2.13.17 regression (?): Scalafix failing in community build

[SethTisue]

reproducible locally using dbuild (also probably reproducible outside dbuild if you locally publish scalameta?)

https://scala-ci.typesafe.com/job/scala-2.13.x-jdk17-integrate-community-build/2616/artifact/logs/scalafix-build.log

[info] - scala/test/removeUnused/RemoveUnusedPatternVars.scala *** FAILED *** (98 milliseconds)

--- obtained
+++ expected
@@ -385,3 +385,3 @@
   def f(v: (Int, (Boolean, String))): Int = v match {
-    case (i, v @ (_, _)) => i
+    case (i, (_, _)) => i
   }

I haven't bisected it but it's in this range:

-nightly=2.13.17-bin-483b4ee
+nightly=2.13.17-bin-c392ee5

eyeballing that range, scala/scala#11032 looks like a possible culprit (@som-snytt)

[som-snytt]

v is no longer reported because that is the selector expression.

Compare

x match {
  case Some(x) =>
}
// generalizes
x match {
  case x: Some[_] =>
}

Remark at scala/scala#11012 (comment)

I comment that it doesn't do tuples (i, j) match { case (i, j) => } but a possible narrowing of the behavior would be to allow it only if there is exactly one pattern var. Then v would warn because i is present (albeit used).

Dotty is not (yet) so liberal, but considers only a top-level Bind in the pattern. Probably I intended to import the behavior (or the improved behavior). (The dotty check has regressions or other tickets with priority.)

[som-snytt]

Although underreporting might be better, I don't mind reverting that one LOC to restore the warning. That is, revert to current dotty.

I'll prepare a quick PR for your consideration.

[som-snytt]

The OP should work as expected, but the PR still allows case Some(x) and similar. In particular, case email(s, _) => for arbitrary extraction of one patvar.