Re: [gentoo-dev] Prevent to need to change all keywords at the same time

[Rich Freeman]

On Thu, Jul 17, 2014 at 10:09 AM, Ian Stakenvicius <axs@g.o> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
> On 17/07/14 08:28 AM, Pacho Ramos wrote:
>> I recently noticed this:
>> https://bugs.gentoo.org/show_bug.cgi?id=502836
>>
>> imlib2 ebuild can only be stabilized in one round for all arches
>> as KEYWORDS are set in eclass depending on E_STATE="release". That
>> has an important drawback as forces all arches to be done at the
>> same time and, since some are much slower than others, forces all
>> to wait for them. And, as that can depend on even more
>> stabilizations (like it's the case) all that bugs blocking the
>> stabilization need to also be done for *all* arches before.
>>
>> I am not sure if any policy exists for this, but I would forbid to
>> make this due this issue. I would instead move to use KEYWORDS en
>> ebuild as done usually.
>>
>> What do you think?
>>
>>
>
> Unless there is some sort of need to synchronize stable keywords
> across multiple packages in an identical fashion, that is -so
> important- it can't be left to AT's and maintainers to ensure the
> stablereq bugs are filed across them all at once on their own, I don't
> see a reason for setting KEYWORDS in an eclass.
>
> So, +1 for moving KEYWORDS to ebuilds.  I'm not sure if "forbidding"
> is necessary, as I think strongly discouraging all overly-complicated
> solutions like this one would suffice. (and yes i know the irony of
> this statement given that I'm in the gx86-multilib project :)

+1000

I think that sticking KEYWORDS in an eclass is something that should
probably never happen.  That isn't to say that it can't happen if
there is some really important reason, but I can see it creating a
number of issues.

If for some reason we have a collection of packages that need to be
synchronized WITHIN an arch I think we should think about ways to make
this easier, but this probably isn't the way to do it.

Rich