Georg Baum wrote:
> Peter Kümmel wrote:
> 
>> I've replaced the complete boost tree (.cpp included) with the version
>> from trunk. But it also makes no difference.
>>
>> BTW should we update the 1.4.x tree to boost 1.34?
> 
> Not before it is released. We might want to update to 1.33.1, we had that
> for some time without problems in trunk, but as long as 1.32 works I see no
> real reason. It was a wild guess that this might be a boost bug, but that
> seems not to be the case, and honestly I would be really surprised if it
> qwas one, because boost is widely used.
> 
>> I think the reason for the crash is :
>>
>> lyx-qt3.exe!`anonymous namespace'::`dynamic atexit destructor for
>> 'theCuts''()  + 0xd bytes     C++ msvcr80d.dll!102020c6()
>> lyx-qt3.exe!LyXServer::`scalar deleting destructor'()  + 0x27 bytes   C++
>> msvcr80d.dll!10201e00()
>> lyx-qt3.exe!lyx_gui::exit(int status=0)  Line 281 + 0xc bytes C++
>> lyx-qt3.exe!QuitLyX(bool noask=false)  Line 220 + 0x7 bytes   C++
>>
>>
>> the free variable theCuts, it is destructed first, and I assume the values
>> are later on used.
> 
> I don't think so. The backtrace tells about a problem in the destructor, not
> in using theCuts. If theCuts would be used after destruction, then the
> destructor would not occur in the backtrace, but the code using theCuts
> afterwards, no?
> 

above backtrace was not complete, it only shows the destruction of theCuts
the more complete is here:

http://www.mail-archive.com/[email protected]/msg101523.html


Nevertheless static (global) variables only make problems.

>> Because the destruction order is not specified by the 
>> standard it is possible that we have no problem with gcc.
>>
>> But it seems the code depends on the order of destruction.
>>
>>
>> I think, when above path really fixes the bug (currently I use the trunk's
>> boost) we could also ship the windows binaries.
> 
> I would prefer to understand what the problem is. Your code seems safe, but
> your guess does not look right to me.
> 
> 
> Georg
> 
> 


-- 
Peter Kümmel

Reply via email to