It was pointed out in #32862 (comment) that that PR uncovered a false positive libc++ tsan warning. It was not initially clear to me why the issue was not happening in the TSAN CI. The reason it does not, is because of DEBUG_LOCKORDER
. Dropping that from the TSAN CI job does shows the issue. i.e #33081.
I am wondering if the TSAN job should be split up, so that TSAN & DEBUG_LOCKORDER
do not interfere with each other. It seems better to not have false positives (or real issues/anything else) in other tooling suppressed, by our own tooling, and instead be able to directly suppress/document issues, and properly report them upstream.