Summary

Importer::get_or_import_symbol can fail, and if it does it returns a useful ResolutionError that tells you why it failed. The message from this error is printed to the terminal if you run Ruff with enough -v arguments on the command line. However, we're currently throwing away this useful information in the implementation of this rule by calling .ok() on the value returned by get_or_import_symbol. There's no reason to do this, as it doesn't really simplify the code; we already have a Diagnostic::try_set_optional_fix() method that's designed exactly for this situation.

The fix also makes an incorrect edit if typing is already imported in the scope being edited -- here, it makes a fix that would introduce a NameError (because Self is not imported anywhere in the file):

~/dev/ruff (main)⚡ [2] % cargo run -- check --no-cache --preview --diff --fix --select=PYI019 --target-version=py313 foo.pyi
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.11s
     Running `target/debug/ruff check --no-cache --preview --diff --fix --select=PYI019 --target-version=py313 foo.pyi`
--- foo.pyi
+++ foo.pyi
@@ -1,5 +1,5 @@
 import typing
 
 class F:
-    def m[S](self: S) -> S: ...
+    def m(self) -> Self: ...

This is because we throw away the second element of the tuple returned by get_or_import_symbol_here:

In this case, the second element would be the string "typing.Self", which is what we should use in the Edit. But instead we just pass in the string "Self", assuming the at the import edit would have added from typing import Self at the top of the module (it only does that if both typing nor Self are currently not imported):

Test Plan

I added some fixtures that exhibit incorrect behaviour on main.