Building math/py-pandas against py3k (WAS: ports/172528: [maintainer update] [patch] Update math/py-pandas to 0.9.0)

[Ruslan Mahmatkhanov]

Hi John,

John W. O'Brien wrote on 17.10.2012 18:17:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 10/12/2012 08:17 AM, John W. O'Brien wrote:
>> On 10/11/2012 03:55 PM, Ruslan Mahmatkhanov wrote:
>>> I delayed the response becaused I decide to check the build with
>>>   python32. It's fore sure that py-pandas pkg-plist should be
>>> patched to make packaging correct, but unfortunately the build is
>>> stuck on some of it's dependencies (py-numpy, maybe) so I wasn't
>>> able to test it.
>>
>> I appreciate your attempt. I'll see what I can do to probe that
>> issue this weekend. It will be the first time I try building
>> everything against python 3, but I've got a fast machine and a load
>> of disk space.
>
> Ruslan (and freebsd-python@),
>
> It looks to me like there are three, maybe four other ports that need
> attention before we can successfully build and test math/py-pandas
> against py3k.
>
> devel/py-dateutil is at 1.5 which is needed for Python < 3, but Python
>> = 3 requires dateutil 2.0.
>
> devel/py-pytz will build, install, and package cleanly, but...
>
[...]
>
> devel/py-nose doesn't package cleanly because of plist problems (looks
> like mainly lingering *.pyo and *.pyc entries). I also tried 1.2.1
> using the patch from ports/172161, but the plist is still not
> py3k-friendly.
>
> math/py-numpy also has plist problems, and fails 14 of 3556 unit tests
> (results available on request).
>
> In summary, devel/py-dateutil is the big blocker. math/py-pandas,
> devel/py-nose, and math/py-numpy need plist patches. devel/py-pytz and
> math/py-numpy have runtime bugs that need further analysis.
>
> As time permits, I will learn the plist magic that makes ports
> py3k-friendly, and secondarily poke at the dateutils upgrade. I'm
> probably out of my depth, but I'm happy to take direction and correction.
>
> Cheers,
> John

Thank you for the investigation. Yes, sadly dateutil upstream developing 
two separate branches for 2.x/3.x, so it will require some bsd.python.mk 
tweaks (like setuptools vs distribute) so the correct dependency picked 
according to python version used.

pytz needs an bug report on launchpad.

nose and numpy issues are solvable, and I will look at it in near time.

-- 
Regards,
Ruslan

Tinderboxing kills... the drives.