Relax numpy version requirements

[adrianmoisey]

Just want to see if tests pass with this

[skvark]

I agree that current implementation is not very good. See my comment in issue #29 how this should be implemented.

[skvark]

FYI the Travis failure related to test dependencies on the Ubuntu container has been fixed in f5b3562.

[adrianmoisey]

Thanks. I've cherry picked it across.
Once I get the CI builds passing, I'll cleanup the commit history,

[skvark]

Note the numpy requirement in requirements.txt installs the latest numpy and it overrides the travis.yml numpy version. The requirement should be removed. requirements.txt is used in these lines and it should be safe to remove these calls completely:

• https://github.com/skvark/opencv-python/blob/master/travis/build-wheels.sh#L14
• https://github.com/skvark/opencv-python/blob/master/travis/build-wheels-osx.sh#L11

For Appveyor I suggest to define a variable which contains the exact numpy version requirement for each build in the matrix and add a new install line for it, e.g. "%PYTHON%/python.exe" -m pip install numpy==%NP_VERSION% after this line https://github.com/skvark/opencv-python/blob/master/appveyor.yml#L51

[adrianmoisey]

Ah right, thanks!
I hadn't looked much at the build scripts.

[adrianmoisey]

I'm finished here. What do you think?
If you're happy with this, I'll squash the commits and fix the git history, as I made a few debug commits.

[skvark]

Looks good. Couple of small additions:

• Appveyor.yml: NP_VERSION can be 1.11.1 for all other Python versions except for 3.6 it should be 1.11.3.
• .travis.yml: Add also TEST_DEPENDS=numpy==1.11.3 for py36 builds in the matrix to make the test dependency consistent with the build dependency.

[skvark]

Do you want to clean up the commits before the merge?

[adrianmoisey]

Yeah, let me do that.
Should I do that in a new branch?
Or should I destroy the history here?

[skvark]

Hmm, I think it's easiest if I do the squashing directly from Github UI unless you want to do it manually with git.

[adrianmoisey]

That works for me 👍

[skvark]

Thanks!

[adrianmoisey]

Thanks for all the help!