Sorry I don't know enough about the ustat usages, resigning as a reviewer.

This will be included in an upcoming documentation change I am making.

In D46961#1101650, @dvyukov wrote:

Making it static const makes the compiler use cxa_guard_acquire & cxa_guard_release which we don't want to be able to stay free of libc++ dependencies.

I meant keeping it as class member (not making it function-local static).

In D46961#1101608, @dvyukov wrote:

• const uptr BatchClassId = SizeClassMap::ClassID(sizeof(QuarantineBatch));
  
  Shouldn't it be constexpr? Or at least static const?

The changes to the local cache were not worth doing. Reverting those.

Addressing a couple of comments from Aleksey.

I fixed at some point COMPILER_RT_HAS_MCRC_FLAG being defined for x86_64, it was due to warnings being disabled, making all the cmake checks pass.
See https://reviews.llvm.org/D46079. It might be what's happening here.

And I forgot:

• the SizeClassInfo mutex should be a StaticSpinMutex since it's zero initialized and doesn't need to be "constructed";

Changing the break condition to Dmitry's suggestion.

Currently not compatible with D45997's COMPILER_RT_ENABLE_WERROR ON.

Add description of the function in the interface header.

Added a very generic test.

Sorry, you were right! The initial version was less complicated.
I'm updating this to fallback to it, with the style changes included.

asan_test.cc leverages an intercepted prctl with the ThreadedTestFunc test which calls:

In D45838#1073685, @dvyukov wrote:

Tests are not linked with interceptors?

In D45838#1073197, @dvyukov wrote:

I am not following. Isn't it a test for prctl interceptor? Why are we removing useful tests?
Perhaps it's not the most critical functionality. And perhaps SanitizerSetThreadName needs to be moved to the test itself. But still it's a test.

Readding SetEnv.

It seems that the existing weak OnPrint could fulfill the purpose of an application defined "callback" .
Right now it's not be used by any of the sanitizers, and having a weak function & 8 callbacks serving the same purpose seems overkill.

Adding one cfi-diag.

Rebasing.

Anyone else for comments?