I'm seeing behavior where the Pyright backend is caching at the granularity of the top-level call. E.g. if I call

pants check my_package/::

and get a successful result, the a repeat of that call will return a cached result, but if I edit any file within

my_package/::

, a new call to

pants check

will run against all files again, even if the changed file is a leaf node in the larger dependency tree. Is this the expected caching behavior of the Pyright backend? Are there plans to make it granular to each source file?

Type checking usually has cross-file dependencies, eg if

file.py

has a

from .sibling import function

import, then type checking

file.py

needs

sibling.py

, and thus changes to

sibling.py

require type checking

file.py

too (and for all transitive downstream code too) So, the checking goals typically work at a whole-repo level. This can definitely be improved eg theoretically could happen in a vaguely per-file way, with appropriate support from pyright (since checking a file actually only needs the definitions from each import, doesn’t need to type check all their internal code) For mypy, pants has support for getting mypy to use it’s own cache, so that rechecking a barely-changed repo is still pretty fast (not as fast as unchanged of course). Does pyright have its own caching? If so, I don’t know if pants enables it

As far as I know Pyright only has an in-memory cache and the maintainers didn’t want to deal with cache invalidation

@ancient-terabyte-9085 if you can find a way to run the

pyright

command line tool (^ with no daemon, i.e. not using

--watch

) that is faster, then it is probably possible and just needs more or less elbow grease since this is a volunteer driven project. It shouldn't be too hard to experiment with the

pyright

CLI and find the answer to 1st order. ^ Pants does not have easy support for daemon tools right now that stay up and running between Pants runs.

Yeah, sharding the type checking by "connected component"/directories of dependencies would probably be feasible. Feel free to a file a feature request issue, although, as John notes, Pants is volunteer driven so there may not be progress on it unless someone (maybe you 😄 ) feels inclined to pick it up.