Details
- Reviewers
hctim - Commits
- rG448c89415b7a: [HWASan] Update use-after-scope test to use ARM64.
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
This test is really hard to validate changes to. I mean, it's basically "$COMPILER produces $BLOB", right? I'm not really sure what are the interesting parts I'm supposed to be looking for.
Is it possible to reduce the test to check the actual conditions?
Like, for standard_lifetime, we have lifetime start -> shadow setup (some minimal UID like __hwasan_generate_tag) -> lifetime end -> shadow teardown (some minimal UID like memset-zero).
This is a auto-generated golden test: ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
I think there is a value in that, to make sure any changes don't exceed the intended scope. There are more focused tests as well in this directory.
Maybe I'm just biased against golden-file tests like this because they're almost impossible to review (apart from "I trust the author"), and they increase churn significantly when one wants to make instrumentation changes. But, I do agree there's some value in the "didn't mess things up" check.
If that's all we're aiming for, would it make sense to have use-after-scope-arm64.ll and use-after-scope-x64.ll?
Or we just put x86 and aarch64 into the same file, by using -target. That will make the file very large (2x the expectation comments), but makes sure they don't diverge. I would prefer that. WDYT?