bug#25879: [PATCH] gnu: Add LLVM and CLANG 3.9.1.

[Roel Janssen]

Ricardo Wurmus writes:

> Roel Janssen <address@hidden> writes:
>
>> Ricardo Wurmus writes:
>>
>>> Roel Janssen <address@hidden> writes:
>>>
>>>> Ricardo Wurmus writes:
>>>>
>>>>> Pjotr Prins <address@hidden> writes:
>>>>>
>>>>>> On Tue, Mar 07, 2017 at 09:06:28PM +0100, Ludovic Courtès wrote:
>>>>>>> > Well, actually, this is just the latest release, so maybe I should
>>>>>>> > update the 3.8.1 recipe to3.9.1 instead.  WDYT?
>>>>>>>
>>>>>>> If the other users of LLVM and Clang (as per ‘guix refresh -l llvm’) can
>>>>>>> cope with it, upgrading sounds better indeed.  Could you check if that
>>>>>>> is the case?
>>>>>>
>>>>>> With LLVM it is probably a good idea to keep the major versions as
>>>>>> packages tend to lag after latest. Many compiler writers are a bit
>>>>>> behind and sometimes people want to use older compilers (like with
>>>>>> Julia).
>>>>>
>>>>> I agree.
>>>>>
>>>>> @Roel: I see that this patch hasn’t been pushed yet.  Is there anything
>>>>> missing or was it just forgotten?
>>>>
>>>> I think the idea was to upgrade, instead of have this newer version next
>>>> to the current version.  The upgrade involves a lot of rebuilding, and I
>>>> am stuck at compiling 'dub' with 3.9.1.
>>>>
>>>> If we can instead apply this patch as (having both 3.8.1 and 3.9.1), we
>>>> can push it, and after that add the darktable patch as well.
>>>
>>> I think it’s fine to have multiple versions of LLVM + Clang around,
>>> especially considering that in my experience many dependent projects
>>> won’t build with later versions without adjustments.  (RStudio, for
>>> example, still insists on the oldest version of Clang that we offer, and
>>> it crashes with later versions.)
>>>
>>> It would be good to keep an eye on this, though, to make sure that we
>>> don’t provide outdated versions that have no users and no maintainer.
>>
>> So, is it OK to push the patch as-is then?
>
> Yes, please!  :)

Pushed in 584da12dc71da745edb13bf748e832b77a0193d7.

> If you can, it would be good to investigate if current users of 3.8
> could be built with 3.9, but this doesn’t have to block this patch in my
> opinion.
>
> If in fact all users of 3.8 can be built with 3.9 without problems you
> can make the change in a follow-up commit.

I am sure that at least 'c-reduce' does not build with 3.9.1.  So I
think we'll need to take this on a per-program basis.  I'll see which
ones can be built with 3.9.1.

Thanks!

Kind regards,
Roel Janssen