PR gdb/11568 - delete thread-specific breakpoints on thread exit
authorPedro Alves <palves@redhat.com>
Tue, 17 Sep 2013 19:32:47 +0000 (19:32 +0000)
committerPedro Alves <palves@redhat.com>
Tue, 17 Sep 2013 19:32:47 +0000 (19:32 +0000)
commit49fa26b0411d990d36f3f6c095d167f3d12afdf4
treed5f61a4078a129b9c605ac0b5d96a18f2c9bcdfc
parente83b28bc76d98e8b1f853ad5d72028c682695a37
PR gdb/11568 - delete thread-specific breakpoints on thread exit

PR gdb/11568 is about thread-specific breakpoints being left behind
when the corresponding thread exits.

Currently:

 (gdb) b start thread 2
 Breakpoint 3 at 0x400614: file thread-specific-bp.c, line 23.
 (gdb) b end
 Breakpoint 4 at 0x40061f: file thread-specific-bp.c, line 29.
 (gdb) c
 Continuing.
 [Thread 0x7ffff7fcb700 (LWP 14925) exited]
 [Switching to Thread 0x7ffff7fcc740 (LWP 14921)]

 Breakpoint 4, end () at thread-specific-bp.c:29
 29      }
 (gdb) info threads
   Id   Target Id         Frame
 * 1    Thread 0x7ffff7fcc740 (LWP 14921) "thread-specific" end () at thread-specific-bp.c:29
 (gdb) info breakpoints
 Num     Type           Disp Enb Address            What
 2       breakpoint     keep y   0x0000000000400614 in start at thread-specific-bp.c:23
         breakpoint already hit 1 time
 3       breakpoint     keep y   0x0000000000400614 in start at thread-specific-bp.c:23 thread 2
         stop only in thread 2
 4       breakpoint     keep y   0x000000000040061f in end at thread-specific-bp.c:29
         breakpoint already hit 1 time

Note that the thread-specific breakpoint 3 stayed around, even though
thread 2 is gone.

There's no way that breakpoint can trigger again (*), so the PR argues
that the breakpoint should just be removed, like local watchpoints.
I'm ambivalent on this -- it could be reasonable to disable the
breakpoint (kind of like breakpoint in shared library code when the
DSO is unloaded), so the user could still use it as visual template
for creating other breakpoints (copy/paste command lists, etc.), or we
could have a way to change to which thread a breakpoint applies.  But,
several people pushed this direction, and I don't plan on arguing...

(*) - actually, there is ...  thread numbers are reset on "run", so
the user could do "break foo thread 2", "run", and expect the
breakpoint to hit again on the second thread.  But given gdb's thread
numbering can't really be stable, that'd only work sufficiently well
for thread 1, so we'd better call it unsupported.

So with the patch, whenever a thread is deleted from GDB's list, GDB
goes through the thread-specific breakpoints and deletes corresponding
breakpoints.  Since this is user-visible, GDB prints out:

  Thread-specific breakpoint 3 deleted - thread 2 is gone.

And of course, we end up with:

 (gdb) info breakpoints
 Num     Type           Disp Enb Address            What
 2       breakpoint     keep y   0x0000000000400614 in start at thread-specific-bp.c:23
         breakpoint already hit 1 time
 4       breakpoint     keep y   0x000000000040061f in end at thread-specific-bp.c:29
         breakpoint already hit 1 time

2013-09-17  Muhammad Waqas <mwaqas@codesourcery.com>
    Pedro Alves  <palves@redhat.com>

PR gdb/11568
* breakpoint.c (remove_threaded_breakpoints): New function.
(_initialize_breakpoint): Attach remove_threaded_breakpoints
as thread_exit observer.

2013-09-17  Muhammad Waqas  <mwaqas@codesourccery.com>
    Jan Kratochvil  <jan.kartochvil@redhat.com>
    Pedro Alves  <palves@redhat.com>

PR gdb/11568
* gdb.thread/thread-specific-bp.c: New file.
* gdb.thread/thread-specific-bp.exp: New file.
gdb/ChangeLog
gdb/breakpoint.c
gdb/testsuite/ChangeLog
gdb/testsuite/gdb.threads/thread-specific-bp.c [new file with mode: 0644]
gdb/testsuite/gdb.threads/thread-specific-bp.exp [new file with mode: 0644]