Scala 3: semantic info for Infix arguments cannot be looked up

[bjaglin]

scalafix/scalafix-tests/shared/src/main/scala/test/SymbolTest.scala

Lines 10 to 14 in 7cfbad5

	trait SymbolTest {
	def shouldBe(right: Any): Unit
	def arg = 1
	this shouldBe (arg)
	}

L13: Term.ApplyInfix(_, Term.Name("shouldBe"), _, arg :: Nil)

Symbol lookup for arg fails against Scala 3.1.1 SemanticDB as the position there excludes surrounding parentheses while 2.x (scalac-semanticdb) and the parser include them

https://github.com/scalacenter/scalafix/pull/1528/files#r846793225

[tgodzik]

Good catch! Did you report it in the Scala 3 compiler?

[bjaglin]

@tgodzik to be honest, I wasn't sure what was at fault here. Is there any spec or tests in the reference implementation I could link to if I create a ticket on dotty?

[tanishiking]

I checked how the Scala3 compiler generates the SemanticDB for infix argument tanishiking/scala3@5781617 (though the latest Scala3 generates it in main branch, there's no change around the position of symbol occurrence these days AFAIK).
But it seems like the symbol position for arg takes the surrounding parentheses into account.

[tanishiking]

I got it, for the following code

trait SymbolTest {
  def shouldBe(right: Any): Unit
  def arg = 1
  this shouldBe (arg)
}

Scala2 (semanticdb-scalac) generates:

trait SymbolTest/*<=_empty_.SymbolTest#*/ { 
  ...
  this shouldBe/*=>_empty_.SymbolTest#shouldBe().*/ (arg)/*=>_empty_.SymbolTest#arg().*/ 
} 

[3:16..3:21): (arg) => _empty_/SymbolTest#arg().

and Scala3 generates:

trait SymbolTest/*<=_empty_.SymbolTest#*/ { 
  ...
  this shouldBe/*=>_empty_.SymbolTest#shouldBe().*/ (arg/*=>_empty_.SymbolTest#arg().*/)
} 

[3:17..3:20): arg -> _empty_/SymbolTest#arg().

[tanishiking]

I personally see the Scala3 generated symbol occurrence for arg seems more accurate, and it would be better to handle on the scalafix side? 🤔

[tgodzik]

Hmm... having just arg seems more resonable, or are we missing something? The alternative would be to change it in the parser and in semanticdb Scala 2 plugin

[rochala]

The parenthesis inclusion is done only by scalameta parser. Both scala2 and scala3 trees at compilation exclude them and properly show position for arg ( excluding the parenthesis ). Scala3 semantic db position is thou different because its based on trees from compiler.
I'm no expert but in my opinion fixing it in scala3 would be a hack.

[tanishiking]

I think scalameta parser should exclude The parentheses (if there're no special reasons), that should fix both scala2 and scala3?

[tgodzik]

I think scalameta parser should exclude The parentheses (if there're no special reasons), that should fix both scala2 and scala3?

Looks like it! We should use the Positions suite for that, are you able to take a look in scalameta @tanishiking ?

[tanishiking]

I found some comments here https://github.com/scalameta/scalameta/blob/1a9ec6f9aa06e20bc786b5c7ef6d7aa500e59eaf/scalameta/parsers/shared/src/main/scala/scala/meta/internal/parsers/ScalametaParser.scala#L1440-L1444

[bjaglin]

Thanks @tanishiking & @tgodzik !