iOS - How to disable TSan for a third party framework

[KelvinJin]

Hey guys, is there a way to disable TSan for a specific third party framework in iOS?

I know that you can suppress a race/thread.. for a function but what if I want to suppress the entire Framework.

I say Framework here because I know you can disable a .o file or a .c file. But what if I have this .framework file...

[dvyukov]

We need to ask Kuba Mracek, but I can't figure out his github username...

[kcc]

pinging Kuba...

[kubamracek]

Is the framework is question actually instrumented with TSan instrumentation? If yes, then just skip instrumenting it (don't pass -fsanitize=thread, or if you have a project-wide -fsanitize then use -fno-sanitize=thread on the particular framework), and TSan should not report anything for that framework.

If that doesn't work, please be more specific about what the actual problem is, and how does your project setup look like.

[KelvinJin]

@kubamracek Thanks for pointing me to the right direction. It turns out that the framework is not instrumented with TSan. (I checked with nm xxx.framework/xxx | grep tsan)

And since ignore_noninstrumented_modules is enabled by default on Darwin (which I think is the same for iOS simulator), TSan shouldn't report any false positive.

I asked the question because I'm having a weird crash related to TSan when running my app. I posted the issue on Swift forum https://forums.swift.org/t/fatal-threadsanitizer-check-failed/16666. I was just trying to make sure this issue is not caused by any 3rd party frameworks I'm using.

[dvyukov]

Regardless of instrumentation, false positives and false negatives, tsan must not CHECK-fail. That's a bug in tsan runtime.

The first crash reminds me of:
https://reviews.llvm.org/D45646#1071868
@kubamracek that's still unfixed, right?

Re second crash, we need to see the CHECK failure message, stack alone is not so useful. But either way, it can well be some complex manifestation of the above bug.

[kubamracek]

Right, that all sounds right. Can you please file separate bugreports for both these issues, ideally with full crashlogs and stderr outputs, and if possible more info about what triggers the issues.

https://reviews.llvm.org/D45646 indeed is still unfixed.