Document revision date: 30 March 2001

You can set a breakpoint on a task body by using one of the following syntaxes to refer to the task body (see Section 17.3.2):

task-type-identifier$TASK_BODY

task-identifier$TASK_BODY

For example, the following command sets a breakpoint on the body of task CHILD. This breakpoint is triggered just before the elaboration of the task's declarative part (also called the task's activation).

DBG> SET BREAK CHILD$TASK_BODY

CHILD$TASK_BODY is a name for the address of the first instruction the task will execute. It is meaningful to set a breakpoint on an instruction, and hence on this name. However, you must not name the task object (for example, CHILD) in a SET BREAK command. The task-object name designates the address of a data item (the task value). Just as it is erroneous to set a breakpoint on an integer object, it is erroneous to set a breakpoint on a task object.

You can monitor the execution of communicating tasks by setting breakpoints or tracepoints on entry calls and accept statements.

There are several points in and around an accept statement where you might want to set a breakpoint or tracepoint. For example, consider the following program segment, which has two accept statements for the same entry, RENDEZVOUS:

8 task body TWO_ACCEPTS is 9 begin 10 for I in 1..2 loop 11 select 12 accept RENDEZVOUS do 13 PUT_LINE("This is the first accept statement"); 14 end RENDEZVOUS; 15 or 16 terminate; 17 end select; 18 end loop; 19 accept RENDEZVOUS do 20 PUT_LINE("This is the second accept statement"); 21 end RENDEZVOUS; 22 end TWO_ACCEPTS;

You can set a breakpoint or tracepoint in the following places in this example:

• At the start of an accept statement (line 12 or 19). By setting a breakpoint or tracepoint here, you can monitor when execution reaches the start of the accept statement, where the accepting task might become suspended before a rendezvous actually occurs.
• At the start of the body (sequence of statements) of an accept statement (line 13 or 20). By setting a breakpoint or tracepoint here, you can monitor when a rendezvous has started---that is, when the accept statement actually begins execution.
• At the end of an accept statement (line 14 or 21). By setting a breakpoint or tracepoint here, you can monitor when the rendezvous has completed, and execution is about to switch back to the caller task.

To set a breakpoint or tracepoint in and around an accept statement, you can specify the associated line number. For example, the following command sets a breakpoint on the start and also on the body of the first accept statement in the preceding example:

DBG> SET BREAK %LINE 12, %LINE 13

To set a breakpoint or a tracepoint on an accept statement body, you can also use the entry name (specifying its expanded name to identify the task body where the entry is declared). For example:

DBG> SET BREAK TWO_ACCEPTS$TASK_BODY.RENDEZVOUS

If there is more than one accept statement for an entry, the debugger treats the entry as an overloaded name. The debugger issues a message indicating that the symbol is overloaded, and you must use the SHOW SYMBOL command to identify the overloaded names that have been assigned by the debugger. For example:

DBG> SHOW SYMBOL RENDEZVOUS overloaded symbol TEST.TWO_ACCEPTS$TASK_BODY.RENDEZVOUS overloaded instance TEST.TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__1 overloaded instance TEST.TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__2

Overloaded names have an integer suffix preceded by two underscores. For more information on overloaded names, see the debugger's online help (type Help Language_Support Ada).

To determine which of these overloaded names is associated with a particular accept statement, use the EXAMINE/SOURCE command. For example:

DBG> EXAMINE/SOURCE TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__1 module TEST_ACCEPTS 12: accept RENDEZVOUS do DBG> EXAMINE/SOURCE TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__2 module TEST_ACCEPTS 19: accept RENDEZVOUS do

In the following example, when the breakpoint triggers, the caller task is evaluated (see Section 17.3.4 for information about the symbol %CALLER_TASK):

DBG> SET BREAK TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__2 - _DBG> DO (EVALUATE %CALLER_TASK)

The following breakpoint triggers only when the caller task is %TASK 2:

DBG> SET BREAK TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__2 - _DBG> WHEN (%CALLER_TASK = %TASK 2)

If the calling task has more than one entry call to the same accept statement, you can use the SHOW TASK/CALLS command to identify the source line where the entry call was issued. For example:

DBG> SET BREAK TWO_ACCEPTS$TASK_BODY.RENDEZVOUS__2 - _DBG> DO (SHOW TASK/CALLS %CALLER_TASK)

17.6.4 Monitoring Task Events

The SET BREAK/EVENT and SET TRACE/EVENT commands enable you to set breakpoints and tracepoints that are triggered by task and exception events. For example, the following command sets tracepoints that trigger whenever task CHILD or %TASK 2 makes a transition to the RUN state:

DBG> SET TRACE/EVENT=RUN CHILD,%TASK 2

When a breakpoint or tracepoint is triggered as a result of an event, the debugger identifies the event and gives additional information.

The following tables list the event-name keywords that you can specify with the SET BREAK/EVENT and SET TRACE/EVENT commands:

• Table 17-6 lists the generic language-independent events common to all tasks.
• Table 17-7 lists the events specific to POSIX Threads tasks.
• Table 17-8 lists the events specific to Ada tasks.

Table 17-6 Generic Low-Level Task Scheduling Events

Event Name	Description
RUN	Triggers when a task is about to run.
PREEMPTED	Triggers when a task is being preempted from the RUN state and its state changes to READY. (See Table 17-3.)
ACTIVATING	Triggers when a task is about to begin its execution.
SUSPENDED	Triggers when a task is about to be suspended.

Table 17-7 POSIX Threads-Specific Events

Event Name	Description
HANDLED	Triggers when an exception is about to be handled in some TRY block.
TERMINATED	Triggers when a task is terminating (including by alert or exception).
EXCEPTION_TERMINATED	Triggers when a task is terminating because of an exception.
FORCED_TERM	Triggers when a task is terminating due to an alert operation.

Table 17-8 Ada-Specific Events

Event Name	Description
HANDLED	Triggers when an exception is about to be handled in some Ada exception handler, including an others handler.
HANDLED_OTHERS	Triggers only when an exception is about to be handled in an others Ada exception handler.
RENDEZVOUS_EXCEPTION	Triggers when an exception begins to propagate out of a rendezvous.
DEPENDENTS_EXCEPTION	Triggers when an exception causes a task to wait for dependent tasks in some scope (includes unhandled exceptions, 1 which, in turn, include special exceptions internal to the Compaq Ada Run-Time Library; for more information, see the Compaq Ada documentation). Often immediately precedes a deadlock.
TERMINATED	Triggers when a task is terminating, whether normally, by an abort statement, or by an exception.
EXCEPTION_TERMINATED	Triggers when a task is terminating due to an unhandled exception. 1
ABORT_TERMINATED	Triggers when a task is terminating due to an abort statement.

In the previous tables, the exception-related events are included for completeness. Only the task events are discussed in the following paragraphs. For more information about the exception events, see the debugger's online help (type Help Language_Support Ada).

You can abbreviate an event-name keyword to the minimum number of characters that make it unique.

The event-name keywords that you can specify with the SET BREAK/EVENT or SET TRACE/EVENT command depend on the current event facility, which is either THREADS or ADA in the case of task events. The appropriate event facility is set automatically when the program is brought under debugger control. The SHOW EVENT_FACILITY command identifies the facility that is currently set and lists the valid event name keywords for that facility (including those for the generic events).

Several examples follow showing the use of the /EVENT qualifier:

DBG> SET BREAK/EVENT=PREEMPTED DBG> GO break on THREADS event PREEMPTED Task %TASK 4 is getting preempted by %TASK 3 . . . DBG> SET BREAK/EVENT=SUSPENDED DBG> GO break on THREADS event SUSPENDED Task %TASK 1 is about to be suspended . . . DBG> SET BREAK/EVENT=TERMINATED DBG> GO break on THREADS event TERMINATED Task %TASK 4 is terminating normally DBG>

The following predefined event breakpoints are set automatically when the program is brought under debugger control:

• EXCEPTION_TERMINATED event breakpoints are predefined for programs that call POSIX Threads routines.
• EXCEPTION_TERMINATED and DEPENDENTS_EXCEPTION event breakpoints are predefined for Ada programs or programs that call Ada routines.

Ada examples of the predefined and other types of event breakpoints follow.

Example of EXCEPTION_TERMINATED Event

When the EXCEPTION_TERMINATED event is triggered, it usually indicates an unanticipated program error. For example:

... break on ADA event EXCEPTION_TERMINATED Task %TASK 2 is terminating because of an exception %ADA-F-EXCCOP, Exception was copied at a "raise;" or "accept" -ADA-F-EXCEPTION, Exception SOME_ERROR -ADA-F-EXCRAIPRI, Exception raised prior to PC = 00000B61 DBG>

Example of DEPENDENTS_EXCEPTION Event (Ada)

For Ada programs, the DEPENDENTS_EXCEPTION event often unexpectedly precedes a deadlock. For example:

... break on ADA event DEPENDENTS_EXCEPTION Task %TASK 2 may await dependent tasks because of this exception: %ADA-F-EXCCOP, Exception was copied at a "raise;" or "accept" -ADA-F-EXCEPTION, Exception SOME_ERROR -ADA-F-EXCRAIPRI, Exception raised prior to PC = 00000B61 DBG>

Example of RENDEZVOUS_EXCEPTION Event (Ada)

For Ada programs, the RENDEZVOUS_EXCEPTION event enables you to see an exception before it leaves a rendezvous (before exception information has been lost due to copying the exception into the calling task). For example:

... break on ADA event RENDEZVOUS_EXCEPTION Exception is propagating out of a rendezvous in task %TASK 2 %ADA-F-CONSTRAINT_ERRO, CONSTRAINT_ERROR -ADA-I-EXCRAIPRI, Exception raised prior to PC = 00000BA6 DBG>

To cancel breakpoints (or tracepoints) set with the /EVENT qualifier, use the CANCEL BREAK/EVENT (or CANCEL TRACE/EVENT) command. Specify the event qualifier and optional task expression in the CANCEL command exactly as you did with the SET command, excluding any WHEN or DO clauses.

You might want to set event breakpoints and tracepoints in a debugger initialization file for your tasking programs. For example:

SET BREAK/EVENT=ACTIVATING SET BREAK/EVENT=HANDLED DO (SHOW CALLS) SET BREAK/EVENT=ABORT_TERMINATED DO (SHOW CALLS) SET BREAK/EVENT=EXCEPTION_TERM DO (SHOW CALLS) SET BREAK/EVENT=TERMINATED

17.7 Additional Task-Debugging Topics

The following sections discuss additional topics related to task debugging:

• Deadlock
• Automatic stack checking
• Using Ctrl/Y

A deadlock is an error condition in which each task in a group of tasks is suspended and no task in the group can resume execution until some other task in the group executes. Deadlock is a typical error in tasking programs (in much the same way that infinite loops are typical errors in programs that use WHILE statements).

A deadlock is easy to detect: it causes your program to appear to suspend, or hang, in midexecution. When deadlock occurs in a program that is running under debugger control, press Ctrl/C to interrupt the deadlock and display the debugger prompt.

In general, the SHOW TASK/ALL command (see Section 17.4) or the SHOW TASK/STATE=SUSPENDED command is useful because it shows which tasks are suspended in your program and why. The command SET TASK/VISIBLE %NEXT_TASK is particularly useful when debugging in screen mode. It enables you to cycle through all tasks and display the code that each task is executing, including the code in which execution is stopped.

The SHOW TASK/FULL command gives detailed task state information, including information about rendezvous, entry calls, and entry index values. The SET BREAK/EVENT or SET TRACE/EVENT command (see Section 17.6.4) enables you to set breakpoints or tracepoints at or near locations that might lead to deadlock. The SET TASK/PRIORITY and SET TASK/RESTORE commands enable you to see if a low-priority task that never runs is causing the deadlock.

Table 17-9 lists a number of tasking deadlock conditions and suggests debugger commands that are useful in diagnosing the cause.

Table 17-9 Ada Tasking Deadlock Conditions and Debugger Commands for Diagnosing Them

Deadlock Condition	Debugger Commands
Self-calling deadlock (a task calls one of its own entries)	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL
Circular-calling deadlock (a task calls another task, which calls the first task)	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL
Dynamic-calling deadlock (a circular series of entry calls exists, and at least one of the calls is a timed or conditional entry call in a loop)	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL
Exception-induced deadlock (an exception prevents a task from answering one of its entry calls, or the propagation of an exception must wait for dependent tasks)	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL SET BREAK/EVENT=DEPENDENTS_EXCEPTION (for Ada programs)
Deadlock because of incorrect run-time calculations for entry indexes, when conditions, and delay statements within select statements	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL EXAMINE
Deadlock due to entries being called in the wrong order	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL
Deadlock due to busy-waiting on a variable used as a flag that is to be set by a lower priority task, and the lower priority task never runs because a higher priority task is always ready to execute	SHOW TASK/ALL SHOW TASK/STATE=SUSPENDED SHOW TASK/FULL SET TASK/PRIORITY SET TASK/RESTORE

17.7.2 Automatic Stack Checking in the Debugger

In tasking programs, an undetected stack overflow can occur in certain circumstances and can lead to unpredictable execution. (For more information on task stack overflow, see the Compaq Ada or POSIX Threads documentation.) The debugger automatically does the following stack checks to help you detect the source of stack overflow problems (if the stack pointer is out of bounds, the debugger displays an error message):

• A stack check is done for the active task after a STEP command executes or a breakpoint triggers (see Section 17.6.1). (This check is not done if you have used the /SILENT qualifier with the STEP or SET BREAKPOINT command.)
• A stack check is done for each task whose state is displayed in a SHOW TASK command. Thus, a SHOW TASK/ALL command automatically causes the stacks of all tasks to be checked.

The following examples show the kinds of error messages displayed by the debugger when a stack check fails. A warning is issued when most of the stack has been used up, even if the stack has not yet overflowed.

warning: %TASK 2 has used up over 90% of its stack SP: 0011194C Stack top at: 00111200 Remaining bytes: 1868 error: %TASK 2 has overflowed its stack SP: 0010E93C Stack top at: 00111200 Remaining bytes: -10436 error: %TASK 2 has underflowed its stack SP: 7FF363A4 Stack base at: 001189FC Stack top at: 00111200

One of the unpredictable events that can happen after a stack overflows is that the stack can then underflow. For example, if a task stack overflows and the stack pointer remains in the top guard area, the operating system will attempt to signal an ACCVIO condition. However, because the top guard area is not a writable area of the stack, the operating system cannot write the signal arguments for the ACCVIO. When this happens, the operating system cuts back the stack: it causes the frame pointer and stack pointer to point to the base of the main program stack area, writes the signal arguments, and then modifies the program counter to force an image exit.

If a time-slice AST or other AST occurs at this instant, execution can resume in a different task, and for a while, the program may continue to execute, although not normally (the task whose stack overflowed may use---and overwrite---the main program stack). The debugger stack checks help you to detect this situation. If you step into a task whose stack has been cut back by the operating system, or if you use the SHOW TASK/ALL command at that time, the debugger issues its stack underflow message.

17.7.3 Using Ctrl/Y When Debugging Ada Tasks

Pressing Ctrl/C is the recommended method of interrupting program execution or a debugger command during a debugging session. This returns control to the debugger; pressing Ctrl/Y returns control to DCL level.

If you interrupt a task debugging session by pressing Ctrl/Y, you might have some problems when you then start the debugger at DCL level with the DEBUG command. In such cases, you should insert the following two lines in the source code at the beginning of your main program to name the Compaq Ada predefined package CONTROL_C_INTERCEPTION:

with CONTROL_C_INTERCEPTION; pragma ELABORATE(CONTROL_C_INTERCEPTION);

For information on this package, see the Compaq Ada documentation.

	privacy and legal statement
4538PRO_037.HTML