Using commas wrong in a type ignore makes there be an error code in there that is the empty string

[wyattscarpenter]

Expected behavior:

[case testErrorCodeWarnUnusedIgnores9_testUnusedIgnoreSloppyFormattingNotEmptyBrackets]
# flags: --warn-unused-ignores
1 + "ok" + "ok".foo # type: ignore[  operator,attr-defined]
1 + "ok" + "ok".foo # type: ignore[  ,operator,attr-defined] # E: Unused "type: ignore" comment  [unused-ignore]
1 + "ok" + "ok".foo # type: ignore[  operator,attr-defined,] # E: Unused "type: ignore" comment  [unused-ignore]
1 + "ok" + "ok".foo # type: ignore[  operator,attr-defined,,] # E: Unused "type: ignore" comment  [unused-ignore]

[case testErrorCodeWarnUnusedIgnores10_testUnusedIgnoreSloppyFormattingEmptyBrackets]
# flags: --warn-unused-ignores
1+1 # type: ignore    # E: Unused "type: ignore" comment  [unused-ignore]
1+1 # type: ignore[]  # E: Unused "type: ignore" comment  [unused-ignore]
1+1 # type: ignore[ ] # E: Unused "type: ignore" comment  [unused-ignore]
1+1 # type: ignore[,,, ] # E: Unused "type: ignore" comment  [unused-ignore]

Actual behavior:

When there is a trailing comma, a leading comma, or other comma abuse, the error is Unused "type: ignore[]" comment instead, because the empty string error code still counts.

I could easily have suppressed this symptom at the site of the unused ignores warning, but this seems like something that should probably be fixed at the source, to avoid other almost-invisible erroneous behavior from popping up later.

[codenamenam]

This would be fixed by #20052

[wyattscarpenter]

Wow, thanks!

Thanks also for fixing my proposed test, which are a little confused in the OP, due to me working on #19904 simultaneously.

[codenamenam]

well, I just made some minor adjustments based on your test cases. Thank you for your contribution!