ENH: add correct versions of numpy installs

[cpcloud]

closes #4142.

should be able to finish up tomorrow, just need to put things on the server and
test

[cpcloud]

@y-p can u have a look-see?

some notes:

• for some reason the locale build won't override the lxml install so i'm uninstalling it so that the wheel gets used and that works
• i'm installing build deps for numpy which i don't think was being done before now: i couldn't get numpy to compile on the vagrant box unless i installed numpy's build deps, it doesn't add a significant amount of time to the build

anyway it seems to be working so lmk if u think i should change anything

here's a travis build:
https://travis-ci.org/cpcloud/pandas/builds/8825273

[ghost]

Very strange about the LOCALE build refusing lxml, there shouldn't be any difference at all
between it and the other 2.7 build.

Adding a JOB_TAG envar to each job in travis.yml and using that to select the requirements file
and url would make more sense IMO then what you opted for.

re numpy build deps, what packages specifically? I've had no problems building numpy
with the stock precise64.box + the packages that the original build script installed.

All in all, If it works reliably, it's good enough.

[cpcloud]

• numpy build-dep was a false alarm, rebuilding packages now...
• JOB_TAG stuff

[cpcloud]

just a note to self: distribute 0.6.35 doesn't work with py3

[ghost]

The amount of pain and suffering python packaging inflicts on kind, honest devs..

note that latest pip stable use 0.6.34 I believe, and 0.6.35 is the minimum version required
by pip beta (which Is why I named that specifically) , distribute itself has released many versions
since, although now it's being merged into setuptools. It's a longshot but you may try one of the
latest versions.

[ghost]

oh, they just released setuptools 0.8 final and pip 1.4rc3, perhaps
everything just works with that.

[cpcloud]

Would changing the versions of pip and using setuptools require a rebuild of packages? if not, great, i'll try it, if so i'm going to stick with what works which is pip installing whatever version of distribute it does for py3 and 0.6.35 for py2

[ghost]

It should absolutley not make a difference. Obviously, you install packages from pypi without
taking notice of what the packager had installed.
I would have thought that the default distribute for py3 would make pip yell, but if you have
a working setup, I'd stick with it too.

[cpcloud]

not sure what the lxml issue is and not worth tracking down for now...leaving the uninstall in

[cpcloud]

@y-p any last words before I merge (after passing)?

[ghost]

yes: thanks for taking care of this!

[cpcloud]

no problem! happy to grow a few gray hairs helping out pandas!

quick recap of changes:

• travis dependencies are now controlled by individual job specific requirements files available in pandas/ci
• py26 now now uses numpy 1.6.1
• py27 locale now uses numpy 1.6.1
• py27 uses numpy 1.7.1 and scikits timeseries (still full deps)
• py32 now uses numpy 1.6.2
• py33 now uses numpy 1.7.1 and runs with full dependencies

[cpcloud]

bombs away