gh-148832: Use -icf=0 in BOLT_APPLY_FLAGS

[ashm-dev]

• Issue: Build: BOLT -icf=1 breaks type-slot function-pointer identity, causes SIGSEGV #148832

[ashm-dev]

@corona10 @erlend-aasland could you please review this PR?

[ashm-dev]

Gentle ping — this is a small, low-risk build-flag change (-icf=0 in BOLT_APPLY_FLAGS) that's been awaiting review for ~4 weeks. @AA-Turner @emmatyping you were both auto-requested as reviewers — would either of you have a moment to take a look?

[corona10]

lgtm!

[geofft]

FYI I totally forgot about this until just now but #134642 uses -icf=safe on LLVM versions that support it (LLVM 20 and up) and leaves the flag out entirely on older versions. Safe ICF doesn't run into the issue with function pointer equality. Do we need to explicitly set -icf=0 instead of leaving the flag out? I think it's off by default.

[ashm-dev]

@geofft AFAIK it's off by default. If you believe leaving it out causes regressions on older LLVM builds, please feel free to verify against the BOLT docs and provide a test case.

[geofft]

No, I believe leaving it out is plenty, just wanted to confirm (since the current behavior / the behavior in this PR is to explicitly say -icf=0).

[ashm-dev]

No, I believe leaving it out is plenty, just wanted to confirm (since the current behavior / the behavior in this PR is to explicitly say -icf=0).

PRs welcome. Unsubscribing from this thread.