yapf

certainly isn't the fastest formatter in the west, and paying for format tools to run twice certainly isn't ideal. Of course, using

--changed-since

minimizes the files being run, so the timing overhead might be "annoying" at best. But this also bleeds into the issue of

lint

-ing dependees with formatters, which is wasteful as well 😬

Trying to think how we could do otherwise

Interesting idea to chew on

Saying nothing of it's difficulty or technical feasibility, but if Pants could query the cache and then make decisions from that, and/or produce synthetic cache results... the engine would have unlimited power (two examples:

lint

not running on files where the

fmt

cache is hot, or

lint

running on the subset of files where the

lint

cache isn't hot)

crack theories aside, I wonder if running

./pants fmt lint

could have an option to imply not running the formatters under

lint

Tangentially related to https://github.com/pantsbuild/pants/issues/10542 (in that Pants could be smarter about multiple-goals-in-one-command)

Well, one way to do this would be to proactively cache the fact that the linter passes if the formatter passes

There are several use cases under discussion where we might want to cache "fake" processes that we didn't actually run

Or more generally cache "facts" that don't necessarily correspond directly to a process

So this is another such use case

But there are complications: If you run

fmt

first, and you have multiple formatters, then formatter B might stomp on the output of formatter A, so that when subsequently formatter A is run as a linter, it will fail!

Obviously that is not a healthy or sustainable formatter setup, but it can happen

in which case not running formatter A the second time is actually incorrect