Duplicate diagnostics in unpacking targets

[dhruvmanila]

Summary

Consider the following code:

class Foo():
    pass

a, b = Foo().nonexistant

Here, the RHS of the assignment has an error unresolved-attribute which is being shown 3 times (2 + 1 where 2 is the number of variables involved in the unpacking):

This is because the standalone expression is being inferred outside the unpacking in infer_assignment_definition (and similarly in infer_for_statement_definition). This inference should only be done for non-unpack targets.

Fixing this would lead to another bug which is that the diagnostics raised during the unpacking query isn't being merged into the outer type inference builder.

Version

No response

[dhruvmanila]

This should be relatively easy to fix once astral-sh/ruff#18041 has landed. This is because that PR adds a Definition ingredient for all attributes which would then allow us to use the infer_unpack_types for all attributes involved in an unpacking.

[dhruvmanila]

I tried fixing this today based on the assumption that assignments containing an attribute expressions now always create a Definition but that isn't true. It's only created if PlaceExpr::try_from is successful, so a call expression like the following doesn't create a Definition. I'll have to re-think the approach that I used here.

class Foo:
    def __init__(self):
        self.x = 1

Foo().x = 2