Re: CVS commit: pkgsrc/lang/gcc12

[Brook Milligan <brook%nmsu.edu@localhost>]

> On Dec 20, 2024, at 07:20, Jason Bacon <jtocino%gmx.com@localhost> wrote:
> 
> On 12/20/24 03:06, Jonathan Perkin wrote:
>> * On 2024-12-19 at 23:11 GMT, Brook Milligan wrote:
>> 
>>> 
>>>> On Dec 19, 2024, at 13:01, maya%netbsd.org@localhost <maya%NetBSD.org@localhost> wrote:
>>>> 
>>>> There are two things changing libgcc/config.host in lang/gcc12:
>>>> patches/patch-libgcc_config.host, but also gcc-12.4.0.diff, a
>>>> downloaded patch file.
>> 
>>> Thanks for noticing this.  The problem is that the downloaded patch
>>> only applies to Darwin-*-aarch64.  I believe my patches are fine
>>> everywhere else, as they are built against the unpatched distfile.
>>> 
>>> I think this means platform-specific patches, but I have no idea how
>>> we handle that.
>> 
>> Right, this is a somewhat tricky situation, where the problem is
>> exacerbated by the fact that gcc12 is the selected gfortran version on
>> Darwin, so this is causing a lot of fallout.
>> 
>> I think I'm going to back this out.  We're just way too deep into the
>> freeze for this serious a regression, and even with a fix we would still
>> be a few days away from any bulk builds to validate the patch.
>> 
>> It can be sorted after the freeze, hopefully along with an update to the
>> gfortran selection that doesn't require the huge aarch64 patchset.
>> 
> 
> For after the branch:
> 
> Do we have an existing tool to convert a diff like the one used here to
> pkgsrc patches?  If not, I think this should not be hard to create.
> Then we won't have a risk of collisions between two different patching
> methods.

Likewise, it seems there should be a method of using patches selectively.  That would have given us a simple solution.

Cheers,
Brook