A few more gcc-13 bugs
Another 6 months have passed since I wrote anything about
gcc-13 should release around April next year (in about 6 months).
Being in the middle I’m dumping another set of bugs bugs I encountered:
fontforgemiscompiled by vectorizer
pcremiscompiled by range analyzer in middle end
proxmark3ICE in C typedef handling of vectored types
av1ICE in gcc vectorizer
boostICE in C++ template expansion
linux-4.19ICE in ia64 backend
llvmtest suite hangup in range analyser
- https://gcc.gnu.org/PR106334: LTO ICE on
nixwhen merging debug info
- https://gcc.gnu.org/PR106540: LTO ICE on
nixwhen merging debug info (again)
- https://gcc.gnu.org/PR106831: decimal code miscompilation on
- https://gcc.gnu.org/PR105852: ICE in template instantiation on
- https://gcc.gnu.org/PR105608: precompiled headers ICE in large macros
- https://gcc.gnu.org/PR106905: vectorizer ICE on
I won’t focus on any of the bugs in detail here as they are quite boring this time.
As usual ICEs are simple to minimize and reduce. Runtime failures are hardest to deal with. Especially tough one was the llvm-14 hangup. It took me a few attempts at minimizing the example. I failed all of them. Luckily someone found a smaller unrelated reproducer and the bug was fixed.
Bug pace makes it about 2 new bugs a week. Still barely manageable :)
As usual here is a set of upcoming changes in
https://gcc.gnu.org/gcc-13/changes.html. A few more
would have to be added to various upstreams.