The first cut failed because accessors were star-typed, or star-crossed.

scala> class C(val xs: Int*) { def f = xs }
defined class C

scala> val c = new C(1,2,3)
c: C = C@7ddd84b5

scala> c.f //print
   $line4.$read.$iw.$iw.c.f // : Seq[Int]

scala> c.xs //print
   $line4.$read.$iw.$iw.c.xs // : Int*

I tested bootstrap build locally thanks to @lrytz instructions on another PR, maybe he can also say how far off base this tweak is. I didn't follow type completers in namers far enough to get a sense of it.

The tweak probably isn't tweaky enough, but one test failure is because it relies on previous star in function:

val testAllS: (Int, =>Int, Int*) => Int = testAll _
test("all s", 8, testAllS(1, 5, 78, 89))

It also re-broke varargs in ctors. I guess that relies on the accessors having star type.

This area fills me with dread, but I've been here before so I'll assign myself to this ticket. Let's move the PR to 2.13.x though.

ping @retronym

hmm i don't see it, which part of the diff fixes the bug?

I see the fix allows star in the signature, but requires Seq in the application. Essentially, instead of disallowing it, the star becomes a kind of alternative syntax for Seq. The spec language would say that the star type means Seq everywhere except in a method application.

Probably that made more sense back in the days of -Yeta-expand-keeps-star. But since that is no longer an option (literally), it should be disallowed except for method parameters.

I kind of like the current fix, but it's rather eccentric.

Added a commit that disallows T* in (T*) => U.

Maybe that should be deprecated until -Xsource:2.14?

The previous commit still does something, compare to last year:

$ skala
Welcome to Scala 2.13.0 (OpenJDK 64-Bit Server VM 1.8.0_181)

scala> class C(val xs: Int*) { def f = xs }
defined class C

scala> val c = new C(1,2,3)
c: C = C@7cea0110

scala> c.f //print
   c.f // : Seq[Int]

scala> c.xs //print
   c.xs // : Seq[Int]

The original report, that import X.f ; f(1,2,3) fails but X.f(1,2,3) succeeds, is fixed by stripping star in typedSelect. But with the parser change, the star type is not expressible, so you can't import something of star type. The print-tab in REPL might see it if I reverted that change; I haven't tried that yet.

It's probably the aliasing bit that retronym found dreadful. I don't see a new test, and I don't remember which tests fail without that change, which I'll have to puzzle over a minute.

You may have noticed 38 Down in the Friday NY Times crossword, "rotary phone", @retronym so-called. I wonder if "compiler engineer" could be considered a retronym. Then is retronym a compiler engineer or the other way around?

closing for inactivity, can be revived & reopened whenever

Since retronym is dreadfully hung up, I moved the other commits to separate PRs.

new PR: #7399