Re: [gentoo-python] [PATCH] Introduce python_setup() as a common name for Python setup function.

[Mike Gilbert]

On Sun, Oct 20, 2013 at 11:47 AM, Michał Górny <mgorny@g.o> wrote:
> Dnia 2013-10-20, o godz. 11:33:45
> Mike Gilbert <floppym@g.o> napisał(a):
>
>> On Sun, Oct 20, 2013 at 11:12 AM, Michał Górny <mgorny@g.o> wrote:
>> > Dnia 2013-10-20, o godz. 10:53:34
>> > Mike Gilbert <floppym@g.o> napisał(a):
>> >
>> >> On Sun, Oct 20, 2013 at 3:36 AM, Michał Górny <mgorny@g.o> wrote:
>> >> > As described in the ml, the new function does:
>> >> >
>> >> > * in python-any-r1, choose and set up the best installed impl,
>> >> >
>> >> > * in python-single-r1, read PYTHON_SINGLE_TARGET and set it up,
>> >> >
>> >> > * in python-r1, call python_export_best + python_wrapper_setup.
>> >> >
>> >> > In other words, it's a common way of getting Python build environment
>> >> > set up in the ebuild scope.
>> >>
>> >> Invoking python_setup from pkg_setup in any eclass seems sub-optimal;
>> >> especially so in python-any-r1. However, I realize there isn't a great
>> >> place to put it.
>> >
>> > I don't understand. Could you rephrase?
>> >
>>
>> The python stuff is generally only used in "src" phases, so pkg_setup
>> is the wrong place for it.. As well, it would be better to set up the
>> python wrappers and environment in a phase where the sandbox and
>> userpriv are active to prevent permissions problems like we had with
>> Paludis recently (there's a bug, but I'm not going to look for it).
>
> I can understand you but can't really think of a good way of doing
> this. The only other reasonable phase for this is src_prepare(),
> and overriding it will result in hell breaking loose...
>

Right. I just wanted to go "on the record" as saying that this is not
optimal, but there is probably no better alternative. Carry on. ^_^