produced from executable a.out
    Thread 0x5 terminated at PC 0x3ff8058b448 by signal SEGV
    (idb) listobj
        section         Start Addr           End Addr
    ------------------------------------------------------------------------------
    a.out
         .text        0x120000000        0x120003fff
         .data        0x140000000        0x140001fff
    
    /usr/shlib/libpthread.so
         .text      0x3ff80550000      0x3ff8058bfff
         .data      0x3ffc0180000      0x3ffc018ffff
         .bss       0x3ffc0190000      0x3ffc01901af
    
    /usr/shlib/libmach.so
         .text      0x3ff80530000      0x3ff8053ffff
         .data      0x3ffc0170000      0x3ffc0173fff
    
    /usr/shlib/libexc.so
         .text      0x3ff807b0000      0x3ff807b5fff
         .data      0x3ffc0210000      0x3ffc0211fff
    
    /usr/shlib/libc.so
         .text      0x3ff80080000      0x3ff8019ffff
         .data      0x3ffc0080000      0x3ffc0093fff
         .bss       0x3ffc0094000      0x3ffc00a040f
    
    (idb) quit
    
  3. Cut, paste, and edit the result into a list of file names. Note that libpthread.so is included, so add /usr/shlib/libpthreaddebug.so to the list.

  4. Create a tar file:

    % tar cfv mybug.tar a.out core \
            /usr/shlib/libpthread.so /usr/shlib/libmach.so \
            /usr/shlib/libexc.so /usr/shlib/libc.so \
            /usr/shlib/libpthreaddebug.so
    a a.out 128 Blocks
    a core 2128 Blocks
    a /usr/shlib/libpthread.so 928 Blocks
    a /usr/shlib/libmach.so 208 Blocks
    a /usr/shlib/libexc.so 96 Blocks
    a /usr/shlib/libc.so symbolic link to ../../shlib/libc.so
    a /usr/shlib/libpthreaddebug.so 592 Blocks
    

    Note that libc.so is a symbolic link. Therefore, use the tar h option to force tar to follow symbolic links as if they were normal files or directories:

    % tar hcfv mybug.tar a.out core \
            /usr/shlib/libpthread.so /usr/shlib/libmach.so \
            /usr/shlib/libexc.so /usr/shlib/libc.so \
            /usr/shlib/libpthreaddebug.so
    a a.out 128 Blocks
    a core 2128 Blocks
    a /usr/shlib/libpthread.so 928 Blocks
    a /usr/shlib/libmach.so 208 Blocks
    a /usr/shlib/libexc.so 96 Blocks
    a /usr/shlib/libc.so 3193 Blocks
    a /usr/shlib/libpthreaddebug.so 592 Blocks
    

    Now you have a package that you can transport.

  5. On the current system, create a directory for debugging, move to that directory, and get the package.

    % mkdir mybug
    % cd mybug
    % mv <wherever>/mybug.tar .
    
  6. Create the necessary subdirectories and unpackage the tar file using the s option:

    % mkdir applibs dbglibs
    % tar xfvs mybug.tar
    blocksize = 256
    x a.out, 65536 bytes, 128 tape blocks
    x core, 1089536 bytes, 2128 tape blocks
    x usr/shlib/libpthread.so, 475136 bytes, 928 tape blocks
    x usr/shlib/libmach.so, 106496 bytes, 208 tape blocks
    x usr/shlib/libexc.so, 49152 bytes, 96 tape blocks
    x usr/shlib/libc.so, 1634400 bytes, 3193 tape blocks
    x usr/shlib/libpthreaddebug.so, 303104 bytes, 592 tape blocks
    
  7. Move the original shared objects into applibs, and make libpthreaddebug.so exist in the dbglibs directory, for example, by linking it to the file in the applibs directory:

    % mv usr/shlib/* applibs
    % ln -s ../applibs/libpthreaddebug.so dbglibs/libpthreaddebug.so
    

    In this example, all shared objects were in usr/shlib/, so no other moving is needed.

  8. Observe the file system:

    % ls -lR
    total 4904
    -rwxr-xr-x   1 user1 groupXX    65536 Sep 17 11:20 a.out*
    drwxrwxr-x   2 user1 groupXX     8192 Sep 17 11:36 applibs/
    -rw-------   1 user1 groupXX  1089536 Sep 17 11:21 core
    drwxrwxr-x   2 user1 groupXX     8192 Sep 17 11:24 dbglibs/
    -rw-rw-r--   1 user1 groupXX  3737600 Sep 17 11:23 mybug.tar
    drwxrwxr-x   3 user1 groupXX     8192 Sep 17 11:36 usr/
    
    ./applibs:
    total 2632
    -rw-r--r--   1 user1 groupXX  1634400 Dec  7  1998 libc.so
    -rw-r--r--   1 user1 groupXX    49152 Jun 26  1998 libexc.so
    -rw-r--r--   1 user1 groupXX   106496 Dec 29  1997 libmach.so
    -rw-r--r--   1 user1 groupXX   475136 Dec  7  1998 libpthread.so
    -rw-r--r--   1 user1 groupXX   303104 Dec  7  1998 libpthreaddebug.so
    
    ./dbglibs:
    total 0
    lrwxrwxrwx   1 user1 groupXX       29 Sep 17 11:24 libpthreaddebug.so@ -> ../applibs/libpthreaddebug.so
    
    ./usr:
    total 8
    drwxrwxr-x   2 user1 groupXX     8192 Sep 17 11:36 shlib/
    
    ./usr/shlib:
    total 0
    %
    

    If other files need to be moved into applibs, do that as well and then re-observe the file system. In this example, there are none.

  9. Now set the environment variables as indicated:

    % env IDB_COREFILE_LIBRARY_PATH=applibs \
    LD_LIBRARY_PATH=dbglibs \
    idb a.out core
    Welcome to the debugger Version n
    ------------------
    object file name: a.out
    core file name: core
    Reading symbolic information ...done
    Core file produced from executable a.out
    Thread 0x5 terminated at PC 0x3ff8058b448 by signal SEGV
    
  10. Issue the listobj command to ensure the application libraries are coming from applibs/. Find any that are not, either from the original system, or unpacked from the tar file but not yet moved into applibs.

    (idb) listobj
        section         Start Addr           End Addr
    ------------------------------------------------------------------------------
    a.out
         .text        0x120000000        0x120003fff
         .data        0x140000000        0x140001fff
    
    applibs/libpthread.so
         .text      0x3ff80550000      0x3ff8058bfff
         .data      0x3ffc0180000      0x3ffc018ffff
         .bss       0x3ffc0190000      0x3ffc01901af
    
    applibs/libmach.so
         .text      0x3ff80530000      0x3ff8053ffff
         .data      0x3ffc0170000      0x3ffc0173fff
    
    applibs/libexc.so
         .text      0x3ff807b0000      0x3ff807b5fff
         .data      0x3ffc0210000      0x3ffc0211fff
    
    applibs/libc.so
         .text      0x3ff80080000      0x3ff8019ffff
         .data      0x3ffc0080000      0x3ffc0093fff
         .bss       0x3ffc0094000      0x3ffc00a040f
    
  11. Now debug as usual:

    (idb) where
    >0  0x3ff8058b448 in nxm_thread_kill(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libpthread.so
    #1  0x3ff80578c58 in pthread_kill(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libpthread.so
    #2  0x3ff8056cd34 in UnknownProcedure3FromFile69(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libpthread.so
    #3  0x3ff807b22d8 in UnknownProcedure4FromFile1(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libexc.so
    #4  0x3ff807b3824 in UnknownProcedure17FromFile1(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libexc.so
    #5  0x3ff807b3864 in exc_unwind(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libexc.so
    #6  0x3ff807b3af0 in exc_raise_signal_exception(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libexc.so
    #7  0x3ff8057a328 in UnknownProcedure6FromFile80(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libpthread.so
    #8  0x3ff800d6a30 in __sigtramp(0x140091c68, 0xb, 0x1, 0x0, 0x0, 0xfffffffffffffcc0) in applibs/libc.so
    #9  0x120001d94 in mandel_val(cr=0.01, ci=0.16, nmin=0, nmax=255) "mb_pi.c":62
    #10 0x12000274c in smp_fill_in_data(raw_mthread=0x11fffe998) "mb_pi.c":338
    #11 0x3ff80582068 in thdBase(0x0, 0x2, 0x0, 0x0, 0xff, 0x1) in applibs/libpthread.so
    (idb) quit
    %
    

Quick Reference for Transporting a Core File

The following sections contain a quick reference for transporting a core file.

First, do the following steps on the original system:

  1. % idb a.out core
  2. (idb) listobj; quit
  3. Cut, paste, and edit into list of filenames.
  4. Add /usr/shlib/libpthreaddebug.so, if libpthread.so
  5. % tar cfvh mybug.tar a.out core <shlibs>

Next, do the following steps on the current system:

  1. % mkdir mybug
  2. % cd mybug
  3. % mv <tarfile> mybug.tar
  4. % mkdir applibs dbglibs
  5. % tar sfvc mybug.tar
  6. % mv usr/shlib/* applibs
  7. % ln -s ../applibs/libptheaddebug.so dbglib/libptheaddebug.so

The ../applibs is not a typo. Think of it as:

% cd dbglibs
% ln -s ../applibs/libpthreaddebug.so libpthreaddebug.so
% cd ..
  1. % env IDB_COREFILE_LIBRARY_PATH=applibs \
    LD_LIBRARY_PATH=dbglibs \
    idb a.out core
  2. (idb) listobj
  3. (idb) <as usual>

Chapter 17 — Kernel Debugging

TBD

Chapter 18 — Machine-Level Debugging

The debugger lets you debug your programs at the machine-code level as well as at the source-code level. Using debugger commands, you can examine and edit values in memory, print the values of all machine registers, and step through program execution one machine instruction at a time.

Only those users familiar with machine-language programming and executable-file-code structure will find low-level debugging useful.

This chapter contains the following sections:

Examining Memory Addresses

You can examine the value contained at an address in memory as follows:

  • The examine commands (/ and ?) display the values stored in memory.
  • The print command, with the appropriate pointer arithmetic, prints the value contained at the address in decimal.
  • The printregs command prints the values of all machine-level registers.

In addition to examining memory, you can also search memory in 32 and 64-bit chunks.

Using the Examine Commands

You can use the examine commands (/ and ?) to print the value contained at the address in one of a number of formats (decimal, octal, hexadecimal, and so on). See Memory Display Commands for more information.

The debugger also maintains the $readtextfile debugger variable that allows you to view the data from the text section of the executable directly from the binary file, rather than reading it from memory.

Using Pointer Arithmetic

You can use C and C++ pointer-type conversions to display the contents of a single address in decimal. Using the print command, the syntax is as follows:

(idb) print *(int *)(address)

Using the same pointer arithmetic, you can use the assign command to alter the contents of a single address. Use the following syntax:

(idb) assign *(int *)(address) = value

The following example shows how to use pointer arithmetic to examine and change the contents of a single address:

(idb) print *(int*)(0x10000000)
 4198916
(idb) assign *(int*)(0x10000000) = 4194744
(idb) print *(int*)(0x10000000)
 4194744
(idb)
Examining Machine-Level Registers

The printregs command prints the values of all machine-level registers. The registers displayed by the debugger are machine dependent. The values are in decimal or hexadecimal, depending on the value of the $hexints variable (the default is 0, decimal). The register aliases are shown; for example, $r1 [$t0]. See the printregs command for more information.

Stepping at the Machine Level

The stepi and nexti commands let you step through program execution incrementally, like the step and next commands. The stepi and nexti commands execute one machine instruction at a time, as opposed to one line of source code. The following example shows stepping at the machine-instruction level:

(idb) stop in main
[#1: stop in main ]
(idb) run
[1] stopped at [main:4 0x120001180]
4     for (i=1 ; i<3 ; i++) {
(idb) stepi
stopped at [main:4 0x120001184] stl     t0, 24(sp)
(idb) [Return]
stopped at [main:5 0x120001188] ldl     a0, 24(sp)
(idb) [Return]
stopped at [main:5 0x12000118c] ldq     t12, -32664(gp)
(idb) [Return]
stopped at [main:5 0x120001190] bsr     ra,
(idb) [Return]
stopped at [factorial:12 0x120001210]   ldah    gp, 8192(t12)
(idb)

At the machine-instruction level, you can step into, rather than over, a function's prologue. While within a function prologue, you may find that the stack trace, variable scope, and parameter list are not correct. Stepping out of the prologue and into the actual function updates the stack trace and variable information kept by the debugger.

Single-stepping through function prologues that initialize large local variables is slow. As a workaround, use the next command.

Chapter 19 — Debugging Parallel Applications

Intel IDB supports debugging of message passing interface (MPI) applications launched by 

  • mpirun, a MPI launcher from mpich, a public domain implementation of MPI.
  • prun, a parallel launcher of Resource Management System (RMS) from Quadrics.

This chapter contains the following sections:

Overview

The biggest challenge of debugging massively parallel applications is coping with large quantities of output from debuggers controlling the parallel application's processes. Intel Debugger helps you do this by condensing (aggregating) similar output into groups. Aggregation is performed by using the following two strategies:

  • Identical output messages are condensed into a single output message. When a condensed message is displayed, it is prefixed with a range of user process IDs (not necessarily consecutive) to which this output applies. All processes with the same output are aggregated into a single and final output message, for example:

    [0-41] Linux Application Debugger for Itanium(R)-based applications, Version XX
     |
    Process range
    
  • Outputs that have different hexadecimal digits, but are otherwise identical, are condensed by aggregating the differing digits into a range, for example:

    [0-41]>2 0x120006d6c in feedback(myid=[0;41],np=42,name=0x11fffe018="mytest") "mytest.c":41
      |                                     |
    Process range	                Value range
    

Another challenge of debugging massively parallel applications is controlling all processes or subsets of the parallel application's processes from the debugger in a consistent manner. The debugger allows you to control all or a subset of your processes through a single user interface. At the startup of a parallel debugging session, Intel IDB does the following:

  1. Detects the topology of your application and attaches a debugger to each of your application's processes.
  2. Builds an n-nary tree with the debuggers as root and leaves with special processes called aggregators in the middle (shown in the following diagram). You can specify the tree's branching factor and the aggregator time delay.

The root debugger is responsible for starting your parallel application and serves as your user interface. The aggregators perform output consolidation as described previously. The leaf debuggers control and query your application processes.

The branching factor is the factor used to build the n-nary tree and determine the number of aggregators in the tree. For example, for 16 processes:

  • Using a branching factor of 8 creates 3 aggregators
  • Using a branching factor of 2 creates 15 aggregators

You can set the value of the $parallel_branchingfactor variable from its default value of 8 to a value equal to or greater than 2 in the Intel IDB initialization file (.idbrc, and so on).

When you delete $parallel_branchingfactor from the Intel IDB initialization file, the branching factor used in the startup mechanism is the default value.

Aggregator delay specifies the time that aggregators wait before they aggregate and send messages down to the next level when not all of the expected messages have been received.

You can change the value of the $parallel_aggregatordelay variable from its default value of 3000 milliseconds in the Intel IDB initialization file (.idbrc, etc.). See Parallel Debugging Tips for more information.

When you delete $parallel_aggregatordelay from the Intel IDB initialization file, the aggregator delay used in the startup mechanism is the default value.

Note: You can only change the values that are set for $parallel_branchingfactor and $parallel_aggregatordelay at startup, in the .idbrc file. After the program has started up, you cannot change these values.

Note: Intel Debugger uses rsh to create the leaf debugger and aggregator processes in the tree structure. Make sure that every node in your cluster has rsh privilege to all other cluster nodes for proper setup of the tree structure. 

Starting a Parallel Debugging Session

To start your parallel application under debugger control, you need to have the environment variable IDB_HOME set to the directory your Intel Debugger is in. When debugging an application launched by mpich, issue the following command at the shell, where N represents number of processes and application is the name of the MPP program you would like to debug:

% mpirun -dbg=idb -np N [other mpich options] application [application arguments] [--idbopt idb options]

Make sure that there is a file called mpirun_dbg.idb in the directory in which mpirun is located. Also note that the Intel IDB option -gdb is not yet supported under parallel debugging session.

When the debugger starts your parallel application, it detects and attaches to all of your application's processes. At this point, your application stops before executing any user code and the debugger displays a prompt.

You can now set any necessary breakpoints and use the continue command to continue the execution of your application.

When debugging an application launched by prun, issue the following command at the shell, where n is the number of processes and N is the number of nodes running the processes.  

% $IDB_HOME/idb [idb options] -parallel `which prun` -n n -N N [other prun options] application [application arguments] 

Using Commands in a Parallel Debugging Session

You can use most Intel Debugger commands just as you would when debugging a non-parallel application. Most commands are passed on to the leaf debuggers and you see aggregated output from them in your user interface. However, there are a few important exceptions.

The following table shows debugger commands that can be accessed remotely, locally, and both remotely and locally for parallel debugging; and Intel Debugger commands that are disabled for parallel debugging.

Remote Local Both Remote and Local Disabled
#
/
?
assign
call
catch
class
cont/conti
delete
delsharedobj
disable
down
dump
enable
examine_address
file
func
goto
history
if
ignore
kill
list
listobj
map/unmap source directory
next/nexti
pop
print
printb
printd
printf
printi
printo
printregs
printt
printx
process
readsharedobj
return
show condition
show mutex
show process
show source directory
show thread
status
step/stepi
stop/stopi
thread
trace/tracei
use/unuse
up
watch
whatis
when/wheni
where/whereis
which
!/history
alias/unalias
edit
export
help
playback
quit
record/unrecord
source
export
set/setenv
sh
unset/unsetenv
attach/detach
kps
load/unload
patch
printenv
rerun
run
snapshot

Remote means commands will be sent to the leaf debuggers. Local means that commands are not sent to the leaf debuggers but are processed by the local Intel IDB.

In addition to the commands listed in the table, you can use four other Intel Debugger commands to assist parallel debugging:

parallel_debugging_command
        : focus_command
        | show_process_set_command
        | show_aggregated_message_command 
        | expand_aggregated_message_command

Working With Sets of Application Processes

When there are many processes, it can be annoying or impractical to enumerate all the processes when one needs to focus on specific processes. Therefore, Intel IDB introduces the concept of "process sets" and "process ranges" to let the user specify a group of processes in a compact form. Moreover, process sets come with the usual set operations, and both the sets and the ranges can be stored in debugger variables for manipulation, reference, or inspection at a later time.

A process set is a bracketed list of process ranges separated by commas.

Note: Because brackets ([]) are part of the process set syntax, this section shows optional syntactic items enclosed in curly braces ({}).

process_set
	: [ ]
	| [ process_range {,...} ]

Note: The set can be empty.

A process range has the following three forms:

process_range
        : *
        | expression
        | { expression } : { expression }

In the first form, the star (*) specifies all processes.

You can use the second form as follows:

  • If expression evaluates to, or can be coerced into an integer p, then the range contains the process with pid p only.
  • If expression evaluates to a process range r, then the process range is the same as r.

You can use the third form to specify a contiguous range of processes. For example, 10:12 stands for the processes associated with pids 10, 11, and 12.

Note: A range whose lower bound is greater than its upper bound is illegal and will be ignored.

Because both the lower bound and the upper bound are optional, you can specify ranges as follows:

Example Represents
:5 All processes whose pid is no greater than 5.
20: All processes whose pid is no less than 20.
: The process set [:] is equivalent to the process set [*].
Using Debugger Variables to Store Process Sets and Ranges

Like storing other data types supported by the debugger, you can store process sets and process ranges in debugger variables using the set command. For example:

	(idb) set $set1 = [:7, 10, 15:20, 30:]
	(idb) print $set1
	[:7, 10, 15:20, 30:]

In addition to using the print command, you can also use the show process set command to inspect the process set stored in a debugger variable. For example:

show_process_set_command
        : show process set debugvar_name
        | show process set all 
        | show process set 

If you do not specify the set name, or if you use the all specifier, the debugger displays all the process sets that are currently stored in debugger variables, as the continued example shows:

(idb) set $set2 = [8:9, 5:2, 22:27]
`5:2' is not a legal process range. Ignored.
(idb) show process set $set2
$set2 = [8:9, 22:27]
(idb) show process set *
$set1 = [:7, 10, 15:20, 30:]
$set2 = [8:9, 22:27]
Process Set Operations

You can use the following three operations on process sets:

Operation Represents Action
+ Set union Takes two sets S1 and S2 and returns a set whose elements are either in S1 or in S2.
- Difference Takes two sets S1 and S2 and returns a set whose elements are in S1 but not in S2.
unary - Negation Takes a single set S and returns the difference of [*] and S.

The following example demonstrates these operations:

(idb) set $set1 = [:10, 15:18, 20:]
(idb) set $set2 = [10:16, 19]
(idb) set $set3 = $set1 + $set2
(idb) print $set3
[*]
(idb) print $set3 - $set2
[:9, 17:18, 20:]
(idb) print -$set2
[:9, 17:18, 20:]
Changing the Current Set with the focus Command

You can use the focus command to change the current process set, which is the set of processes whose debuggers receive the remote command entered at the root debugger:

focus_command
        : focus expression
        | focus all
        | focus

The first form of the command sets the current process set to the set resulting from the evaluation of the given expression. The second form sets the current process set to the set that includes all processes. The third form displays the current process set.

Working with Aggregated Messages

As mentioned in the Overview, the root debugger collects the outputs from the leaf debuggers and presents you with an aggregated output. In most cases, this aggregation works fine, but it can be an impediment if you want to know the exact output from certain leaf debuggers.

To remedy this, the debugger assigns a unique number (called a message_id) to each aggregated message and saves the message in the message_id_list. You can use the following commands to inspect the message list and expand its entries:

show_aggregated_message_command
        : show aggregated message message_id_list
        | show aggregated message all
        | show aggregated message 

message_id_list
        : expression {,...}

The first form of the command displays the aggregated messages in the list whose message IDs match the numbers specified in the message_id_list. The second form displays all the aggregated messages in the list. If no message_id is specified, the debugger shows the most recently added (newest) message.

expand_aggregated_message_command
        : expand aggregated message  message_id_list
        | expand aggregated message 

This command expands the specified messages. If no message_id is specified, the debugger expands the most recently added (newest) message.

You can control the length of the message list using the $aggregatedmsghistory debugger variable. If you set this variable to the default (0), the debugger records as many messages as the system will allow.

Parallel Debugging Tips

This section contains the following tips for debugging parallel applications:

Tip 1. How to Obtain Better Aggregate Outputs

If the debugger outputs are not aggregated as you would expect them to be, you can increase the value of the $parallel_aggregatordelay debugger variable, whose value is the expiration time (in milliseconds) for each of the aggregators when the aggregators have not received all the expected messages. Because the default value of the $parallel_aggregatordelay is 3000 milliseconds, you should not normally have a problem with the aggregation delay.

Tip 2. How to Synchronize Processes

If the processes become unsynchronized in the debugging session (for example, if you use the focus command on a subset of the total set and then use a next or some other motion command), the easiest way to get the processes back together is to use a cont to a future location where all processes have to go. The following example shows how the output from processes is not identical because different processes are at different locations in the program. Using the cont to command synchronizes the processes and aggregates the messages.

(idb) next
(idb)    [4:5,12] stopped at [int feedbackToDebugger(int, int, char*):17 0x120006bf4]
   [0:3,6:11] [3] stopped at [int feedbackToDebugger(int, int, char*):15 0x120006bf0]
   [4:5,12]      17   int pathSize = 1000;
   [0:3,6:11]      15   int i = 0;

(idb) l
(idb)    [0:3,6:11]      16   char path[1000];
   [4:5,12]      18   char hostname[1000];
   [0:3,6:11]      17   int pathSize = 1000;
   [4:5,12]      19   int hostnameSize = 1000;
   [0:3,6:11]      18   char hostname[1000];
   [4:5,12]      20
   [0:3,6:11]      19   int hostnameSize = 1000;
   [4:5,12]      21   volatile int debuggerAttached = 0;
   [0:3,6:11]      20
   [4:5,12]      22
   [0:3,6:11]      21   volatile int debuggerAttached = 0;
   [4:5,12]      23   gethostname(hostname,hostnameSize);
%3 [0:12]      [22;24]
   [0:3,6:11]      23   gethostname(hostname,hostnameSize);
   [4:5,12]      25   getcwd(path,pathSize);
   [0:3,6:11]      24
   [4:5,12]      26   strcat(path,"/");
   [0:3,6:11]      25   getcwd(path,pathSize);
   [4:5,12]      27   strcat(path,name);
   [0:3,6:11]      26   strcat(path,"/");
   [4:5,12]      28
   [0:3,6:11]      27   strcat(path,name);
   [4:5,12]      29   // Print myid pid into idbAttach.myid
   [0:3,6:11]      28
   [4:5,12]      30   sprintf(filename,"idbAttach.%d",myid);
   [0:3,6:11]      29   // Print myid pid into idbAttach.myid
   [4:5,12]      31   file = fopen(filename,"w");
   [0:3,6:11]      30   sprintf(filename,"idbAttach.%d",myid);
   [4:5,12]      32   if (file == NULL) {
   [0:3,6:11]      31   file = fopen(filename,"w");
   [4:5,12]      33     fprintf(stderr,"smg98: can't open %s for %s\n",filename, "w");
   [0:3,6:11]      32   if (file == NULL) {
   [4:5,12]      34     exit(1)
   [0:3,6:11]      33     fprintf(stderr,"smg98: can't open %s for %s\n",filename, "w");
   [4:5,12]      35   }
   [12]      36   fprintf(file," %ld %ld %s %s\n", myid, getpid(), hostname, path);
   [12]      37   fclose(file);
   [12]      38
   [4:5]      36   fprintf(file," %ld %ld %s %s\n", myid, getpid(), hostname, path);
   [0:3,6:11]      34     exit(1);
   [0:3,6:11]      35   }
   [4:5]      37   fclose(file);
   [0:3,6:11]      36   fprintf(file," %ld %ld %s %s\n", myid, getpid(), hostname, path);
   [4:5]      38

(idb) cont to 36
   [0:13] stopped at [int feedbackToDebugger(int, int, char*):36 0x120006cb8]
   [0:13]      36   fprintf(file," %ld %ld %s %s\n", myid, getpid(), hostname, path);

(idb) next
(idb)    [0:13] stopped at [int feedbackToDebugger(int, int, char*):37 0x120006d0c]
   [0:13]      37   fclose(file);
Tip 3. How to find the sources in a parallel debugging session

The debugger will not be able to display the source lines if it cannot find the source file in the directory specified in the application binary file or in the directory in which the binary resides.

Specifying the -I option in the command line fixes the problem. Note that when launching a debugging session through mpirun, this option should be following the flag --idbopt. See Starting a Parallel Debugging Session for details on specifying idb options in a command line.

Alternatively, applying the use command or the map source directory command to all the leaf debuggers can overcome the problem as well. For example,

(idb) w
Source file not found or not readable, tried...
    ./cpi.c
    /usr/users/smith/idb-sandbox/test/src/common/Funct/bin/cpi.c
(Cannot find source file mpirun.c)
(idb) use /usr/proj/debug/idb/test/src/common/Funct/src
   [0:7] Directory search path for source files:
   [0:7]  . /usr/users/smith/idb-sandbox/test/src/common/Funct/bin /usr/proj/debug/idb/test/src/common/Funct/src
(idb) w
   [0:7]      20
   [0:7]      21 double f(double);
   [0:7]      22
   [0:7]      23 int main(int argc, char *argv[])
   [0:7]      24 {
   [0:7]      25     int done = 0, n, myid, numprocs, i;
   [0:7]      26     double PI25DT = 3.141592653589793238462643;
   [0:7]      27     double mypi, pi, h, sum, x;
   [0:7]      28     double startwtime = 0.0, endwtime;
   [0:7]      29     int  namelen;

Parallel Debugging Examples

The first example is a parallel debugging session started by mpirun. Click on the links within the example for explanation.

% mpirun -dbg=idb -np 8 cpi
Linux Application Debugger for Itanium(R)-based applications, Version XX
Reading symbolic information ...done
stopped at [void* MPIR_Breakpoint(void):101 0x40000000000b3060]
    101 {
Process has exited
(idb)
   [0:7] Linux Application Debugger for Itanium(R)-based applications, Version XX
   [0:7] ------------------
   [0:7] object file name: /home/nsl/smith/mpich-1.2.4/examples/cpi
   [0:7] Reading symbolic information ...   [0:7] done
%1 [0:7] Attached to process id [30596;30636]  ....
   [1:7] stopped at [ 0x20000000001ef962]
   [0] stopped at [void* MPIR_Breakpoint(void):101 0x40000000000b3060]
   [0]     101 {

(idb)
   [0:7] stopped at [int main(int, char**):20 0x4000000000003520]
   [0:7]      20     MPI_Init(&argc,&argv);

(idb)
   [0:7]      16     double startwtime = 0.0, endwtime;
   [0:7]      17     int  namelen;
   [0:7]      18     char processor_name[MPI_MAX_PROCESSOR_NAME];
   [0:7]      19
   [0:7] >    20     MPI_Init(&argc,&argv);
   [0:7]      21     MPI_Comm_size(MPI_COMM_WORLD,&numprocs);
   [0:7]      22     MPI_Comm_rank(MPI_COMM_WORLD,&myid);
   [0:7]      23     MPI_Get_processor_name(processor_name,&namelen);
   [0:7]      24

(idb) stop in f
(idb)
   [0:7] [#1: stop in double f(double) ]

(idb) focus [0:3]
[0:3]>
[0:3]>  cont
[0:3]>  Process 3 on nht6005.spt.intel.com
Process 2 on nht6005.spt.intel.com
Process 0 on nht6005.spt.intel.com
Process 1 on nht6005.spt.intel.com

   [0:3] [1] stopped at [double f(double):7 0x4000000000003390]
   [0:3]       7 {

[0:3]>  where
[0:3]>
   [0:3] >0  0x4000000000003390 in f(a=<no value>) "cpi.c":7
%2 [0:3] #1  0x4000000000003a30 in main(argc=0, argv=0x[0;80000fffffffba7c]) "cpi.c":51
   [0:3] #2  0x20000000000906b0 in /lib/libc.so.6.1
   [0:3] #3  0x4000000000003220 in _start(...) in /home/nsl/smith/mpich-1.2.4/examples/cpi

[0:3]>  focus [4:7]
[4:7]>
[4:7]>  cont
[4:7]>  Process 7 on nht6005.spt.intel.com
Process 4 on nht6005.spt.intel.com
Process 6 on nht6005.spt.intel.com
Process 5 on nht6005.spt.intel.com

   [4:7] [1] stopped at [double f(double):7 0x4000000000003390]
   [4:7]       7 {

[4:7]>  where
[4:7]>
   [4:7] >0  0x4000000000003390 in f(a=<no value>) "cpi.c":7
%3 [4:7] #1  0x4000000000003a30 in main(argc=0, argv=0x[0;80000fffffffba7c]) "cpi.c":51
   [4:7] #2  0x20000000000906b0 in /lib/libc.so.6.1
   [4:7] #3  0x4000000000003220 in _start(...) in /home/nsl/smith/mpich-1.2.4/examples/cpi

[4:7]>  focus [*]
[0:7]>
[0:7]>  next
[0:7]>
   [0:7] stopped at [double f(double):8 0x40000000000033b1]
   [0:7]       8     return (4.0 / (1.0 + a*a));

[0:7]>  where
[0:7]>
%4 [0:7] >0  0x40000000000033b1 in f(a=[0.0050000000000000001;0.074999999999999997]) "cpi.c":8
%5 [0:7] #1  0x4000000000003a30 in main(argc=1, argv=0x[80000fffffffb768;6000000000014a50]) "cpi.c":51
   [0:7] #2  0x20000000000906b0 in /lib/libc.so.6.1
   [0:7] #3  0x4000000000003220 in _start(...) in /home/nsl/smith/mpich-1.2.4/examples/cpi

[0:7]>  show aggregated message
%1 [0:7] Attached to process id [30596;30636]  ....
%2 [0:3] #1  0x4000000000003a30 in main(argc=0, argv=0x[0;80000fffffffba7c]) "cpi.c":51
%3 [4:7] #1  0x4000000000003a30 in main(argc=0, argv=0x[0;80000fffffffba7c]) "cpi.c":51
%4 [0:7] >0  0x40000000000033b1 in f(a=[0.0050000000000000001;0.074999999999999997]) "cpi.c":8
%5 [0:7] #1  0x4000000000003a30 in main(argc=1, argv=0x[80000fffffffb768;6000000000014a50]) "cpi.c":51
[0:7]>
[0:7]>  expand aggregated message 1
%1 [0:7] Attached to process id [30596;30636]  ....
 [3] Attached to process id 30612  ....
 [2] Attached to process id 30606  ....
 [0] Attached to process id 30596  ....
 [1] Attached to process id 30600  ....
 [4] Attached to process id 30618  ....
 [5] Attached to process id 30624  ....
 [7] Attached to process id 30636  ....
 [6] Attached to process id 30630  ....
[0:7]>  disable 1
[0:7]>
[0:7]>  cont
[0:7]>  pi is approximately 3.1416009869231249, Error is 0.0000083333333318
wall clock time = 69.300781

   [0:7] Process has exited with status 0

[0:7]>  quit

The following are explanatory notes from the previous example:

Component of Example Meaning
-np 8 This parallel session creates 8 processes.
[0:7] This is a message from processes 0 to 7.
%1 This aggregated message contains messages with differing portions (in this case, the process id's are different from process to process), and 1 is the message id.
focus [0:3] This focus command sets the current process set to include processes 0, 1, 2, and 3.
[0:3]> This prompt shows the current process set.
show aggregated message This show aggregated message command displays all the aggregated messages saved in the message list.
expand aggregated message 1 This expand aggregated message command expands the aggregated message with message id 1.

 

The second example demonstrates how to start a parallel debugging session with prun.

% idb -parallel `which prun` -n 16 -N 8 ./cpi
Linux Application Debugger for Itanium(R)-based applications, Version 7.0,
Build 20021118
Reading symbolic information ...done
stopped at [void _rms_breakpoint(void):2150 0x20000000001913e0]
Source file not found or not readable, tried...
    ./loader.cc
    /usr/bin/loader.cc
(Cannot find source file loader.cc)
stopped at [void _rms_breakpoint(void):2150 0x20000000001913e0]
Source file not found or not readable, tried...
    ./loader.cc
    /usr/bin/loader.cc
(Cannot find source file loader.cc)
Process has exited
(idb)
   [0:15] Linux Application Debugger for Itanium(R)-based applications,
Version 7.0, Build 20021118
   [0:15] ------------------
   [0:15] object file name: cpi
   [0:15] Reading symbolic information ...   [0:15] done
   [0:15]      13     int done = 0, n, myid, numprocs, i;

(idb) where
(idb)
   [0:15] >0  0x4000000000000c60 in _start(...) in cpi

(idb) stop in main
(idb)
   [0:15] [#1: stop in int main(int, char**) ]

(idb) cont
(idb)
   [0:15] [1] stopped at [int main(int, char**):13 0x4000000000000f52]
   [0:15]      13     int done = 0, n, myid, numprocs, i;

In this examples, the first couple of messages about not being able to find certain file can be ignored; they are caused by the fact that this file ususally does not exist on a production system.

Using the mpirun_dbg.idb Startup File

The latest mpich distribution should come with the idb startup file mpirun_dbg.idb. If it does not, or if you are using an older distribution of mpich, you can create the idb startup file by saving the following script as mpirun_dbg.idb in the directory in which mpirun resides:

#! /bin/sh

cmdLineArgs=""
p4pgfile=""
p4workdir=""
prognamemain=""

while [ 1 -le $# ] ; do
  arg=$1
  shift
  case $arg in
     -cmdlineargs)
        cmdLineArgs="$1"
        shift
	;;
     -p4pg)
        p4pgfile="$1"
	shift
	;;
     -p4wd)
        p4workdir="$1"
	shift
	;;
     -progname)
        prognamemain="$1"
	shift
	;;
  esac
done
#
if [ -n "$IDB_HOME" ] ; then
    ldbdir=$IDB_HOME
    idb=$ldbdir/idb
    if [ -f $ldbdir/idb.cat ] && [ -r $ldbdir/idb.cat ] ; then
      if [ -n "$NLSPATH" ];  then
	 nlsmore=$NLSPATH
      else
	 nlsmore=""
      fi
      NLSPATH=$ldbdir/$nlsmore
    fi
else
    idb="idb"
fi
#
#
# Need to `eval echo $cmdLineArgs` to undo evil quoting done in mpirun.args
#
$idb -parallel $prognamemain -p4pg $p4pgfile -p4wd $p4workdir -mpichtv `eval echo $cmdLineArgs`

Appendixes

Appendix 1 — Debugger Variables

The debugger has the following predefined variables. Conventionally, an Intel IDB variable name is an identifier with a leading dollar sign ($).

Variable Default Setting Description
$aggregatedmsghistory
0
Controls the length of the aggregated message list. If set to the default (0), the debugger records as many messages as the system will allow.
$ascii
1
Prints ASCII or all ISO Latin-1.
$beep
1
Beeps on illegal command line editing.
$catchexecs
0
Stops execution on program exec.
$catchforkinfork
0
Notifies you as soon as the forked process is created (otherwise you are notified when the call finishes).
$catchforks
0
Notifies you on program fork and stops child.
$childprocess
0
When the debugger detects a fork, it assigns the child process ID to $childprocess.
$curevent
0
Displays the current breakpoint number.
$curfile
(null)
Displays the current source file.
$curfilepath
(null)
Displays the current source file access path.
$curline
0
Displays the current source line.
$curpc
0
Displays the current point of program execution.
$curprocess
0
Displays the current process ID.
$cursrcline
0
Displays the last source line at end of most recent source listing.
$cursrcpc
0
Displays the PC address at end of most recent machine code listing.
$curthread
0
Displays the current thread ID.
$dbxoutputformat
0
Displays various data structures in dbx format.
$dbxuse
0
Replaces current use paths.
$decints
0
Displays integers in decimal radix.
$doverbosehelp
1
Displays the help menu front page.
$editline
1
Enables command line editing.
$eventecho
1
Echoes events with event numbers.
$exitonterminationofprocesswithpid
None
If set to process ID (pid), when that process terminates, the debugger exits.
$floatshrinking
1
If set to the default (1), the debugger prints binary floating point numbers using the shortest possible decimal number. If set to 0, the debugger prints the decimal number which is the closest representation in the number of decimal digits available of the internal binary number.
$framesearchlimit
0
Defines the maximum number of call frames by which to extend normal language-based identifier lookups.
$funcsig
1
Displays function signature at breakpoint.
$givedebughints
1
Displays hints on debugger features.
$hasmeta
0
Interprets multibyte characters.
$hexints
0
Displays integers in hex radix.
$historylines
20
Defines the number of commands to show for history.
$indent
1
Prints structures with indentation.
$lang
"None"
Defines the programming language of current routine.
$lasteventmade
0
Displays the number of last (successful) breakpoint definition.
$lc_ctype
"C"
Displays the current locale information.
$listwindow
20
Displays the number of lines to show for list.
$main
"main"
Displays the name of the first routine in the program.
$maxstrlen
128
Defines the largest string to print fully.
$memorymatchall
0
When set to non-zero, displays all memory matches in the specified range. Otherwise, only the first memory match is displayed.
$octints
0
Displays integers in octal radix.
$overloadmenu
1
Prompts for choice of overloaded C++ name.
$page
1
Paginates debugger terminal output.
$pagewindow
0
Defines the number of lines per output page. The default of 0 causes the debugger to query the terminal for the page size.
$parallel_branchingfactor
8
Specifies the factor used to build the n-nary tree and determine the number of aggregators in the tree.
$parallel_aggregatordelay
3000 milliseconds
Specifies the length of time that aggregators wait before they aggregate and send messages down to the next level when not all the expected messages have been received.
$parentprocess
0
When the debugger detects a fork, it assigns the parent process ID to $parentprocess.
$pimode
0
Echoes input to log file on playback input.
$prompt
"(idb) "
Specifies debugger prompt.
$readtextfile
0
If set to non-zero, instructions are read from the text area of the binary file rather than from the memory image.
$regstyle
1
Controls the format of register names during disassembly. Valid settings are:
  • 0 = compiler names, for example, t0, ra, or zero.
  • 1 = hardware names, for example, r1, r26, or r31.
  • 2 = assembly names, for example, $1, $26, or $31.
$repeatmode
1
Repeats previous command when you press the Return key.
$reportsotrans
0
Report when an event was changed because a shared object was either opened or closed.
$showlineonstartup
0
Displays the first executable line in main.
$showwelcomemsg
1
Displays welcome message at startup time.
$stackargs
1
Shows arguments in the call stack if 1.
$statusargs
1
Prints breakpoints with parameters if 1.
$stepg0
0
Steps over routines with minimal symbols.
$stoponattach
0
Stops the running process on attach.
$stopparentonfork
0
Stops parent process execution on fork. When set to a nonzero value, this variable instructs the debugger to stop the parent process after it forks a child process. The child process continues to run if $catchforks is not set, otherwise stops. The default is 0.
$symbolsearchlimit
100
Specifies the maximum number of symbols that will be returned by the whereis command for a regular expression search. The default value is 100; a value of 0 indicates no limit.
$threadlevel
decthreads
Specifies POSIX threads (DECthreads) or native threads.
$usedynamictypes
1
Evaluates using C++ static or dynamic type.
$verbose
0
Produces even more output.

Appendix 2 — Debugger Aliases

The debugger has the following predefined aliases:

(idb) alias
F1	print
F2	print 'F2 executes the command "F2 selected-text" - define alias F2'
F3	print 'F3 executes the command "F3 selected-text" - define alias F3'
S	next
Si	nexti
W	list $curline - 10:20
a	assign
att	attach
b	stop at
bp	stop in
c	cont
d	delete
det	detach
e	file
exit	quit
f	func
g	goto
h	history
j	status
l	list
li	($cursrcpc)/10 i; set $cursrcpc = $cursrcpc + 20
n	next
ni	nexti
p	print
pb	printb
pd	printd
pi	printi
plist	show process all
po	printo
pr	printregs
ps	printf "%s",
pt	printt
px	printx
q	quit
r	rerun
ri	record input
ro	record output
s	step
si	stepi
source	playback input
sw	switch
switch	process
t	where
tlist	show thread
ts	where thread all
tset	thread
tstack	where thread all
u	list $curline - 9:10
w	list $curline - 5:10
wi	($curpc - 7)/10 i
wm	watch memory
wv	watch variable

Appendix 3 — corefile_listobj.c Example

You can use the following example as an alternative to the listobj command for cases in which the debugger cannot be run on the original system. See the Transporting Core Files section for more information.

/*
  cc corefile_listobj.c -lxproc -o corefile_listobj
 */


#include <stdio.h>
#include <stdarg.h>
#include <stdlib.h>

#include <errno.h>

typedef unsigned long vma_t;

/* core file format */

#include <sys/user.h>
#include <sys/core.h>

/* dynamic loader hookup */

#include <loader.h>
typedef int (*ldr_reader_func)(vma_t from,
                               void * to,
                               long nbytes,
                               int is_string);
extern pid_t ldr_core_process();
extern int ldr_set_core_reader(ldr_reader_func reader);

/********************************************************************/

static FILE *                   corefile;
static struct core_filehdr      corehdr;
static int                      nsections;
static struct core_scnhdr *     section_headers;

int
open_corefile(const char * corename)
{
  size_t nread;

  corefile = fopen(corename, "rb");
  if (!corefile) {
    perror("Opening corefile");
    return -1;
  }
  nread = fread(&corehdr, sizeof(corehdr), 1, corefile);
  if (nread != 1) {
    perror("fread() of corefile header");
    return -1;
  }
  if (strncmp(corehdr.magic, "Core", 4) != 0) {
    fprintf(stderr, "Corefile header magic is not \"Core\"\n");
    return -1;
  }
  nsections = corehdr.nscns;
  section_headers = calloc(nsections, sizeof(section_headers[0]));
  if (!section_headers) {
    perror("Allocating corefile section headers");
    return -1;
  }
  nread = fread(section_headers, sizeof(section_headers[0]),
                nsections, corefile);
  if (nread != nsections) {
    perror("fread() of corefile section headers");
    return -1;
  }

  return 0;
}

static int
section_type_has_memory(int type)
{
  switch (type) {
  case SCNTEXT: case SCNDATA: case SCNRGN: case SCNSTACK:
    return 1;
  case SCNREGS: case SCNOVFL:
  default:
    return 0;
  }
}

static int
read_from_corefile(vma_t from,
                   void * to,
                   long nbytes,
                   int is_string)
{
  vma_t getter = from;
  char * putter = (char *) to;
  long to_go = nbytes;
  int secnum;
  size_t nxfer;

try_for_more:
  while (to_go > 0) {
    for (secnum = 0; secnum < nsections; secnum += 1) {
      if (section_type_has_memory(section_headers[secnum].scntype)) {
        vma_t vaddr = (vma_t) section_headers[secnum].vaddr;
        vma_t size  = (vma_t) section_headers[secnum].size;
        if (vaddr <= getter && getter < vaddr+size) {
          vma_t this_time = (size < to_go ? size : to_go);
          long file_offset = section_headers[secnum].scnptr+(getter-vaddr);
          if (fseek(corefile, file_offset, SEEK_SET) != 0) {
            perror("fseek() for corefile read");
            return -1;
          }
          nxfer = fread(putter, 1, this_time, corefile);
          if (nxfer != this_time) {
            perror("fread() of corefile data ");
            return -1;
          }
          to_go -= this_time;
          getter += this_time;
          putter += this_time;
          goto try_for_more;
        }
      }
    }
    fprintf("Couldn't find core address for %#lx\n", getter);
    return -1;
  }
  return 0;
}

int
main(int argc, char* argv[])
{
  pid_t process;

  if (argc != 2) {
    fprintf(stderr, "Usage is %s <corefile>\n", argv[0]);
    return 1;
  }
  if (open_corefile(argv[1]) < 0)
    return -1;

  process = ldr_core_process();
  ldr_set_core_reader(read_from_corefile);

  if (ldr_xattach(process) < 0) {
    perror("Attaching to corefile");
    return 1;
  } else {
    ldr_module_t mod_id = LDR_NULL_MODULE;
    ldr_module_info_t info;
    size_t ret_size;
    while (1) {
      if (ldr_next_module(process, &mod_id) < 0) {
        perror("ldr_next_module");
        return 1;
      }
      if (mod_id == LDR_NULL_MODULE)
        break;
      if (ldr_inq_module(process, mod_id, &info,
                         sizeof(info), &ret_size) < 0) {
        perror("ldr_inq_module");
        return 1;
      }
      printf("%s\n", info.lmi_name);
    }
    ldr_xdetach(process);
    return 0;
  }
}

Appendix 4 — Array Navigation Example

The debugger provides parameterized aliases and debugger variables of arbitrary types. Clever use of these can do almost any list traversal.

For example, here is how to navigate an array:

alias elt(e_) "{ p e_ }"
alias pa0(a)  "{ set $a = &a[0]; set $i = 0; elt($a[$i]); set $i = $i+1 }"
alias pan     "{ elt($a[$i]); set $i = $i+1 }"
pa0
pan
pan
pan

%idb a.out
...
(idb) alias elt(e_) "{ p e_ }"
(idb) alias a0(a)   "{ set $a = &a[0]; set $i = 0; elt($a[$i]); set $i = $i+1 }"
(idb) alias pan     "{ elt($a[$i]); set $i = $i+1 }"
...
(idb) pa0(a)
struct S {
  next = 0x140000178;
}
(idb) pan
struct S {
  next = 0x140000180;
}
(idb)
struct S {
  next = 0x140000188;
}
(idb)
struct S {
  next = 0x140000190;
}