automatic musl-test

automatic musl-test

[Waldemar Brodkorb]

Hi musl hackers,

I finished up my first version of the replacement for
musl-test-framework. It is more general and can be used to test
other software than musl-libc. You can get it here:
http://www.openadk.org/cgi-bin/gitweb.cgi?p=adk-test-framework.git;a=summary
( git clone git://openadk.org/git/adk-test-framework )

A report generated from today, is here:
http://openadk.org/musl-test/

There is only sh4/sh4eb tests missing, I am still fighting
with a size limitation of the initramfs when using
qemu-system-sh4{,eb}.

Have fun,
        Waldemar

Re: automatic musl-test

[Szabolcs Nagy]

* Waldemar Brodkorb <wbx@openadk.org> [2014-04-09 17:07:59 +0200]:
> I finished up my first version of the replacement for
> musl-test-framework. It is more general and can be used to test
> other software than musl-libc. You can get it here:
> http://www.openadk.org/cgi-bin/gitweb.cgi?p=adk-test-framework.git;a=summary
> ( git clone git://openadk.org/git/adk-test-framework )
> 
> A report generated from today, is here:
> http://openadk.org/musl-test/
> 

hm something is wrong with the i686-git float tests

those are supposed to work with x87 and sse2 arithmetics as well

even functions with asm implementation fail which should not
depend on any compiler flags..

either the test system is broken somehow or you are not linking
to the latest musl

> There is only sh4/sh4eb tests missing, I am still fighting
> with a size limitation of the initramfs when using
> qemu-system-sh4{,eb}.
> 
> Have fun,
>         Waldemar

Re: automatic musl-test

[Szabolcs Nagy]

* Szabolcs Nagy <nsz@port70.net> [2014-04-09 22:16:24 +0200]:
> * Waldemar Brodkorb <wbx@openadk.org> [2014-04-09 17:07:59 +0200]:
> > A report generated from today, is here:
> > http://openadk.org/musl-test/
> > 
> 
> hm something is wrong with the i686-git float tests
> 

can you send me some test binaries built by your toolchain?

src/math/fenv.exe
src/math/fenv-static.exe
src/math/exp.exe
src/math/exp-static.exe
src/functional/strtod.exe
src/functional/strtod-static.exe

static binaries are not built for src/math by default but you
can build them like

  make src/math/exp-static.exe

i suspect the test system is broken (maybe the i686 emulator?)

Re: automatic musl-test

[Szabolcs Nagy]

* Szabolcs Nagy <nsz@port70.net> [2014-04-09 22:36:53 +0200]:

> * Szabolcs Nagy <nsz@port70.net> [2014-04-09 22:16:24 +0200]:
> > * Waldemar Brodkorb <wbx@openadk.org> [2014-04-09 17:07:59 +0200]:
> > > A report generated from today, is here:
> > > http://openadk.org/musl-test/
> > > 
> > 
> > hm something is wrong with the i686-git float tests
> > 
> 
> can you send me some test binaries built by your toolchain?
> 

ok i downloaded the adk-test-framework and the i686
root fs has glibc in it..

that explains all the errors :)

may be you should report the bugs to glibc..

(i knew about several fp errors in glibc but i never had
a recent enough glibc to test them)

Re: automatic musl-test

[Waldemar Brodkorb]

Hi, 
Szabolcs Nagy wrote,

> * Szabolcs Nagy <nsz@port70.net> [2014-04-09 22:36:53 +0200]:
> 
> > * Szabolcs Nagy <nsz@port70.net> [2014-04-09 22:16:24 +0200]:
> > > * Waldemar Brodkorb <wbx@openadk.org> [2014-04-09 17:07:59 +0200]:
> > > > A report generated from today, is here:
> > > > http://openadk.org/musl-test/
> > > > 
> > > 
> > > hm something is wrong with the i686-git float tests
> > > 
> > 
> > can you send me some test binaries built by your toolchain?
> > 
> 
> ok i downloaded the adk-test-framework and the i686
> root fs has glibc in it..

Sure, the base system is glibc.
 
> that explains all the errors :)
> 
> may be you should report the bugs to glibc..
> 
> (i knew about several fp errors in glibc but i never had
> a recent enough glibc to test them)

Hmm, but the libc-test is compiled with musl libc via a special
specs file. So no glibc code involved while executing the test
binaries.
$ readelf -a libc-test/src/math/fenv.exe|grep interpreter
      [Requesting program interpreter: /lib/ld-musl-i686.so.1]

Upps, and here we see, that the wrong interpreter is linked into the
binary. I fixed it. Get a new REPORT.

best regards
 Waldemar