I disagree with the change made to block tests on linting:
0db530a
We did discuss this elsewhere the other day, I didn't realize the change had been made here until now and is spreading to other repos. (We should probably use PRs in this repo.)
Reasons to run lint in parallel:
- All jobs finish faster.
- Optimize for human time (the most important and expensive metric IMHO).
- CI lint failures are, or can be, rare due to linting in editors and local lint runs.
Reasons to block tests on lint were:
- Wastes testing time/resources if lint issues are found and need to be fixed.
- Lint job is fast so it doesn't slow humans down.
I don't have data for the rarity of lint failures. But if true, I think we should optimize for human time as much as possible and run everything in parallel. To that end, I have sometimes been waiting for coverage tests to finish and pondered just making those run in parallel as well. Only reason I hadn't proposed that yet is I wanted to test what it does in the failure case.
I disagree with the change made to block tests on linting:
0db530a
We did discuss this elsewhere the other day, I didn't realize the change had been made here until now and is spreading to other repos. (We should probably use PRs in this repo.)
Reasons to run lint in parallel:
Reasons to block tests on lint were:
I don't have data for the rarity of lint failures. But if true, I think we should optimize for human time as much as possible and run everything in parallel. To that end, I have sometimes been waiting for coverage tests to finish and pondered just making those run in parallel as well. Only reason I hadn't proposed that yet is I wanted to test what it does in the failure case.