Re: Is SANITIZE=leak make test unreliable for anyone else?

From: Jeff King <peff@peff.net>
Date: Oct 4, 2023, 13:21 UTC
Message-ID: <20231004132132.GC607079@coredump.intra.peff.net>
In-Reply-To: <878r8j2mu1.fsf@email.froward.int.ebiederm.org>

On Tue, Oct 03, 2023 at 08:33:26PM -0500, Eric W. Biederman wrote:

Show 5 quoted lines

> My big question is:
> 
>     Do other people see random test failures when SANITIZE=leak is enabled?
> 
> Is it just me?

Yes, I've seen this. You mentioned that you were testing with v2.42, which lacks 370ef7e40d (test-lib: ignore uninteresting LSan output, 2023-08-28). Try using the current version of 'master', or just cherry-picking that commit onto v2.42.

A few other tips to avoid confusing results (though they at least do not vary from run to run):

  - use the LEAK_LOG option, since you otherwise miss some cases (it
    looks like you already are from what you posted above)

  - gcc and clang sometimes produce different results. Right now I get
    no leak from gcc on t9004, but clang reports one (I think clang is
    right here)

  - turn off compiler optimizations; we've had cases where code
    reordering/removal creates false positives. Oh, hmm, I forgot we do
    this by default since d3775de074 (Makefile: force -O0 when compiling
    with SANITIZE=leak, 2022-10-18), so your v2.42 should be covered.

-Peff