Supercollider fails to build

[Felipe Sateler]

On Sat, Jul 21, 2012 at 5:07 PM, Dan S <danstowell+debmm at gmail.com> wrote:
> 2012/7/21 Felipe Sateler <fsateler at debian.org>:
>> On Sat, Jul 21, 2012 at 4:18 AM, Dan S <danstowell+debmm at gmail.com> wrote:
>>> 2012/7/21 Felipe Sateler <fsateler at debian.org>:
>>>> On Thu, Jul 12, 2012 at 5:50 PM, Dan S <danstowell+debmm at gmail.com> wrote:
>>>>>
>>>>> 2012/7/10 Felipe Sateler <fsateler at debian.org>:
>>>>> > OK, so we have more build failures, but this time different on
>>>>> > different architectures:
>>>>> >
>>>>> > 1. Arm*: qreal is not double but float. Patch is in the repository.
>>>>> > 2. PowerPc64: Undefined variable order. It seems that there is a bogus
>>>>> > call in line 255 of
>>>>> > external_libraries/boost_lockfree/boost/atomic/detail/gcc-ppc.hpp, but
>>>>> > I'm not sure.
>>>>> > 3. ia64: missing ifdef case in
>>>>> > external_libraries/nova-simd/benchmarks/perf_counter.hpp (starting at
>>>>> > line 466). I don't know how to fix this.
>>>>> > 4. all big-endian archs: Typo in
>>>>> > external_libraries/boost_endian/boost/integer/endian.hpp:285. It seems
>>>>> > like an extra ) crept in in the least-tested #ifdef path. Patch in the
>>>>> > repository too
>>>>>
>>>>> I believe issues 2 and 3 are now fixed, with the small patches I added in
>>>>> git.
>>>>> The upstream fixes are different (importing a more recent version of
>>>>> boost.atomic) but the patches would be big and complex.
>>>>
>>>> Uploaded.
>>>
>>> Thanks
>>
>> ... and it fails again :(
>>
>> https://buildd.debian.org/status/package.php?p=supercollider
>
> OK. I'll have to pass the powerpc asm issue onwards.

Googling tells me that apparently there is some bug in gcc causes it
to generate too big asm sections, which is why the assembler chokes.
Applying

--- a/external_libraries/boost_lockfree/boost/atomic/detail/gcc-ppc.hpp
+++ b/external_libraries/boost_lockfree/boost/atomic/detail/gcc-ppc.hpp
@@ -124,7 +124,7 @@ public:
 		T &expected,
 		T desired,
 		memory_order success_order,
-		memory_order failure_order) volatile
+		memory_order failure_order) volatile __attribute__((noinline))

Seems to fix the issue. SC builds on the porterbox with that change.

> For the others:
>
> The ia64 issue might be the same as this in powertop:
> <http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=570662> and so I
> propose untested patch [1].

Indeed, supercollider builds with this patch applied.

> For the arm*s, I'd suggest adding [2] to your existing
> "arm-qreal-is-float.patch". There may be more "double"-->"qreal"
> changes needed.

I'm testing the patch. Arm is slow to build, though.

-- 

Saludos,
Felipe Sateler