CI: Sanitize for undefined behavior

[whitequark]

What are the reasons/motivation for this change?

Yosys includes some amount of buggy code that invokes undefined behavior, e.g. #4844. This should be fixed and stay fixed.

Explain how this is achieved.

By running the Undefined Behavior Sanitizer on each commit/PR. UBSan does not have false positives and has a slim impact on performance, so this is done unconditionally.

If applicable, please suggest to reviewers how they can test the change.

Add SANITIZER = undefined to your Makefile.conf, then run make clean && make && UBSAN_OPTIONS=halt_on_error=1 make test.

[whitequark]

We should probably exclude the whole libs/* subtree from UBSan checking since it's not like we're going to fix those bugs anyway. I think we should use -fsanitize-ignorelist=FILE where FILE contains:

src:libs/*

or maybe

[alignment]
src:libs/fst/*

What should FILE be? I'm thinking .sanitize-ignorelist or something.

[whitequark]

The CI does fail in an unfortunately opaque way (the ubsan output gets redirected to tests/simple/memory.err) but I've confirmed that this is indeed #4844:

kernel/celledges.cc:262:30: runtime error: shift exponent 32 is too large for 32-bit type 'int'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior kernel/celledges.cc:262:30 in 

[KrystalDelusion]

There is a Report errors step in the tests yaml that runs find tests/**/*.err -print -exec cat {} \;. It doesn't always work, but in this case it does give the ubsan output you're looking for:
https://github.com/YosysHQ/yosys/actions/runs/12762313299/job/35571164274?pr=4845#step:12:4995

[whitequark]

Ah, that's handy.

[povik]

The CI is green now, someone please review

[KrystalDelusion]

I think

[alignment]
src:libs/fst/*

in .sanitize-ignorelist is good (with appropriate gating I guess?)

[povik]

The one instance of UB has been easy to patch. I'd only start the exclusion list if issues with that library keep coming back.

[KrystalDelusion]

Oh! I missed that you patched the library 😅 lgtm