This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Compiler for testing a bootstrapped glibc?
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Brooks Moses <bmoses at google dot com>
- Cc: Carlos O'Donell <carlos at redhat dot com>, libc-alpha <libc-alpha at sourceware dot org>
- Date: Thu, 6 Mar 2014 14:07:33 +0000
- Subject: Re: Compiler for testing a bootstrapped glibc?
- Authentication-results: sourceware.org; auth=none
- References: <CAOxa4KrctdEQWfXbfd7Xk-=cFcskzUSMASCFaa2Lgf3USPPmKg at mail dot gmail dot com> <530FA88D dot 9050009 at redhat dot com> <CAOxa4KpL5uO5okGqKbwXqwk_BBLV2eVZruWkP4fjj3Ns0JQXcQ at mail dot gmail dot com>
On Wed, 5 Mar 2014, Brooks Moses wrote:
> Joseph, I've forgotten from test your patch sequence, but I think you
> mentioned it: How far away are we from having all test files starting
> with tst-* or a related pattern? It seems to me that if we made sure
That's not a goal for the present patch series. Making all test outputs
end with .out (unless there's some reason a particular output can't end
with .out) is a goal (well, a cleanup that fits naturally in the series,
rather than something actually needed for getting PASS/FAIL summaries),
but one that's waiting on review of
<https://sourceware.org/ml/libc-alpha/2014-03/msg00010.html>.
> that was always followed, we could put in separate rules for building
> test programs that would (optionally) use a different compiler, and
> thereby at least avoid the rebuild-glibc step. If that seems
> feasible, I'd be happy to work on a patch.
My notion of support for installed glibc testing (not fully designed) is
that we should make sure that all dependencies etc. on files that glibc
installs go through makefile variables, so that for "make check
test-installed-sysroot=/some/where" the build process links against
installed files rather than build-tree ones. In that model, installed
testing would involve reconfiguring with the final compiler, but not
rebuilding (and if the dependencies were correct, "make check" would not
trigger rebuilding any of glibc - there would be another makefile target
to save files from the build process that are genuinely needed at test
time but not appropriate for normal installation).
(It should be possible to support both that and separate rules for
building test programs, if desired. I did wonder about separate rules
being of use to apply -Werror to installed code but not tests, but my
inclination now would be to apply -Werror everywhere and simply use
-Wno-error=whatever for tests that have an actual need to test deprecated
interfaces or otherwise do something that generates warnings.)
--
Joseph S. Myers
joseph@codesourcery.com