Anyone looking at the 'setup.exe' 2.510.2.1 crashing problem?

[Christopher Faylor]

On Thu, Sep 08, 2005 at 09:05:27AM -0700, Brian Dessent wrote:
>Dave Korn wrote:
>
>>   Or perhaps that "__result may be used uninitialised" warning in the stl
>> stuff is for real after all, and the iterator is actually returning an
>> uninitialised value.
>
>Sadly, it looks like something really is broken in gcc 3.4.4, and this
>warning is not benign.  By inserting a bunch of debug statements, I
>found that the crash induced with cgf's steps occurs at the "ExitThread
>(0)" at the bottom of do_ini_thread_reflector.  I installed gcc 3.3.3
>and can no longer reproduce the crash.
>
>See also <http://gcc.gnu.org/ml/gcc/2005-09/msg00159.html>.  It looks as
>if there might be problems with STL code and -O2 and sjlj exception
>handling.
>
>I'm going to release a setup.exe built with 3.3.3 right now, as a
>temporary work around.

Have you tried building with -Wstrict-aliasing?  I found some problems in
newlib that were tickled with gcc 3.4.4 that way.

cgf