On Thu, Mar 07, 2019 at 09:07:15AM +0100, Stephen Kitt wrote: > Hi Bill, > > On Mon, 4 Mar 2019 00:39:00 +0100, Bill Allombert <ballo...@debian.org> wrote: > > Since the upgrade to 8.2.0-17, gcc-mingw-w64-x86-64 miscompiles PARI/GP. > > <https://pari.math.u-bordeaux.fr> > > > > The native linux version is fine though. > > The bug is caused by longjmp causing wine to crash with > > > > 0198:err:seh:setup_exception stack overflow 320 bytes in thread 0198 eip > > 000000007bc9a41c esp 00000000001314d0 stack 0x130000-0x131000-0x330000 > > > > when called in a function doing an unrelated setjmp _after_ the longjmp > > (in other functions longjmp work fine), and it does not occurs with the > > debug build. > > > > One of the logs for the broken version is at > > <https://pari.math.u-bordeaux.fr/buildlogs/e9a1cdf2-df47-4a47-8324-df35d1648c58> > > (the test-suite fails under wine). > > I use the following kit to crossbuild: > > https://pari.math.u-bordeaux.fr/pub/pari/windows/paricrossmingwkit.tgz > > > > So if you have a suggestion... > > I don’t have any practical suggestion right now, but given that there’s a > test suite I’ll bisect to identify the source of the problem.
Thanks! I tried __builtin_setjmp/__builtin_longjmp and while it seems to work __builtin_longjmp only allows val to be 1, which would require to change the code. Cheers, -- Bill. <ballo...@debian.org> Imagine a large red swirl here.
