Add Python 3.7 testing to travis - #7607

Hi there @pllim 👋 - thanks for the pull request! I'm just a friendly 🤖 that checks for issues related to the changelog and making sure that this pull request is milestoned and labeled correctly. This is mainly intended for the maintainers, so if you are not a maintainer you can ignore this, and a maintainer will let you know if any action is required on your part 😃.

Everything looks good from my point of view! 👍

If there are any issues with this message, please report them here.

So it seems that people would like to use astropy with 3.7, we may want to warm this PR up and see whether everything goes well.

When it works, we'll have to think about rearranging the actual CI matrix for master. Do we need to support Python 3.7 for 2.x series? Do we still need to test Python 3.5, and so forth?

• I think we need to support 3.7 on LTS, but if others disagree, honestly I'm happy to go either way.
• We need to keep 3.5, the roadmap in APE10 switches to 3.6+ in v4.0 (as we won't do a 3.3)

There is also a new numpy out, but I would rather deal with that martix reshuffle in another PR.

Also, another related question for 3.7 testing whether we want to swich to it on circleCI and/or appevyor, too, or leave those as is?

@pllim - I've rebased this to make sure the tests pick up the changes from #7617

you just hit the same travis silly I just did in sunpy 🤦‍♀️ ...

@pllim - to answer the other question about pip installing for python2: in #7416 I've originally used the 3.7-dev environment on travis, changing it to the nonexistent 3.7 meant that the pip version it picked up was some default and apparently a 2.7 version of it...

OK, so now all the debug commits are gone. The only outstanding issue is whether it should be the clang job that's the 3.7, or another one further down the matrix.

I'm sorry that this ended up as a PR hijack.

I'll manually add the python3.7 tests to the travis configs in 3.0.x and LTS given that this version is significantly diverted already.

No problem, @bsipocz ! Oh, all green. 🎉

Personally, I wouldn't bother making LTS compatible with Python 3.7. If user is so keen to use Python 3.7, they can also upgrade to Astropy 3.x. Astropy 2.x is really only for those stuck in Python 2 for whatever reason. But I guess the final say should come from @astropy/coordinators ?

Well, it looks to be already compatible, all backport were smooth.

Yay! Agreed that it might be useful to look at the whole test suite to see if we're not duplicating too much, but would seem out of scope here.

@mhvk - The new numpy is not available on conda yet, I would rather wait for it than do the reshuffle twice.
Whether we wait with this PR for that or merge as is, I don't mind.

My comment was meant to suggest I was very happy with merging as is!