respect noqa

[jakkdl]

fixes #181 Implemented:

• respects noqa
• new pyright error about how to in-line silence stuff.
  • it would be nice to catch errors-due-to-updated-linters with a bot, instead of me randomly encountering them when writing new PRs and being surprised.
• various small test improvements
  • changes parametrized params to test_eval so summaries are more informative than e.g. test_eval[False, True, TRIO100, path7]
  • sort errors on columns late, since columns have a habit of getting messed up and thrown around (and might be ignored anyway)
  • ignore columns when checking with autofix
  • the scenario here is two errors on the same line, the first which modifies the line, causing the second to have a different column

remaining TODO:

• don't autofix noqa'd errors
  • ... ah shit, I have to kind of completely change when errors get suppressed to fix this - since visitors have to know if an error is noqad when they raise/autofix it.
• support noqa-trio (or similar / depends on discussion in #181)
  • handle NQA102 for noqa-trio
  • separate error, or command-line flag, if code not enabled
  • check that ignored code is valid/exists
  • maybe give separate error if it's not enabled, see above
• autofixing two errors in the same line seems broken
  • TRIO100 autofix looks like it's eating comments
  • TRIO911 autofix does not get executed, although that's possibly due to it being a SimpleStatementSuite
  • might fix these in a separate PR which i rebase this onto, or just do NOAUTOFIX
• noqas with invalid code is parsed as bare noqa (but that's same as flake8, so /shrug )
• report old regex101.com link upstream in PyCQA/flake8
• don't suppress errors when running as a plugin
• --disable-noqa

[Zac-HD]

it would be nice to catch errors-due-to-updated-linters with a bot, instead of me randomly encountering them when writing new PRs and being surprised.

Yeah, I think it's time that we started pinning our dependencies. Steal the pip-tools setup from shed in a new PR? The idea is to pin all the transitive deps, and then have a tox env that updates them - for now we can just run that manually every so often.

[Zac-HD]

don't autofix noqa'd errors

For simplicity I think we can skip this too for now; if it's autofixable I think we're OK with doing that and if I turn out to be wrong, well, that's a future problem 😝

[jakkdl]

don't autofix noqa'd errors

For simplicity I think we can skip this too for now; if it's autofixable I think we're OK with doing that and if I turn out to be wrong, well, that's a future problem stuck_out_tongue_closed_eyes

well, if you have any TRIO91x errors noqa'd they'll be autofixed - although I suppose extra lowlevel checkpoints won't do anything. noqa'd TRIO100 being autofixed is quite a bit worse, but maybe not too common. But I can postpone it to a separate PR regardless.

[jakkdl]

Oh actually, current solution is very incompatible with flake8-noqa as it suppresses errors even when run non-standalone...

[jakkdl]

But feel free to merge and test against live code, and if it works up to you how badly you want to break flake8-noqa, or wait until followup PRs.

[jakkdl]

nowait, this also needs to ignore ast visitor errors. but can fix that in another PR

[Zac-HD]

Looks like we've got some useful fixes here for the other PRs, so merging this + #184 now. I'll test locally ~monday (I hope 🤞), and cut a release either then or after #185 if that turns out to be really necessary. Thanks again!