Deduplicate error codes for ignore-without-code

[cdce8p]

Description

#11633 added the awesome ignore-without-code option. During testing I noticed that in some cases an error code would be displayed multiple times. This can be rather annoying, since it's only necessary to ignore an error code once.

# current
error: "type: ignore" comment without error code (currently ignored: [union-attr, arg-type, union-attr])

# new
error: "type: ignore" comment without error code (currently ignored: [arg-type, union-attr])

/CC: @PeterJCLaw

Test Plan

Added a new test case.

[hauntsaninja]

Thanks!

Separately, I found
"type: ignore" comment without error code (currently ignored: [arg-type, union-attr])
just a little confusing. Maybe something like:
"type: ignore" comment without error code (use "type: ignore[arg-type, union-attr]" instead)
is clearer?

[cdce8p]

Thanks!

Separately, I found "type: ignore" comment without error code (currently ignored: [arg-type, union-attr]) just a little confusing. Maybe something like: "type: ignore" comment without error code (use "type: ignore[arg-type, union-attr]" instead) is clearer?

Think that would make sense. I could update this PR tomorrow or open a new one.

[hauntsaninja]

Let's just update this one :-)

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[hauntsaninja]

Thank you!

[PeterJCLaw]

Separately, I found "type: ignore" comment without error code (currently ignored: [arg-type, union-attr]) just a little confusing. Maybe something like: "type: ignore" comment without error code (use "type: ignore[arg-type, union-attr]" instead) is clearer?

For context: it was intentional not to have the error message just tell you to update the ignore, on the grounds that it's likely users would have existing ignores that may not be quite right and just updating the ignore comment blindly may not be a great idea (implemented in 6f39906). The message was thus aimed at making the process of converting the ignores into more specific ones needing just a bit more human input, to encourage review at the same time.
I still think that's useful, though very open to other spellings of the message :)

[cdce8p]

Separately, I found "type: ignore" comment without error code (currently ignored: [arg-type, union-attr]) just a little confusing. Maybe something like: "type: ignore" comment without error code (use "type: ignore[arg-type, union-attr]" instead) is clearer?

For context: it was intentional not to have the error message just tell you to update the ignore, on the grounds that it's likely users would have existing ignores that may not be quite right and just updating the ignore comment blindly may not be a great idea (implemented in 6f39906). The message was thus aimed at making the process of converting the ignores into more specific ones needing just a bit more human input, to encourage review at the same time. I still think that's useful, though very open to other spellings of the message :)

It's a good point. Although from my test updating 100+ ignores, it wasn't all that practical tbh. It's a rather large codebase with only partial annotated code. The main benefit I see with it is not ignoring any new type errors that might come up once you start adding more annotations. That's just my opinion though. If there is a desire to revert it, I wouldn't oppose it.

[hauntsaninja]

Do you feel the following captures the spirit of the original better?
"type: ignore" comment without error code (consider "type: ignore[arg-type, union-attr]" instead)

[cdce8p]

Opened a followup PR: #12216