Re: [HelenOS-devel] Nightly builds

[Jakub Jermář]

On 05/05/2017 10:59 AM, Vojtech Horky wrote:
> Python is broken after the merge of the VFS improvements. When I was
> looking at it, it wasn't clear to me why it was failing as it was.
> If anyone knows any better, I'd be very grateful.
> 
> I remember we discussed that on IRC. The reason seemed obvious
> (undefined functions when linking) but I have no idea why it fails for
> these functions now as they have not changed at all from Python's point
> of view.

Well, it was not clear - at least to me - why these are suddenly
undefined and only for Python. Our native code has no issue with these
symbols.

> Also pcc seems to be broken more than necessary.
> 
> AFAIK PCC is broken for quite a long time but we have (more or less
> working) GCC and thus no one cares that much about PCC.

So we should either fix it or remove it altogether. There is no point in
keeping it defunct in the coastline.

> But when building does not make sense? I think we should aim for having
> the whole matrix in the green. And the purpose of the CI build is to
> know the overall status.

For example when PCC does not support it. Some of its source code
components seem to be unable to build when one of mach_i386, mach_ppc32
(?), mach_amd64 and mach_mips32 (?) is not defined (and I don't think it
supports cross-compilation or at least we do not build it that way):

pcc-1.0.0/os/helenos/ccconfig.h:
#if defined(mach_ia32)

#define CPPMDADD { "-D__i386__", NULL, }

#define DYNLINKER { NULL }

#elif defined(mach_ppc32)

#define CPPMDADD { "-D__ppc__", NULL, }

#define DYNLINKER { NULL }

#elif defined(mach_amd64)

#define CPPMDADD { "-D__x86_64__", NULL, }

#define DYNLINKER { NULL }

#elif defined(mach_mips32)

#define CPPMDADD { "-D__mips__", NULL, }

#define DYNLINKER { NULL }

#else

#error defines for arch missing

#endif

Jakub

___
HelenOS-devel mailing list
HelenOS-devel@lists.modry.cz
http://lists.modry.cz/listinfo/helenos-devel

Re: [HelenOS-devel] Nightly builds

[Vojtech Horky]

Hello, Jakub.

2017-05-04 18:10 GMT+02:00 Jakub Jermář :

> Hi Vojta,
>
> On 05/04/2017 08:41 AM, Vojtech Horky wrote:
>
>> I have updated the scripts at ci.helenos.org 
>> with the new Pythonic version. We now offer downloadable images of HelenOS
>> for all architectures there.
>>
>> The page also provides
>> - tarballs with Coastline packages,
>> - logs from the compilation,
>> - results of automated tests,
>> - and special images (e.g. ISO for ia32 with GCC).
>>
>> The rebuild is done daily but only when new revision was pushed either to
>> mainline or coastline.
>>
>
> This is a very good news!
>
> - overview matrix (except for Python, it looks healthy now),
>>
>
> Python is broken after the merge of the VFS improvements. When I was
> looking at it, it wasn't clear to me why it was failing as it was. If
> anyone knows any better, I'd be very grateful.
>
I remember we discussed that on IRC. The reason seemed obvious (undefined
functions when linking) but I have no idea why it fails for these functions
now as they have not changed at all from Python's point of view.


Also pcc seems to be broken more than necessary.
>
AFAIK PCC is broken for quite a long time but we have (more or less
working) GCC and thus no one cares that much about PCC.



> Perhaps the ci should have a means to disable CI builds for certain
> architectures. Or the harbours should not attempt to build when it does not
> make sense.
>
Currently, it works the other way around - you can select to build for only
specific architectures/harbours.

But when building does not make sense? I think we should aim for having the
whole matrix in the green. And the purpose of the CI build is to know the
overall status.

Cheers,
- Vojta
___
HelenOS-devel mailing list
HelenOS-devel@lists.modry.cz
http://lists.modry.cz/listinfo/helenos-devel