This is a fun one: Pants puts all the code's requirements in the venv when running mypy, which I believe is the correct thing to do. But I'm now adopting Pants in a repo that already runs mypy in a standalone venv that has a bespoke subset of the code's full requirements. So mypy passes quickly in that setup (and they have a few

# type: ignore

to help it along), but fails slowly under Pants (partly because those

# type: ignore

are unneeded)

Why does it fail "slowly under Pants" ? Mypy cache, or because

mypy

is doing additional work?

AFAICT because mypy is doing a lot more work, analyzing all those requirements

I would expect

mypy

to only look at types when required. That's easily measured. Have

mypy

run in a sandbox with a bunch of reqs, but the file is empty. That should be fast. Then do the same thing, but with a file with a bunch of imports all `# type: ignore`d. I would expect that to be fast as well

It's faster on a clean run, and much faster on a second run with mypy cache

Thats unexpected.

which is the same as the time that process takes under Pants

bad dog