[Bug tree-optimization/114872] [13/14/15 Regression] Miscompilation with -O2 after commit r13-8037

["sjames at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114872

--- Comment #8 from Sam James <sjames at gcc dot gnu.org> ---
element.i is unfortunatley huge. It's hard to analyse things without a
standalone testcase, but it's even harder without _something_ one can run.

I'd suggest:
1) giving instructions to reproduce the crash assuming someone knows nothing
about sage and cython;
2) taking sage and cannibalising it (first to directly call
`libgap.AbelianGroup(0,0,0)`, then you can cut things down with lots of
removals + gdb so that the important caller of Py_XDECREF gets passed with the
same args; it's harder if there's a lot of state involved though, of course)
3) build element.i with -fdump-tree-all -fdump-unnumbered -fdump-noaddr (can
also try e.g. -da, see
https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html) at the bad commit
and before, and diff the produced dumps, and show us the *first* dump which
differs between the two