[Texmacs-dev] Non-reproducible crash in texmacs-win32

[David Allouche]

On Sat, Jun 14, 2003 at 11:23:25PM -0400, Bill Page wrote:
> 
> Second, I do seem to have a strange problem that
> I am having trouble duplicating on demand. After
> poking around at the tool bar and loading some help
> text (e.g. configuration), then looking around for
> some way to change the cursor, giving up and then
> open a new document and entering a sentence or two,
> suddenly texmacs exits, no reason or error message
> given. Just gone. It has happened twice now under
> more or less the same circumstances but I am still
> not able to make it happen reliably. Sometimes I
> can work for 15 minutes or more before it happens.
> Other times it might have in the first 30 seconds.
> Any idea about how I can give you more useful
> feedback?

As far as I know, texmacs-win32 is compiled using Visual Studio, so we
should be careful comparing win32 problems with experiences we had
with g++.

However, what you describe looks awfully like the instability caused
by the buggy inliner of g++-3. If I had a windows system on which to
do some testing (and the time) I would probably try:

  -- compiling texmacs with the fast-allocator disabled. That caused
     the bugs induced by g++ to be much easier to reproduce,
     especially when using a malloc debugging utility.

  -- compiling texmacs without optimizations. This worked around the
     buggy inlining code in g++.

Well, since the compiler is completely different, this speculation
should not be taken too seriously. But it seems that the very peculiar
use of C++ made by texmacs (esp. its weird reference counting system)
makes it vulnerable to bugs in rarely used execution paths of
compilers.

-- 
                                                            -- ddaa