Re: RFC: fix PR backtrace/15558

[Pedro Alves <palves@redhat.com>]

On 04/14/2014 07:06 PM, Tom Tromey wrote:
>>>>>> "Andrew" == Andrew Pinski <pinskia@gmail.com> writes:
> 
> Andrew>   What happened to this patch, I don't see any reference to it latter
> Andrew> on?  I think I am running into the same problem as this patch is
> Andrew> fixing.
> 
> I never got back to fixing it according to Pedro's review.
> 
> As I recall it isn't entirely trivial because for a good fix one would
> need to expose some new Python methods for use by the frame filter code.

Hmm.  Not sure what you're thinking of.  We discussed auditing all
get_prev_frame uses and see if they are better replaced by
get_prev_frame_1, but I don't think that should hold back
fixing the inline frame unwinder.  In the particular case of the
inline frame unwinder, I believe we should simply be making
inline_frame_this_id call get_prev_frame_1 instead of
get_prev_frame.  Let me give it a try.  Hmm, your test doesn't
trigger the bug for me as is, but
I can trigger it in the CLI without Python.  Ah, a "flushregs"
is missing to force re-unwinding and recomputing the frame id.
A sec and I'll post something.

-- 
Pedro Alves