Re: [gentoo-python] python-distutils-ng -- the future?

[Ben de Groot <yngwin@gentoo.org>]

On 9 September 2012 06:00, Michał Górny <mgorny@gentoo.org> wrote:
> I've just looked through python-distutils-ng again and I think it
> deserves at least a bit of changes, if not complete redesign.

Sounds to me like you want to write a new eclass, not a few changes to
the existing one. Major changes to an eclass are a recipe for disaster
(just consider the messy history of the python and distutils
eclasses).

> It is currently used by 22 packages in Gentoo, and probably a few
> in the overlays. I'd honestly prefer to avoid further packages doing
> that before we decide on how to proceed. That's why the first patch I'm
> suggesting somehow marks is as work-in-progress and asks people not to
> use it yet.

Please don't. As it is, the python-distutils-ng is much more usable
than the overcomplicated python and distutils eclasses. I recommend
everyone to use it for new ebuilds.

> The second and third patches just fix one of the formal problems.
>
> The fourth patch completely removes PYTHON_OPTIONAL support. It is not
> used by any of the mentioned packages right now and that way I'd like to
> avoid it being used by anything else, before we redesign it.

Sounds good.

> The major issue I see with it is that it hardcodes the 'python' flag.
> That's a bit mistaken, I believe. Not that I see much use for
> an optional Python support in package using distutils, and if there is
> one, I believe developers will find it more comfortable to do:
>
>     use python && python-distutils-ng_src_compile
>
> than expect the eclass to do that check for them.

I agree here.

> We also should think about USE-dependencies more (I will think about it
> myself, trying to find some working solution better than one suggested
> by Arfrever [the <<[use]>>]). It will be hard thinking...
>
> And one thing certainly needing discussing is extracting some
> common-python code to a separate eclass, without the distutils part.
> To be honest, I'd rather move the whole 'optionality' of Python there.

Sounds good.

> And lastly, the name. Considering the amount of work that needs to be
> done, we might also decide to 'lastrite' the eclass and work on a new
> one, replacing 'ng' with something saner.

Yes, once your new eclass(es) is/are in place, we can migrate existing
ebuilds and lastrite the current eclass(es). Unfortunately, we don't
have a method for versioning eclasses in place, so we have to do it
the clunky way of introducing new eclasses with new names.

-- 
Cheers,

Ben | yngwin
Gentoo developer
Gentoo Qt project lead, Gentoo Wiki admin