Update CI config

[hugovk]

• There's no need for the "drop the dot" workaround, it will use the interpreter tox is installed to
• Also test on Python 3.8

[codecov]

Codecov Report

Merging #83 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master      #83   +/-   ##
=======================================
  Coverage   70.37%   70.37%           
=======================================
  Files           5        5           
  Lines         297      297           
  Branches       35       35           
=======================================
  Hits          209      209           
  Misses         88       88

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update df6400f...fc68f1b. Read the comment docs.

[ofek]

Nice! Mind if we change on: [push, pull_request] to on: [push] to remove double builds on PRs? See #84

Would that fix it?

[hugovk]

I think that would mean we wouldn't get builds from forks.

For example, this PR was submitted from my fork, and there's only pull_requests shown.

Another example: here's a PR in another repo that also has on: [push, pull_request], from someone who's not a project member: python-pillow/Pillow#4398. There's only pull_request builds shown (click the green tick next to the commit message).

[ofek]

Ah here you go, I found the magic 🙂 #85

[hugovk]

From #85:

on:
  push:
    branches:
    - master
  pull_request:
    branches:
    - master

Will that prevent PRs building from feature branches? And from forks?

[ofek]

Forks no (I think, try it here!) but feature branches on this repo that have not opened PRs yet yes (I expect no CI until PR so that's good imo)

[hugovk]

Usually, the recommended workflow for contributors is to fork the repo, create a feature branch, and then a PR.

So I think it's very important to test against forks and feature branches.

The tests are so quick to run (between 22s - 1m), and we have up to 20 jobs in parallel, I'd be inclined to leave it to test more rather than less.

[ofek]

Fair point, though I think it's atypical for contributors (or anyone really) to look at CI without opening a PR. I know I never do 🙂

Feel free to merge whenever