Testing of musl

Testing of musl

[Miroslav Chabrecek]

[-- Attachment #1: Type: text/plain, Size: 440 bytes --]

Hello Everyone,

I tried to run libc-tests (http://nsz.repo.hu/git/?p=libc-test;a=summary)
on x86_64 architecture with musl library.
In results, there are many FAILED tests:
~5 Compilation API errors
~5. Runtime errors
~ 182 Math precision errors

So, I'm asking myself, if I run them in wrong way?
Do you run the test before release and try to have zero failures or what is
the testing process?

Thank you for answering my questions,
Miro

[-- Attachment #2: Type: text/html, Size: 1585 bytes --]

Re: Testing of musl

[Rich Felker]

On Wed, Oct 16, 2019 at 09:17:24AM +0200, Miroslav Chabrecek wrote:
> Hello Everyone,
> 
> I tried to run libc-tests (http://nsz.repo.hu/git/?p=libc-test;a=summary)
> on x86_64 architecture with musl library.
> In results, there are many FAILED tests:
> ~5 Compilation API errors
> ~5. Runtime errors
> ~ 182 Math precision errors
> 
> So, I'm asking myself, if I run them in wrong way?
> Do you run the test before release and try to have zero failures or what is
> the testing process?

That sounds roughly expected. Keep in mind this is not a musl test
suite that's intended to pass 100% when the tests are written
(although it is generally intended on the musl side that musl fix
failures), but a mostly libc-agnostic testset that aims to test
standard libc interfaces as heavily/pedantically as possible.

The API errors are confstr/pathconf/sysconf keys we don't have defined
because we were waiting for glibc to assign numbers for them so our
numbering would be aligned.

The strptime errors are new functionality (IIRC not in a published
standard yet) musl has not implemented.

The math/* errors are all minor precision or status flags corner
cases.

musl/pleval is a test of musl internals that's no longer possible
because we hid the internal symbols, but it does work (and pass) with
static linking.

malloc-brk-fail and possibly a few other similar tests fail on some
systems but do not indicate a bug in musl; rather they indicate that
the test has some issues with how it evaluates what it's trying to
measure.

Rich

Re: Testing of musl

[Szabolcs Nagy]

* Rich Felker <dalias@libc.org> [2019-10-16 07:44:49 -0400]:
> On Wed, Oct 16, 2019 at 09:17:24AM +0200, Miroslav Chabrecek wrote:
> > Hello Everyone,
> > 
> > I tried to run libc-tests (http://nsz.repo.hu/git/?p=libc-test;a=summary)
> > on x86_64 architecture with musl library.
> > In results, there are many FAILED tests:
> > ~5 Compilation API errors
> > ~5. Runtime errors
> > ~ 182 Math precision errors

you should grep out the lines starting with X from the math errors
those are expected and harmless.

(most of the other math errors are also expected, but i havent
filtered them out yet for various reasons)

> malloc-brk-fail and possibly a few other similar tests fail on some
> systems but do not indicate a bug in musl; rather they indicate that
> the test has some issues with how it evaluates what it's trying to
> measure.

i'd say it's a linux kernel bug.

but since it's unlikely to be fixed i might need to find another
way to test out-of-memory conditions.