On Wed, Apr 05, 2017 at 02:35:12AM -0400, al davis wrote:
> So the reason for the crash is probably that during exit, it ran the
> destructor, then attempted to use the object that has already been
> destroyed.

yes. if you switch to "wrong link order" with gcc (for linux/gcc, that
is globals.cc at the end), it segfaults all the same. just gcc seems to
optimize out the _map=NULL assignment. or something like that. in
effect, you cannot "see" it segfaulting with -O2, try -O0.

> It's still suspicious because I don't know what std::map does in cases
> like this.

the real problem seems to be the link order. what is the motivation to
reverse the link order (freebsd+clang?) in the first place? shouldn't
this break other projects as well?

and then: automake+libtool is trying hard to provide a platform
independent linking experience. gnucap+autotools does use libtool for
that particular step. it would be interesting to know if they provide
countermeasures. or the other way round: they might be interested in
this sort of linker mis(?)behaviour...

cheers
felix

_______________________________________________
Gnucap-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/gnucap-devel

Reply via email to