BLD: Don't upgrade numpy with pip install -U patsy

[jseabold]

Untested...thoughts? This should help keep a pip upgrade of statsmodels from upgrading numpy if we add patsy to install_requires.

statsmodels/statsmodels#588
statsmodels/statsmodels#549

[jseabold]

Maybe moot for us now, but might still be worth a look.

[njsmith]

The situation with pip install -U is ridiculous. Nonetheless, IIUC this isn't causing any practical problems for you at the moment, right, because you don't and won't use install_requires=["patsy"]? If that's right then I'm just going to keep letting this sit until either (a) the pip folk get their act together, or (b) it starts causing practical problems.

They do at least document a (still ridiculous) workaround for the broken -U behaviour now: http://www.pip-installer.org/en/latest/cookbook.html#non-recursive-upgrades

[jseabold]

I was debating adding install_requires, hence the PR, but ultimately didn't, so no this won't affect us.

[njsmith]

Actually even if you did add install_requires, if you used the hack in this
PR then it would fix the issue for you regardless of what patsy did. If
statsmodels says its wants numpy==np.version and patsy says it wants any
old numpy, then pip will AND those and decline to upgrade numpy in any case.

On Tue, Mar 26, 2013 at 5:04 PM, Skipper Seabold
[email protected]:

I was debating adding install_requires, hence the PR, but ultimately
didn't, so no this won't affect us.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/16#issuecomment-15471487
.