This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Remove Linuxism from tst-tls-atexit
- From: Roland McGrath <roland at hack dot frob dot com>
- To: Siddhesh Poyarekar <siddhesh at redhat dot com>
- Cc: libc-alpha at sourceware dot org
- Date: Tue, 14 Jul 2015 18:00:28 -0700 (PDT)
- Subject: Re: [PATCH] Remove Linuxism from tst-tls-atexit
- Authentication-results: sourceware.org; auth=none
- References: <1436883383-6903-1-git-send-email-siddhesh at redhat dot com> <20150714194025 dot 4608E2C39FA at topped-with-meat dot com> <20150715001837 dot GV2634 at spoyarek dot pnq dot redhat dot com>
> EXPECTED_SIGNAL makes SIGSEGV OK for the entire test case and we don't
> want that. The dlclose inside the thread function LOAD for example
> will cause a segfault in __tls_call_dtors if it incorrectly unloads
> the DSO.
Hmm. Perhaps there is too much being tested in this one test, then?
One of the things on my list is to clean up the various test cases that
rely on signal handling when they are not testing something related to
signal handling. It makes those tests unusable on non-POSIX
configurations (i.e. currently only NaCl) that do not support signals.
So adding another such case is kind of going backwards. In this case,
going from gratuitous Linuxism to just thorough POSIXism is still an
incremental improvement (it should fix it for Hurd, e.g.). So that is
fine for now if it really is unavoidably much harder to write a good
test without catching a signal. But still not ideal.
Thanks,
Roland