Use PyTest

[bonjourmauko]

Technical changes

• Use pytest to run tests.
• Declare nose a dependency so dependee libraries can move away from nose.
• nose is in maintenance mode
• nose2 was a spin-off of unittest2

Note: openfisca-run-test still depends on nose.

[fpagnoux]

👍

[fpagnoux]

⚠️ No, by adding format-style here, we are actually not flaking the code anymore in the CI! ⚠️

Now tests pass with bad style because autopep8 fixes the style before flake8, but the style fix are not commited!

[Morendil]

Thanks for catching this!

A slightly longer comment in the main discussion.

[bonjourmauko]

Oops ! Thanks for pointing that out @fpagnoux

[fpagnoux]

@maukoquiroga @Morendil part of the changes are problematic as they de facto remove code flaking for the CI.

[Morendil]

Hi @fpagnoux and thanks! I'd like to make a comment here at the "learning review" level, as this is not the first time something goes to merge that we have reason to regret afterwards.

Repeated incidents are a good reason to change the conversation from "shit sometimes happens" to "what do we need to change". As you already noted, one item up for discussion is "don't rush to merge", and it's a reasonable position. At least one common characteristic in the two incidents was a short amount of time between PR created and PR merged.

There's a weaker signal that I don't want to get overlooked, which is a tendency to "over-improve", that is, both PRs went beyond the strictly necessary to achieve the stated objective. An alternative approach would have been to split the work beetween one PR to add PyTest, and strictly that, and a second one refactoring the Makefile. I'm all too aware that one factor in that kind of decision is the transaction costs - each PR requires waiting for CI, getting a review, occasionally being stuck for a while on "changes requested", updating Changelog, rebasing, etc. - all of which involve wait times of various lengths.

The larger objective I'm working on at the moment (streamlining contribution, on France in particular) requires a sharper than usual awareness of these various costs from friction, and I make a conscious effort to compensate in the other direction - by approving, fixing what I can, rebasing, updating Changelogs, working at odd hours. In both incidents this (specifically over-eager approve) was an aggravating factor.

[bonjourmauko]

Thanks for your vigilance @fpagnoux

Thanks four your tremendous work @Morendil

Error was carried from openfisca/openfisca-france#1178, my fault.

Repeated incidents are a good reason to change the conversation from "shit sometimes happens" to "what do we need to change". As you already noted, one item up for discussion is "don't rush to merge", and it's a reasonable position. At least one common characteristic in the two incidents was a short amount of time between PR created and PR merged.

👍

There's a weaker signal that I don't want to get overlooked, which is a tendency to "over-improve", that is, both PRs went beyond the strictly necessary to achieve the stated objective. An alternative approach would have been to split the work beetween one PR to add PyTest, and strictly that, and a second one refactoring the Makefile. I'm all too aware that one factor in that kind of decision is the transaction costs - each PR requires waiting for CI, getting a review, occasionally being stuck for a while on "changes requested", updating Changelog, rebasing, etc. - all of which involve wait times of various lengths.

To be completely honest I was lazy and copy pasted the same changes over four pull requests, in an attempt to normalise (style, etc.). Error carried from the first one, that's not yet approved.

The larger objective I'm working on at the moment (streamlining contribution, on France in particular) requires a sharper than usual awareness of these various costs from friction, and I make a conscious effort to compensate in the other direction - by approving, fixing what I can, rebasing, updating Changelogs, working at odd hours. In both incidents this (specifically over-eager approve) was an aggravating factor.

Reducing rotten backlog is awesome. I think this is a good opportunity to assess the maturity we need to give for contributors to help improve others' contributions.

But this is all learning, and we have IMHO a happy problem 😃 .

That being said, I'll clean my dog shit now.