13 Oct
2016
13 Oct
'16
7:19 p.m.
Hi Ignacy,
Ignacy Gawędzki wrote,
On Thu, Oct 13, 2016 at 06:10:15PM +0200, thus spake
Ignacy Gawedzki:
Sorry I visited ELCE 2016 in Berlin, so I had no time to give you
the required information.
I will sent you buildroot instructions later.
If you want you can try it with my embedded-test script (you need
mksh and qemu installed):
git clone git://git.embedded-test.org/git/embedded-test
cd embedded-test
git clone git://git.uclibc-ng.org/git/uclibc-ng
mksh embedded-test.sh --arch=mipsel --libc=uclibc-ng \
--libc-source=uclibc-ng --test=libc
Only 3 tests should fail.
Then apply your patch and run it again, it will automatically
do a complete clean build.
best regards
Waldemar
On Thu, Oct 13, 2016 at 05:32:35PM +0200, thus
spake Ignacy Gawedzki:
Okay, apart from the fact that you're not supposed to call isfinite
nor isnormal with anything except floating point arguments (tests in
glibc do this by using intermediate variables initialized with the
integer literals), and should cast those literals explicitly to FLOAT,
I discovered that I can disable glibc tests by passing UCLIBC_ONLY=1
(these errors were triggered by my local native glibc). :/
But still, some tests that run a mipsel binary using qemu-mipsel fail
miserably. Some because they just assume the binary is run natively
(as test/crypt which defines a WRAPPER variable to run the binary
using env), which is simply not working when SIMULATOR_uclibc equals
some qemu binary. Some other because of reasons unknown to me at this
time: test/unistd/clone (somehow the child processes seem to be
un-wait4-able with qemu).
I'd *really* like to reproduce the failed tests because of my patch,
but without any more information (actual uclibc-ng's .config,
buildroot's .config, exact way to run the testsuite for mipsel on an
amd64 machine), I'm fighting with unrelated problems so far. On Thu, Oct 13, 2016 at 05:17:32PM +0200, thus
spake Ignacy Gawedzki:
I fixed a few things in test/Makefile and test/Rules.mak (patch
coming soon) and it looks like make check now works with a separate
build directory. But still, I'm getting errors in test/math regarding
__builtin_isfinite and the like not getting floating-point arguments.
To me it looks like the test code is wrong, but I may be wrong myself.
Any idea how to fix that? On Thu, Oct 13, 2016 at 12:44:38PM +0200, thus
spake Waldemar Brodkorb:
> Hi Ignacy,
>
> I tried v4, but it adds regressions for other architectures.
> For example MIPS Little endian running uClibc-ng testsuite.
>
> Any idea? Can you try to reproduce the regressions on your side?
I just tried by building a buildroot toolchain for mipsel. So far
make check fails even without my patch telling me that it doesn't know
how to make target test/Makefile, needed by termios.
It looks just like make check doesn't work with a separate build
directory (make O=/path/to/somewhere/else check). Do you confirm?