Add big codebases that are not formatted with Black

[JelleZijlstra]

Currently we run diff-shades on repos that are already formatted with Black, but I think it would be useful to add some non-Blackened projects, so we get a sense of how we format code that isn't already mostly Black-like. This is especially relevant for functionality that interacts with the magic trailing comma, like psf/black#2368.

Some ideas:

• https://github.com/google/pytype, almost a million lines of two-space indents
• CPython, which surely has lots of interesting stuff in its test suite. I think some files there currently crash Black, though.
• https://github.com/cython/cython, lots of code but I'm not sure it has anything unique

It looks like diff-shades might get slow enough that we should be selective in what we include.

[ichard26]

We could steal even more ideas from mypy's integration of mypy-primer which is to spread out the projects over multiple parallel jobs but that'd introduce a fair amount of complexity (and would worsen the number of jobs issue psf/black already has).

[ichard26]

Another way to deal with the slowness would be to compile the baseline/target revision before running it against the projects. Assuming perfect efficiency this could save 6-10 minutes for an uncached run or 3-5 minutes for a run where the baseline was cached. This effectively means we are 100% trusting mypyc but I guess we were heading in that direction already :)

[ichard26]

One more optimization I thought about last night was to reuse the code stored in the analyses instead of recloning the projects. This wouldn't help runs where the baseline wasn't cached but this would help quick iterations on PRs saving 20 to 60 seconds depending on network performance. I'd have to store the pyproject.toml contents too but that's relatively simple compared to more parallelism.

[ichard26]

For what it's worth, we probably want to avoid any projects beyond half a million lines of code since they are slow to format and start to worsen the signal to noise / time ratio. IMO maintaining / improving variety is more important. If we really want to add bigger projects we might want to cut down on the line count using --exclude.