This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH] test-skeleton.c: Do not enable M_PERTURB
- From: Florian Weimer <fweimer at redhat dot com>
- To: Siddhesh Poyarekar <siddhesh at sourceware dot org>
- Cc: libc-alpha at sourceware dot org
- Date: Thu, 23 Jun 2016 17:05:33 +0200
- Subject: Re: [PATCH] test-skeleton.c: Do not enable M_PERTURB
- Authentication-results: sourceware.org; auth=none
- References: <20160623144042 dot E649F41C390E5 at oldenburg dot str dot redhat dot com> <20160623144927 dot GA17911 at devel dot intra dot reserved-bit dot com>
On 06/23/2016 04:49 PM, Siddhesh Poyarekar wrote:
On Thu, Jun 23, 2016 at 04:40:42PM +0200, Florian Weimer wrote:
Over all, this decreases the realism of the tests because
it ensures that freshly allocated memory has a well-defined
bit pattern. It also causes malloc to take internal paths
different from regular application usage, and therefore
reduces malloc test coverage.
The well-defined bit pattern is more likely to catch any bad tests
though, which might make it valuable.
It could also cover up bugs which would otherwise be visible with fresh
allocations which contain only zeros.
Ideally, we'd run all tests twice, with different settings from the
environment (and also with valgrind).