Fix issue with explicit dependency when using parametrize groups.

[kaos]

When both a target and a explicitly provided dependency is using parametrize groups (the **parametrize(..) syntax), and the dependency does not explicitly provide the parametrization to use, pick the matching parametrization group with the target if available, rather than just the first one.

Closes #20739

[kaos]

Could pick to 2.19.x if we are still running releases for that line..?

[huonw]

I don't understand the mechanics here in detail... but what happens if the parametrizations are slightly different, with different names or fields. For instance, here's a semi-plausible example:

python_sources(
  ...,
  **parametrize("a", resolve="a", skip_mypy=True), # mypy doesn't work in this resolve
  **parametrize("b", resolve="b", skip_mypy=False),
)

python_tests(
   ...,
  # Want to run the tests in a few different environments, so different name & more fields than the source
  **parametrize("a-1", resolve="a", skip_mypy=True, environment="some_docker"),
  **parametrize("a-2", resolve="a", skip_mypy=True, environment="another_docker"),
  # fewer fields than the source, because didn't add the skip_mypy=False explicitly
  **parametrize("b", resolve="b"),

)

Could pick to 2.19.x if we are still running releases for that line..?

Might as well 🤷‍♂️ releases are easy enough now.

[kaos]

I don't understand the mechanics here in detail... but what happens if the parametrizations are slightly different, with different names or fields. For instance, here's a semi-plausible example:

python_sources(
  ...,
  **parametrize("a", resolve="a", skip_mypy=True), # mypy doesn't work in this resolve
  **parametrize("b", resolve="b", skip_mypy=False),
)

python_tests(
   ...,
  # Want to run the tests in a few different environments, so different name & more fields than the source
  **parametrize("a-1", resolve="a", skip_mypy=True, environment="some_docker"),
  **parametrize("a-2", resolve="a", skip_mypy=True, environment="another_docker"),
  # fewer fields than the source, because didn't add the skip_mypy=False explicitly
  **parametrize("b", resolve="b"),

)

We only match on the a vs a-1 and b here (i.e. the "name" of the parametrization group), not the fields. (if you go to the source and see field value matching stuff, that's for the "regular" field parametrizations like resolve=parametrize("one", "two"))
So it would pick the "first" one when there is not a matching group on both ends. (which is the same behavior as now.)

As the actual parameter values are not part of the address for grouped parametrizations, we don't know them as we only work with addresses for the matching, so this is the best we can do, currently, I think (with little effort). (or at least, it's a good first step which we can follow up on to improve later if we want..)

Could pick to 2.19.x if we are still running releases for that line..?

Might as well 🤷‍♂️ releases are easy enough now.

✔️

[cognifloyd]

That would not be a fun bug to run into! This fix makes sense to me.

[WorkerPants]

I tried to automatically cherry-pick this change back to each relevant milestone, so that it is available in those older releases of Pants.

✔️ 2.19.x

Successfully opened #20778.

✔️ 2.20.x

Successfully opened #20779.

---

Thanks again for your contributions!

🤖 Beep Boop here's my run link