Re: [matplotlib-devel] building with python-2.3

[Andrew S. <str...@as...>]

Robert Kern wrote:
> Andrew Straw wrote:
>
>> 1) revert to the old way. The primary issues with this are a) 
>> "package_data" is supported as standard Python from 2.4 on, and the old 
>> way required carrying our own distutils command and b) we switched the 
>> data directory to have a nested structure, which required code changes 
>> and repository layout changes that would have to be undone.
>> 2) make our own distutils monkeypatch a la setuptools. Looking at 
>> setuptools/dist.py, this doesn't look trivial -- certainly beyond my 
>> free bandwidth capacity.
>>     
>
> Actually, it ought to be pretty trivial without setuptools (but compatible with
> setuptools, AFAICT). Here is a Cookbook recipe that ought to work:
>
>   http://wiki.python.org/moin/DistutilsInstallDataScattered
>
>   
That's exactly "the old way", referred to in point #1.
> IPython does something similar and possibly better.
>
>   http://ipython.scipy.org/svn/ipython/ipython/trunk/setupext/install_data_ext.py
>   
 From a quick look at the code, it's hard to determine whether this new 
distutils command (install_data_ext) can handle installation to a nested 
directory structure. Can it? If it can, there's still the question of 
whether we want to continue rolling our own solution or simply using 
Python >= 2.4's standard "package_data".

AFAICT, the monkeypatching setuptools does for 2.3 to support 
"package_data" goes beyond adding a new distutils command. (I don't 
consider adding a distutils command to be monkeypatching -- that's just 
extending distutils is a pre-designed way.) Instead, setuptools actually 
replaces the distutils.dist.Distribution class with 
setuptools.dist.Distribution.