Re: [PATCH] Disable thread specific breakpoints when thread dies

[Daniel Jacobowitz <drow@false.org>]

On Fri, Jan 13, 2006 at 09:11:35PM +0100, Mark Kettenis wrote:
> > Date: Fri, 13 Jan 2006 17:33:26 +0000
> > From: Andrew STUBBS <andrew.stubbs@st.com>
> > 
> > Daniel Jacobowitz wrote:
> > >>> You shouldn't need to use the target method here.  Does valid_thread_id
> > >>> work?
> > >>>
> > >>> Also, please remember the space before opening parentheses.
> > >> The thread still seems to have a valid ID after it has died. You can 
> > >> even do 'b 8 t 4' after the program has exited. It does give an error 
> > >> for threads which never existed though.
> > > 
> > > Why does that happen?  It is presumably a bug.
> > > 
> > 
> > I have looked into this. The problem is that the threads are only 
> > deleted from the table when 'info threads' is used. The target method 
> > works because that queries the target, not GDB's internal state, and 
> > always gets the right answer (at least in our target interface).

At a guess, we should be calling whatever the appropriate function is
to empty the threads list in the generic bits of GDB, around where we
call the target mourn_inferior hook.  Does that make sense?

> > I am happy, therefore, that the attached patch, with valid_thread_id(), 
> > is correct, and will work once this other problem has been solved (or if 
> > the user types 'info threads').
> > 
> > OK to commit?
> 
> Sorry, but I don't think we should commit a patch that's just papering
> over some other more serious problem, perhaps perhaps if there's some
> pressing need to do so.

Sorry about the confusion here - Mark, the patch Andrew posted is not a
workaround, but exactly the opposite.  It's a correct patch that ought
to work, but doesn't because of a different bug, that he has
volunteered to track down next week.

With that clarified, do you have any objection to the patch?

-- 
Daniel Jacobowitz
CodeSourcery