Mailing List Archive: Road towards EAPI 3 main tree approval

As many of you know EAPI 3 has been waiting for Portage to implement it
for many months now and I asked zmedico for his estimate on whether it
will be done this year:

19:50 < Betelgeuse> zmedico: Let's put it this way. How likely in
percentages is EAPI 2 this year?
19:50 < Betelgeuse> s/2/3/
19:51 < zmedico> I'd give it 80% chance I guess

As far as I understand paludis already has the features implemented so
maybe we could give EAPI 3 some field testing in some overlays to see
how it works in practice from ebuild writer point of view. Paludis does
not recognize it atm but probably easy to get a new revision/version out:

20:01 < zmedico> Betelgeuse: I'd guess they could test it as
EAPI=paludis-3_pre or something like that
20:01 < dleverton> Betelgeuse: it doesn't install the config file, just
uses it for tests, so that's correct

So what's the general sentiment for giving this some spin in overlays?
Maybe having some usage would build some social pressure on zmedico :)

Regards,
Petteri

On Sun, 30 Aug 2009 23:03:42 +0300
Petteri RÃ¤ty <betelgeuse@gentoo.org> wrote:
> As far as I understand paludis already has the features implemented so
> maybe we could give EAPI 3 some field testing in some overlays to see
> how it works in practice from ebuild writer point of view.

The main issue with that is eclasses. Quite a few eclasses have
hard-coded lists of 'case ${EAPI:-0} in' lists. These would need to be
updated. So if you go this route:

> 20:01 < zmedico> Betelgeuse: I'd guess they could test it as
> EAPI=paludis-3_pre or something like that

then you'll have to clutter up the main tree with paludis-3_pre. And if
instead you call it '3', it means we have to guarantee we're not going
to make any last minute changes.

Also, as I recall there're two outstanding issues with EAPI 3 that
would need to be addressed by the Council.

First, there's the question of package.mask etc as directories. Some
people were after either doing a quicky EAPI 2.1 or reopening EAPI 3 to
allow that for the 10.0 profiles. I don't know whether that's still
considered worth doing or whether it's an EAPI 4 thing.

Second, there's the issue with 'nonfatal' and 'die' [1]. EAPI 3 as
originally worded made 'die' ignore 'nonfatal' (so that people wouldn't
have to modify all their eclass functions to take into account that
suddenly callers could override their dies), but some crappy wording in
the original spec (which has been addressed) meant that some people
didn't realise that, and would prefer it if we did that differently.

> Paludis does not recognize it atm but probably easy to get a new
> revision/version out:

Yup, can do that easily enough if there's interest once the above has
been addressed.

[1]: http://archives.gentoo.org/gentoo-dev/msg_358b12a494173ab82f3e7d1b2b6b5bf9.xml

--
Ciaran McCreesh

Mailing List Archive

Attached Files:

Attached Files: