Hi, On 22/07/18 11:51, Jacob Nevins wrote: > Marko Lindqvist wrote (Dec 2016): >> As the crash is in atexit handlers when the program is closed, this >> seems like upstream bug #25364: http://gna.org/bugs/?25364 > > gna.org has gone away, but the Internet Archive captured the final state > of this ticket, which is that nothing has been done upstream to address > this. We have no ticket for it in our current tracker. > <https://web.archive.org/web/https://gna.org/bugs/?25364> > > The conversation in Gna indicated that this was some interaction with > distro defaults, rather than a simple upstream bug: > > mir3x: > | I prepared 3 patchs, try if any of them works. > | It just crashes at destructor of static QString, no idea why, but I saw > | similar reports when qt was built for wrong arch. > > mir3x: > | ML can u check with gcc -v if your gcc has --enable-__cxa_atexit ? > | If no then it must be that issue. > > Marko: [gcc -v output, presumably from Debian testing, not including > --enable-__cxa_atexit] > > mir3x: > | "--enable-__cxa_atexit" is required for the G++ compiler to produce > | static destructors that fully comply with the C++ specification. > | So p1.patch might work. > | But its probably some debian bug, all static c++ classes will segfault. > > I've attached the 'p1.patch' referred to, rescued from gna.org. I don't > have any record that anyone tried it.
Not sure about the cause of this bug, but --enable-__cxa_atexit has been enabled by default in gcc for over 10 years now (see gcc/config.gcc in the gcc source). I doubt the bug is related to that. A total guess, but most of the "crash at exit" bugs I've seen were caused by some memory corruption during program execution which is only highlighted when the destructors are run. James
signature.asc
Description: OpenPGP digital signature
