In V2 land, what is the unit of work for code quality tools like isort and Pylint? Would isort get a new invocation for every single Python Target, or like in V1 would it run over the whole set of targets in one invocation?

that should be dependent on what’s faster, not the engine executing it. in v1 junit tests we have

--fast

which makes that choice (all at once or per target) configurable

You mean that both are potentially valid and that we could decide which execution model to use? We would test out both models (all targets per invoke vs. one target per invoke) and determine which is faster?

In general per-target is good because it means that we get caching with fine granularity

definitely want to bias toward per-target, yea. it's also generally more correct

And then who in V2 land determines whether the target should be ran against again? E.g. if we just ran isort on

foo.py

, then we could avoid running it again. Would this be local caching?

caching in general

As a general rule, also, all `@rule`s should be 100% cacheable in all situations, but `@console_rule`s will be unconditionally run, so if you’re planning to e.g. use

isort

to change file contents, rather than just check whether something is correct, it should be the

@console_rule

which does that, and we have not entirely worked out what API it should use to actually manipulate the filesystem

unless you were to import and use

subprocess

in a

@(console_)rule

, or use

open

, it shouldn't be able to do that damage