Pattern sequence 'Pattern2*' is allowed by grammar, but behaves strange

[SrTobi]

Compiler version

3.0.0-RC2-bin-20210217-83cb8ff-NIGHTLY

Minimized code

Seq(1, 2) match
    case Seq(x, y*) => println(y) // prints List(2) which looks correct

Seq(1, 2) match
  case Seq(x, (y)*) => println(y) // parses, but gives a compiler error that y in the pattern is not found 

val y = 3
Seq(1, 2) match
  case Seq(x, (y)*) => println(y) // compiles and outputs 3, which is curious because what did (y) match exactly then?

Output

Output for

Seq(1, 2) match
  case Seq(x, (y)*) => println(y) 

[error] -- [E006] Not Found Error: /dotty-test/src/main/scala/Main.scala:8:17 
[error] 8 |    case Seq(x, (y)*) => println(y)
[error]   |                 ^
[error]   |                 Not found: y
[error] -- [E006] Not Found Error: /dotty-test/src/main/scala/Main.scala:8:33 
[error] 8 |    case Seq(x, (y)*) => println(y)
[error]   |                                 ^
[error]   |                                 Not found: y
[error] two errors found
[error] two errors found
[error] (Compile / compileIncremental) Compilation failed

Expectation

Not 100% sure what to expect... maybe something where the pattern matched multiple times?
It just looks only half implemented to me...

[liufengyun]

For simplicity and consistency with Scala 2, maybe it's better to restore the following restrictions:

• In pat: T, restrict pat to a pattern variable
• in pat*, restrict pat to a pattern variable

There are currently no known use cases for the new features. If there is a strong need, the feature can come after 3.0, as it does not change the existing syntax.

[SrTobi]

What is supposed to happen here anyway? Why did the definition change here from Scala 2 which only allowed id/_?

[sjrd]

It seems to be an accident. There is nothing except a pattern variable, or an _, that could possibly make sense in that context.

It was probably a side effect of allowing patterns on the left of : T, where @odersky argued that it made sense and should not be reverted. But for * it makes no sense.

[liufengyun]

The PR that introduced pat*:

#11240 (comment)