[Distutils] setup.cfg new format proposal

[Sridhar Ratnakumar]

On Wed, 16 Sep 2009 09:14:32 -0700, P.J. Eby <pje at telecommunity.com> wrote:

> So if I resume, so far the uses cases are:
>  - the OS given by os.name and sys.platform  
> (linux/mac/windows/riscos/ce/ etc..)
> - the architecture, given by os.uname() (32/64 bits)
> - the python version, given sys.version_info
>  Do we have anything besides dependencies that change based on the  
> above?  If not, then we might be able to address this with the "extras"  
> syntax mechanism already present in install_requires, and we might able  
> to do it without even changing that syntax.

+1. This whole conditional semantics in setup.cfg (or a new setup.cfg.in)  
is unneeded if the requirements are small.

Just reproducing an related discussion from another thread:

On Tue, 15 Sep 2009 14:12:46 -0700, P.J. Eby <pje at telecommunity.com> wrote:

> At 01:46 PM 9/15/2009 -0700, Sridhar Ratnakumar wrote:
>> install_requires = ['lxml', 'multiprocessing[pyver<(2,6)]', 'argparse',
>> 'pywin32[platform.name=windows]']
>
> If you spelled that [python-lt-26] and [platform-windows], it'd even be  
> syntax compatible with today's setuptools; there'd just need to be a way  
> to specify default "extra" tags to be always installed.  (Technically,  
> '.', and '_' characters would be usable as well as '-'.)

I prefer an explicit 'pyver<(2,6)' and it may be better to do a minor  
release of setuptools addressing this forward compatibility (with '<',  
'(', etc..).

-srid