This is the mail archive of the
gdb-patches@sources.redhat.com
mailing list for the GDB project.
Re: [RFA] use frame IDs to detect function calls while stepping
- From: Joel Brobecker <brobecker at gnat dot com>
- To: Andrew Cagney <cagney at gnu dot org>
- Cc: Elena Zannoni <ezannoni at redhat dot com>,gdb-patches at sources dot redhat dot com
- Date: Mon, 1 Mar 2004 11:48:01 -0800
- Subject: Re: [RFA] use frame IDs to detect function calls while stepping
- References: <20040205044119.GC18961@gnat.com> <20040205171324.GF18961@gnat.com> <16418.37058.65446.669052@localhost.redhat.com> <20040207040049.GH18961@gnat.com> <403F60F1.7020902@gnu.org>
> >>> > + if (IN_SOLIB_CALL_TRAMPOLINE (stop_pc, ecs->stop_func_name))
> >>> > + {
> >>> > + /* We landed in a shared library call trampoline, so it
> >>> > + is a subroutine call. */
> >>> > + handle_step_into_function (ecs);
> >>> > + return;
> >>> > + }
> >>>
> >>>
> >>>I am not sure I understand why the case above is not covered by the
> >>>test below. Is this to fix regression #1? I.e multiple frames?
> >
> >
> >Hmmm, good question...
> >
> >Although it does fix regression #2, it is not the only reason why we
> >added this test. It was also based on logic (see "After ... here is
> >what we found", in my previous message).
> >
> >I should admit that in the case at hand (regression #2), the unwinder
> >is having a hard time unwinding out of this code, and causes the
> >frame ID check to fail. I don't remember seeing several levels of
> >function call.
> >
> >However, I still thought that this test was necessary because we could
> >just as well have reached this trampoline one or more levels of function
> >call down, just as we end up stepping into undebuggable code in
> >regression #1.
>
> I'd not noticed this issue. Hmm, if GDB's single stepping then the
> second test should cover this case. It's when GDB is free running that
> we might find ourselves stopped IN_SOLIB_CALL_TRAMPOLINE. If it is the
> latter case then I'm not sure that silently single stepping away from
> where the program stopped is being helpful.
>
> Can you try the testsuite without that check? If the results are ok
> then (with other changes) commit it. If its not we need to re-think
> whats happening :-( Yes, this will mean it goes into 6.1.
Yes, there is a regression (regression #2 in my previous message) if
we leave that test out. It's been a while since I posted that patch,
so I don't remember the details anymore :-/. I'll dig again later today.
I did rerun the testsuite without it to double-check that my
recollection was right. And I also ran the testsuite with the frame_id
patch you recently posted, hoping that it might solve the extra
regression. Unfortunately, I am sorry to report that it actually
introduced another 5 or 6 regressions...
On the other hand, as soon as I add the check back, we're down to
zero regression (that is, with your frame_id patch as well).
More details later today about the failing test when the test above
is removed.
--
Joel