Version of matplotlib

[adamcpovey]

Is there a reason matplotlib is required to be version < 1.9? Version 2.0 came out in Jan 2017...

It's causing me problems when installing cis as anaconda installs matplotlib 2.2.0 as it's dependency isn't version restricted but easy_install rejects the environment.

[ajdawson]

The conda-forge build of iris (definitely the easiest way to install iris by the way) allows any version of matplotlib.

I suspect this had something to do with making our test suite pass. You can safely ignore this version restriction for your own use.

[adamcpovey]

I can ignore it but I don't know how to get Travis to do so (https://travis-ci.org/cedadev/cis/jobs/368711354).

[ajdawson]

OK understood. That is pretty annoying since you've already got iris 2.0.0 installed by the time setuptools is complaining...

Since your Travis build has already installed all your dependencies you could try doing the final install of your package (cis) without dependencies which may avoid the issue, I'm kind of guessing though.

The v2.0.0 of iris was the first release published on PyPI so we're still working out some issues, and I will confess to not fully understanding this one. Perhaps someone else can say something more informative @bjlittle @pelson @QuLogic ?

[pelson]

$ python setup.py --quiet install
zip_safe flag not set; analyzing archive contents...
cis.cis_main: module references __file__
cis.plotting.plot: module references __file__
cis.test.integration_test_data: module references __file__
cis.test.plot_tests.idiff: module references __file__
cis.test.plot_tests.test_plotting: module references __file__
error: matplotlib 2.2.2 is installed but matplotlib<1.9 is required by set(['scitools-iris'])

I believe there is a --no-deps flag with setuptools: python setup.py --quiet install --no-deps should be what is needed in this case.

In addition, I can recommend adding zip_safe=False in your setup.py to prevent setuptools having to inspect your code to determine whether it can install cis as a tarball.

---

As for whether Iris supports matplotlib 2.0, we don't know. Our dependencies (including cartopy), do, but we don't currently have travis testing to confirm it or not. This is high on my agenda, as it is certainly causing headaches in a lot of areas. Watch this space (@bjlittle has been talking about getting this done ASAP).

Hope that helps,

[bjlittle]

Now that #2989 has been merged, our travis-ci testing is fully functional and we've got coverage for our graphical tests again.

We're now in the position to unpin matplotlib for iris and easily see the impact, so I'm going to start working on that now 🎉

@pelson Did we want to start a milestone for 2.1 and include this issue as part of that deliverable?

[pelson]

Did we want to start a milestone for 2.1 and include this issue as part of that deliverable?

Nope (I feel uncomfortable holding up releases for work that isn't yet done). But if the work is done and merged, I'm happy to cut a release. 😄

[bjlittle]

Yup, sounds like a healthy approach to me, thanks.

[pelson]

Fixed in v2.1