Re: [gentoo-dev] [RFC] EAPI={3,4} offset-prefix semantics mandatory?

[Jeremy Olexa <darkside@gentoo.org>]

On Tue, 15 Dec 2009 19:59:44 +0100, Fabian Groffen <grobian@gentoo.org>
wrote:
> With the current route where EAPI=3 will simply be EAPI=2 +
> offset-prefix support, and EAPI=4 will be EAPI=3 + some other stuff, the
> following question arose:
> 
>   Should an ebuild using an EAPI that has offset-prefix support make the
>   use of that support mandatory or optional?

As a Gentoo Linux developer, I certainly wouldn't want to say that my
EAPI-3+ ebuilds are guaranteed to work on Gentoo Prefix platforms without
testing(keywording) it.

As a Gentoo Prefix user, I certainly wouldn't want devs that don't know
how to test an ebuild to work on Gentoo Prefix to say that it works.
As a Gentoo Prefix developer, I certainly don't want to fix bugs by people
that don't know how test on (or have access to) Gentoo Prefix.

In the end, every Gentoo Prefix arch needs a specific keyword anyway.
Unless that changes, I can't say that I see many benefits to making
offset-prefix support mandatory in EAPI-3. OTOH, if it is mandatory and I
screw it up for Gentoo Prefix platforms, no one will know until it gets
keyworded (and hence tested).

-Jeremy


> In other words, one can perfectly fine write an ebuild EAPI=3 that will
> not work in an offset-prefix install, due to improper absence of
EPREFIX,
> ED and EROOT.  Should we allow this formally, or not?
> 
> Why is this a problem?  Simply because it can be done, but more because
> EAPI=4 will introduce features a developer would like to use/rely on,
> while she/he does not want, or is not able to write the ebuild in a
> Prefix conforming way.
> 
> The pros for forcing ebuilds to be offset-prefix aware are:
> - an ebuild having EAPI >= 3 (as it looks now) is supposed to work
>   for Prefix users
> - hence also obviously is (supposed to be) checked for Prefix
> - repoman might be able to check for obvious mistakes regarding
>   offset-prefix installations
> 
> The cons:
> - all developers need to be aware of how Prefix works, and be able to
>   write ebuilds for it (I can post all the answers to the Prefix quiz)
> - basically requires a Prefix to be setup to test
> - it will stop developers to some degree to use higher EAPIs in the
>   worst case
> 
> The pros for allowing ebuilds that have an offset-prefix aware EAPI to
> ignore the offset-prefix are:
> - easy drop-in replacement for devs, basically the contra of all the
>   cons of the previous approach.
> 
> The cons:
> - not immediately clear which ebuild is offset-prefix aware (could look
>   at Prefix keywords)
> - needs proper rules; an ebuild that has offset-prefix support may not
>   have this support removed again (breaks Prefix users, how to enforce?)
> - ebuilds may get offset-prefix support at a later stage, which may not
>   entirely be understood/noticed by (their maintaining) devs
> 
> Please voice your opinion and share your insights, if any.