Skip to content

Fix msan false positive in the swiss table#2092

Open
AustinSchuh wants to merge 1 commit into
abseil:masterfrom
AustinSchuh:swiss_table_msan
Open

Fix msan false positive in the swiss table#2092
AustinSchuh wants to merge 1 commit into
abseil:masterfrom
AustinSchuh:swiss_table_msan

Conversation

@AustinSchuh

Copy link
Copy Markdown

The swiss table implementation is well optimized and uses SIMD instructions. This doesn't play well with MSAN. There is already a macro to ignore uninitialized warnings, treat MSAN the same way.

The swiss table implementation is well optimized and uses SIMD
instructions.  This doesn't play well with MSAN.  There is already a
macro to ignore uninitialized warnings, treat MSAN the same way.

Signed-off-by: Austin Schuh <austin.linux@gmail.com>
@mkruskal-google

Copy link
Copy Markdown
Contributor

Thanks for the PR! Can you add a test case reproducing this problem as well? We don't have MSAN tests for the OSS code atm, but I can verify that internally

@mkruskal-google mkruskal-google self-requested a review June 27, 2026 19:56
@AustinSchuh

AustinSchuh commented Jun 27, 2026

Copy link
Copy Markdown
Author

Thanks for the prod. I've been trying to isolate this down to a single test I can upstream, and I think I've conflated 2 issues. A decent number of absl tests fail when I'm building optimized with msan, but pass under fastbuild. I think I conflated that with this fix. This makes progress under -c opt, but has probably unintended consequences on fastbuild (ignoring more issues). Happy to close this out if you want.

The latest https://github.com/bazel-contrib/toolchains_llvm now has support for building with msan. That's how I'm doing my tests.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants