Auto-deleting watchpoints

Auto-deleting watchpoints

[Mihai Basa]

Greetings all!

I'd like to question a certain behaviour of gdb, and I quote from the 
man(ual):

      "GDB automatically deletes watchpoints that watch local (automatic)
    variables, or expressions that involve such variables, when they go out
    of scope, that is, when the execution leaves the block in which these
    variables were defined."
    
This includes deleting local watchpoints even when the program makes a call to 
a library function (say, sqrt()). I believe this auto-deletion _severly_ 
reduces the practicality of watchpoints, because they simply go away on the 
first call they hit!
   
I am not totally aware about the way hardware watchpoints behave when the 
processor switches Code-Segments (which is what happens?), but couldn't gdb at 
least re-enable the watchpoint when it re-enters scope? Is there a bad 
side-effect to this that I can't see?

    
Highest regards!
Mihai Basa

Re: Auto-deleting watchpoints

[Daniel Jacobowitz]

On Tue, Dec 02, 2003 at 08:40:23PM +0000, Mihai Basa wrote:
> Greetings all!
> 
> I'd like to question a certain behaviour of gdb, and I quote from the 
> man(ual):
> 
>       "GDB automatically deletes watchpoints that watch local (automatic)
>     variables, or expressions that involve such variables, when they go out
>     of scope, that is, when the execution leaves the block in which these
>     variables were defined."
>     
> This includes deleting local watchpoints even when the program makes a call to 
> a library function (say, sqrt()). I believe this auto-deletion _severly_ 
> reduces the practicality of watchpoints, because they simply go away on the 
> first call they hit!

That is not what is supposed to happen.  The watchpoint should stay
until the function containing the local variable has exited.  If you
have a testcase where that doesn't work, file a bug report (and try the
latest version of GDB), please.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: Auto-deleting watchpoints

[Eli Zaretskii]

> Date: Tue, 2 Dec 2003 15:44:51 -0500
> From: Daniel Jacobowitz <drow@mvista.com>
> >     
> > This includes deleting local watchpoints even when the program makes a call to 
> > a library function (say, sqrt()). I believe this auto-deletion _severly_ 
> > reduces the practicality of watchpoints, because they simply go away on the 
> > first call they hit!
> 
> That is not what is supposed to happen.  The watchpoint should stay
> until the function containing the local variable has exited.

Right.  And I have simple test cases to prove it.

Mihai, you probably discovered a bug.  Please send a test case.

RE: Auto-deleting watchpoints

[Mihai Basa]

>===== Original Message From Eli Zaretskii <eliz@elta.co.il> =====
>> Date: Tue, 2 Dec 2003 15:44:51 -0500
>> From: Daniel Jacobowitz <drow@mvista.com>
>> >
>> > This includes deleting local watchpoints even when the program makes a 
call to
>> > a library function (say, sqrt()). I believe this auto-deletion _severly_
>> > reduces the practicality of watchpoints, because they simply go away on 
the
>> > first call they hit!
>>
>> That is not what is supposed to happen.  The watchpoint should stay
>> until the function containing the local variable has exited.
>
>Right.  And I have simple test cases to prove it.
>
>Mihai, you probably discovered a bug.  Please send a test case.


I've tried this with gdb-6.0, and the fenomenon seems to have disappeared, but 
I've only tested it lightly. I didn't really manage to isolate the bug, as it 
only appeared in certain configurations of the source code, and showed up in 
different places (so the test case is not simple at all). It appears that 
gdb-6.0 has somehow solved this.

To summarize what was happening in gdb-5.3:
 A watchpoint is set on an element of a malloc'ed array. The pointer to this 
array is sent to a subroutine, where, after a while, the previously-set watch 
gets deleted on a call to printf. The message produced is:

        Watchpoint 2 deleted because the program has left the block in
        which its expression is valid.
        0x4008e00e in vfprintf () from /lib/i686/libc.so.6

and the source-line on which the watchpoint to "active_id[123]" got deleted 
is:

        printf("from i = %i at %lf  at vel= %lf ", i,P[i].x[0] ,P[i].u[0]);

(it has nothing to do with active_id[123]). The backtrace was:
        (gdb) backtrace
        #0  0x4008e00e in vfprintf () from /lib/i686/libc.so.6
        #1  0x00000000 in ?? ()


Now gdb-6.0 displays the 'watchpoint deleted' message only when it hits the 
end of the program (in a library function called "_fini()").


Regards,
Mihai Basa