Hi again,
I'm not sure whether anyone is paying attention, but here goes...
Steffen Mueller wrote:
there have been occasional bad crashes of an application in whose
development I'm involved in (Padre). With your generous help, I managed
to build a debugging perl+wxWidgets+Wx combination and finally obtained
a backtraces for such crashes. I always suspected a combination of
ithreads with wx to be the culprit and unfortunately, it turns out I was
right. It seems to happen only when a background thread is joined.
Specifically during the resulting cleanup. I'm wildly speculating now:
Maybe some of the Wx objects with underlying non-ithread-aware C++
implementation are being cloned on thread creation and dubiously freed
on thread exit. Or maybe it's due to Thaw being called on a window that
wasn't Freeze'd (see backtrace).
You can find an example backtrace with some additional debugging info here:
http://padre.perlide.org/attachment/ticket/287/backtrace
This blog post gave me an idea:
http://peter.makholm.net/2009/05/15/persistence-and-objects-from-xs-modules/
What if the WindowUpdateLocker object is cloned as part of a thread
creation and the clone is cleaned up on thread exit? This could go wrong
in various ways.
Let's suppose the underlying C++ object isn't cloned correctly and just
the Perl-level object is cloned. Maybe it shouldn't have been cloned at
all in the first place. After all, the Wx stuff shouldn't be in the
non-GUI thread.
Scenario 1:
The thread will DESTROY (=> C++ delete) it on join(). Then, the next
access to it from the main/GUI thread should crash with a segmentation
fault due to a double free. This is not the case (see backtrace). It
crashes in the thread cleanup phase.
Scenario 2:
The object is destroyed analogously to scenario 1, but by the main/GUI
thread *while the worker thread is still running*. After all, a
WindowUpdateLocker should be a short-lived object. Then, the crash (i.e.
second delete) would appear during global destruction of the interpreter
that's running in the thread.
Does this make any sense or am I barking up the wrong tree? Any ideas on
how to debug this? Will valgrind catch this? I'm a bit of a newbie in
terms of this heavy C debugging, particularly if it involves perl, wx,
and concurrency.
Cheers,
Steffen
