LDC on ARM and PowerPC

[Alex Rønne Petersen]

Hey folks,

What's the state of LDC on these two architectures? Is it anywhere
near a point where we should consider adding them to the ci.lycus.org
LDC job?

Regards,
Alex

[Kai Nacke]

On 25.05.2013 07:52, Alex R�nne Petersen wrote:
> What's the state of LDC on these two architectures? Is it anywhere
> near a point where we should consider adding them to the ci.lycus.org
> LDC job?

Hi Alex,

LDC on Linux/PPC64 is nearly finished - currently only 5% of the tests
do not pass. A CI server for this architecture would be really cool to
avoid regressions. :-)

PPC32 and ARM are only in its early stages. Both miss important
implementations like vararg support and thread support. David did some
work on ARM but this is not yet committed.

Status is also documented on the wiki:
http://wiki.dlang.org/LDC#Project_status

Regards
Kai

[David Nadlinger]

On 25 May 2013, at 9:44, Kai Nacke wrote:
> LDC on Linux/PPC64 is nearly finished - currently only 5% of the tests
> do not pass. A CI server for this architecture would be really cool to
> avoid regressions. :-)

Which 5% of the tests?

Would it make sense to "XFAIL" them (i.e. report a GitHub issue and
selectively disable them) so a CI system could actually provide a binary
regression indicator?

> PPC32 and ARM are only in its early stages. Both miss important
> implementations like vararg support and thread support. David did some
> work on ARM but this is not yet committed.

Sadly, yes. I hope to be able to clean up and (re-)commit my stuff
tomorrow, but I really can't promise anything.

GC, threading, … is completely untested, but Phobos "Hello World"
works out of the box, which I think is an important step as it allows
other people to come in and iteratively work on fixes rather easily.

David

[Kai Nacke]

On 25.05.2013 20:43, David Nadlinger wrote:
> On 25 May 2013, at 9:44, Kai Nacke wrote:
>> LDC on Linux/PPC64 is nearly finished - currently only 5% of the tests
>> do not pass. A CI server for this architecture would be really cool to
>> avoid regressions. :-)
>
> Which 5% of the tests?

Here is the failure list of my last run:

95% tests passed, 54 tests failed out of 1036

Total Test time (real) = 9288.83 sec

The following tests FAILED:
18 - druntime_src_core_thread_debug_run (Failed)
20 - druntime_src_core_thread_release_run (Failed)
38 - druntime_src_core_math_debug_run (Failed)
40 - druntime_src_core_math_release_run (Failed)
640 - phobos_std_bitmanip_release_run (Failed)
642 - phobos_std_regex_debug_run (Failed)
644 - phobos_std_regex_release_run (Failed)
656 - phobos_std_file_release_run (Failed)
673 - phobos_std_numeric_debug_build (Failed)
674 - phobos_std_numeric_debug_run (Not Run)
675 - phobos_std_numeric_release_build (Failed)
676 - phobos_std_numeric_release_run (Not Run)
680 - phobos_std_functional_release_run (Failed)
706 - phobos_std_regexp_debug_run (Failed)
708 - phobos_std_regexp_release_run (Failed)
746 - phobos_std_complex_debug_run (Failed)
748 - phobos_std_complex_release_run (Failed)
758 - phobos_std_getopt_debug_run (Failed)
760 - phobos_std_getopt_release_run (Failed)
766 - phobos_std_conv_debug_run (Failed)
768 - phobos_std_conv_release_run (Failed)
770 - phobos_std_outbuffer_debug_run (Failed)
772 - phobos_std_outbuffer_release_run (Failed)
786 - phobos_std_csv_debug_run (Failed)
788 - phobos_std_csv_release_run (Failed)
790 - phobos_std_mathspecial_debug_run (Failed)
792 - phobos_std_mathspecial_release_run (Failed)
793 - phobos_std_typecons_debug_build (OTHER_FAULT)
794 - phobos_std_typecons_debug_run (Not Run)
795 - phobos_std_typecons_release_build (OTHER_FAULT)
796 - phobos_std_typecons_release_run (Not Run)
814 - phobos_std_datetime_debug_run (SEGFAULT)
816 - phobos_std_datetime_release_run (SEGFAULT)
834 - phobos_std_format_debug_run (Failed)
836 - phobos_std_format_release_run (Failed)
838 - phobos_std_random_debug_run (Failed)
840 - phobos_std_random_release_run (Failed)
849 - phobos_std_math_debug_build (Failed)
850 - phobos_std_math_debug_run (Not Run)
851 - phobos_std_math_release_build (Failed)
852 - phobos_std_math_release_run (Not Run)
868 - phobos_std_range_release_run (Failed)
902 - phobos_std_net_isemail_debug_run (SEGFAULT)
904 - phobos_std_net_isemail_release_run (SEGFAULT)
921 - phobos_std_internal_math_errorfunction_debug_build (Failed)
922 - phobos_std_internal_math_errorfunction_debug_run (Not Run)
923 - phobos_std_internal_math_errorfunction_release_build (Failed)
924 - phobos_std_internal_math_errorfunction_release_run (Not Run)
929 - phobos_std_internal_math_gammafunction_debug_build (Failed)
930 - phobos_std_internal_math_gammafunction_debug_run (Not Run)
931 - phobos_std_internal_math_gammafunction_release_build (Failed)
932 - phobos_std_internal_math_gammafunction_release_run (Not Run)
1034 - dmd-testsuite_debug (Failed)
1036 - dmd-testsuite_release (Failed)
Errors while running CTest

> Would it make sense to "XFAIL" them (i.e. report a GitHub issue and
> selectively disable them) so a CI system could actually provide a binary
> regression indicator?

Can this be done on an architecture base? At least the math precision
failures are really annoying and hard to fix.

The reasons for the failures are still mixed. core.thread fails because
of the not-yet-fixed fiber support. core.math is simply math precision.
std.bitmanip looks like an optimizer bug. std.regex has still an endian
issue. And so on...

Even with these failure a CI server can help. The std.typecons failure
is triggered because of the assert I included in the memcpy function...
(for sure, something other is going wrong here)

Kai

[Kai Nacke]

Hi Alex!

With the merge of 2.063 and the release of LLVM 3.3 I think it is a good
time to add Linux/PPC64 to the set of CI builds. I would try to fix or
disable the failing tests in order to get a clean build.

A precondition is that the build environment has a LLVM 3.3 package. The
code generator of earlier LLVM versions is not able to compile the IR
generated by LDC. Would this be possible?

Thanks for your support.

Regards
Kai

On 25.05.2013 07:52, Alex R�nne Petersen wrote:

> Hey folks,

>
> What's the state of LDC on these two architectures? Is it anywhere
> near a point where we should consider adding them to the ci.lycus.org
> LDC job?
>

> Regards,
> Alex
>