Bug#730830: ocaml: Please add powerpcspe to the list of native architectures

[Stéphane Glondu <glondu@debian.org>]

clone 730830 -1
retitle -1 FTBFS with recent binutils-dev
severity -1 serious
thanks

Le 01/12/2013 11:09, Roland Stigge a écrit :
>>>> Aren't those registers also used by generated code?
>>>
>>> The registers can't be used since they don't exist and the assembler
>>> would show errors if someone would try to use them. Now (with my patch),
>>> the assembler doesn't show errors anymore.
>>>
>>>> Does ocamlopt
>>>> produce working executables with this patch? Does the whole test-suite pass?
>>>
>>> The package builds fine, including successful dh_auto_test target.
>>> Please tell what else I should run as test-suite.
>>
>> The test suite is not run during the package build process with version
>> < 4. Please try with the version from experimental.
> 
> Thanks for the hint. Some tests are failing, and the whole build
> timeouts at some point on the buildd. (I'm attaching the build log.) I
> guess the native arch:powerpcspe should wait until this is fixed?

almabench is very intensive and also times out on some buildds. I am
planning to disable it on $(SLOW_ARCHITECTURES) (defined in
debian/rules, which includes powerpc). I can add powerpcspce there as well.

Thanks for the build log. It doesn't look severely broken, but test
failures must be further investigated:

 * Native backtrace tests are completely broken, but it is also the
   case on powerpc but it seems to have been fixed upstream.

 * tests/lib-bigarray/fftba.ml: no idea what's going on. It would be
   nice to get tests/lib-bigarray/fftba.output and compare with
   tests/lib-bigarray/fftba.reference.

 * tests/misc/nucleic.ml: fails with
     Fatal error: exception Assert_failure("nucleic.ml", 3205, 9)
   which is very weird and looks difficult to figure out.

The first point is explained and would not prevent a switch, but I think
the last two should at least be explained before making powerpcspe
native. Is there a powerpcspe porterbox available for DDs?

> In this case, feel free to close this bug by just fixing the general FTBFS.

There are two bugs, actually. Cloning this one.


Cheers,

-- 
Stéphane