Nice

The suppression range here is a bit misleading because it suggests that all codes are unused. I think I would either:

• Highlight the entire suppression diagnostic. That makes it clear that something's up with the suppression and not with an individual code
• Keep underlining the codes, but only group adjacent codes (in which case we'd emit two diagnostics in this case). I slightly prefer this as it gives more precise ranges.

Don't bother with this if this matches what we do for noqa today.

It looks like noqa highlights the entire noqa comment:

RUF100 [*] Unused `noqa` directive (unused: `F401`, `E741`)
 --> foo.py:2:12
  |
1 | def f():
2 |     x = 1  # noqa: F401, F841, E741
  |            ^^^^^^^^^^^^^^^^^^^^^^^^
  |
help: Remove unused `noqa` directive
1 | def f():
  -     x = 1  # noqa: F401, F841, E741
2 +     x = 1  # noqa: F841

I would then suggest aligning with noqa (which I think should be relatively easy?)

Yeah I liked the idea in #21873 (comment) of extending the range to only the adjacent codes (if I understood correctly), but aligning with noqa makes sense too.

I've updated the highlighting logic so that:

• if only one code exists or only one is being removed, it highlights that one code
• otherwise, it highlights the entire suppression

I think this matches noqa now.

This is probably better suited for a follow-up, and I thought I saw some previous discussion of this, but do these need to be Strings? I tried making UnusedCodes::disabled a Vec<&'a str> and it seems to be working locally. Combining that with into_values makes this a little nicer:

I didn't like this either, but since it would also affect the noqa system I'd prefer to put that change in a different PR

Yeah I liked the idea in #21873 (comment) of extending the range to only the adjacent codes (if I understood correctly), but aligning with noqa makes sense too.