Travis: test against coveragepy 4.x

[blueyed]

No description provided.

[dbarnett]

I couldn't tell, was there a reason we needed a particular version range? If so, could you capture that reason in comments somewhere?

[blueyed]

@dbarnett
It currently runs against older coveragepy only, i.e. it fails with the current version - as this PR shows.

[dbarnett]

Travis runs are failing saying "No module named coverage". Is there something missing still?

[dbarnett]

Did you end up figuring out the failures any further?

We do have issues like #39 where the problem turned out to be specific to newer versions of coveragepy.

[blueyed]

I do not remember.
With https://travis-ci.org/github/google/vim-coverage/jobs/396767645 however it looks like it worked to test with coverage.py 4, which caused an error then - likely what I wanted: testing with an older version to see/show that there are compatibility issues / cover them on CI.
Likely related to #35.

[blueyed]

Rebased (to fix conflict).

[dbarnett]

Don't you need to execute this after downloading it? Otherwise I think you're running the system version of pip below.

Here's a working version from one of our other .travis.yml files: https://github.com/google/vim-codefmt/blob/c84463dbed1408abb4003b2105211488d47423d9/.travis.yml#L18
That fetches for version 3.5 explicitly or it blows up on Travis not running python >=3.6, and we sudo to install the packages "system-wide" on the container.

[blueyed]

I've removed the wget - it seems using the system Python is fine here.

Have not looked at the other reference - I am not sure if putting more time into the Travis config is worth it, given that it likely will run out of OSS credits in the long run anyway, and needs to be replaced by e.g. GitHub Actions then.

[dbarnett]

I am not sure if putting more time into the Travis config is worth it, given that it likely will run out of OSS credits in the long run anyway...

Oof, I hadn't realized that was a risk.

I'll probably still pick this up and try to get it working at some point regardless. It will still be useful to figure out a working set of commands for testing against different coveragepy versions, whatever CI system we may end up using. Anyway, thanks for getting the ball rolling!