Add support for Python 3.7

[lelit]

This enables building binary wheels for Python 3.7, now at its rc1 release.

It is included in the latest manylinux1 Docker image, see pypa/manylinux#196.

Accordingly to https://www.appveyor.com/docs/build-environment/#python it is not yet available on Appveyor.

[lelit]

It fails, because the dockcross/manylinux1 images do not carry py37 yet.

[joerick]

awesome work @lelit! any idea when the dockcross image will be published?

[lelit]

No, unfortunately they pushed an update just before the standard manylinux1 gained support for 3.7, and from https://hub.docker.com/r/dockcross/manylinux-x64/tags/ it's not clear how often that happens.

[YannickJadoul]

Nice! :-)
Is this urgent to include (given that AppVeyor still doesn't have it either), or could we wait 1 or 2 more weeks for the full release (cfr. https://www.python.org/dev/peps/pep-0537/)?

EDIT: Personally, I don't care about dockcross that much, to be fair. It's mainly meant to test the alternative docker image option, so we can just temporarily disable that test, as far as I'm concerned.

[hroncok]

Is this urgent to include (given that AppVeyor still doesn't have it either), or could we wait 1 or 2 more weeks for the full release

Please, release this. Getting 3.7 builds building at rc1 stage means we can find more 3.7 bugs before the actual release.

[YannickJadoul]

Please, release this.

Fair enough. Let's just temporarily disable that one test, then?

[lelit]

Fair enough. Let's just temporarily disable that one test, then?

Would be it any better if we just ignore/skip unavailable versions within a given image? In that way, there would be less pressure on cibuildwheel to follow whatever Docker image one is using (I mean for example the cases when "imagexyz dropped Python 3.4"...).

[YannickJadoul]

Myeah, but wouldn't you want to know by having that as an error and manually disabling that version for that build, instead of silently ignoring what could be a potential oversight?

[lelit]

Possibly yes... I was just thinking aloud (I do not use alternative manylinuxXX images myself): the question is about whether it should be always an hard error or if it could be downgraded (by default, or with an option) to a warning message instead.

[YannickJadoul]

Yes, no, sorry, didn't mean to shut that suggestion down completely, but just my 2 cents on that. Since CIBW_SKIP (and e.g. cp37-manylinux*) is so easy, I thought there's maybe something to say for "explicit is better than implicit"?

[joerick]

Yeah, let's just disable that test on Python 3.7 for now then :)

[joerick]

The 06_docker_images test is really just there to check that the CIBW_MANYLINUX1_XXX_IMAGE options are honored, it's up to the user to create a valid config.

[joerick]

Released as 0.9.0!

[hroncok]

Thank You!

[lelit]

Great, many thanks!