Re: [gentoo-alt] [PREFIX] the future of Prefix

[Michael Haubenwallner]

On Wed, 2008-10-08 at 10:50 +0200, Fabian Groffen wrote:

> Well, I personally think we need to put the EAPI problem[1] in there,
> the extra variables[2] and functions[3] for Portage.  Later concern are
> our keywords[4] and our ground braking core ebuild changes[5].  If the
> consensus is to accept our EAPI thing, devs have to keep ebuilds that
> are EAPI=?prefix.  That's the deal.  One of the problems is that we need
> to touch quite a lot of ebuilds, as well as the issue from[1].
>
> I think we can start with merging trivial non-intrusive patches for
> compilation stuff to packages to gentoo-x86 already.  We should probably
> at the same time push them upstream and annotate the patches where not
> done with that upstream tracker/link.
>
>
> [1] prefix is orthogonal to all other existing EAPIs, e.g. it can be
> overlayed other EAPIs.  Problem here is do we do EAPI="prefix 1" or
> EAPI="prefix-1", and what do we do with base-system ebuilds?  We need
> them, but they need to remain EAPI=0 (EAPI absent) as well for obvious
> reasons.  What's best to do here?

<thoughts>
I'm unsure if stating the prefix-awareness in EAPI is right at all, as
it is orthogonal and works with any EAPI.
One could think of leaving this indicator to the KEYWORDS - and call
prefix an "arch" somehow: When an ebuild becomes keyworded XXX-YYYY,
then it is prefix-aware.
This implies that we keep the "XXX-linux" keyword[4.1].
What to do then with current non-prefix XXX-fbsd keywords - don't know
anything about the activity of that archs. Eventually they could do the
same as what you did with XXX-macos.
Or we rename our keywords to "x86+linux", where "+" stands for "prefix"?
</thoughts>

> [2] EPREFIX, ED, EROOT, can we agree on their name and definition?
>
> [3] eprefixify, often used in prefix part of EAPI

ok.

> [4] we extended to full XXX-YYYY for clarity, probably won't get the
> -YYYY part if "-linux", but we also added x64- for XXX-, which will
> introduce a big war (also sort of whatever, the keyword is the most
> important).  Last point, the introduction of all of our keywords and
> profiles (quite some) and maybe ekeyword's sorting order.

[4.1] I'm afraid we should not share the same keyword for both "Gentoo
Linux" and "Gentoo Prefix on any Linux" within the same tree, although
it's just a feeling ATM.

> [5] some base-system ebuilds and eclasses need non-trivial changes for
> our arches or Prefix (think of GCC ebuild/eclasses), need agreement here
> from their maintainers

ok.

/haubi/
-- 
Michael Haubenwallner
Gentoo on a different level