Closures parsed as self-type annotation

[Blaisorblade]

I'm not sure I care and I can't conceive a use case (at least, not without involving some weird macro), but the following is valid for scalac and rejected (at parse-time) by dotty:

class Foo {
  (a: Int, b: String) => a
}

The reason is that dotty parses (what looks like) an expression, then sees => and "reinterprets" the parsed expression as a self-type annotation.

I'm not so concerned about the feature, I'm more concerned about aligning the grammar with the implementation.

I noticed this while trying to document errors for self-type annotations.

So this gives something like:

> run -explain tests/Foo.scala
[...]
[info] Running dotty.tools.dotc.Main -explain tests/Foo.scala
-- Error: tests/Foo.scala ----------------------------------------------------------------------------------------------
2 |  (a: Int, b: String) => a
  |  ^^^^^^^^^^^^^^^^^^^
  |  not a legal self type clause
-- [E006] Missing Identifier Error: tests/Foo.scala --------------------------------------------------------------------
2 |  (a: Int, b: String) => a
  |                         ^
  |                         not found: a

Explanation
===========
An identifier for `a` is missing. This means that something
has either been misspelt or you're forgetting an import


two errors found
java.lang.RuntimeException: Nonzero exit code returned from runner: 1

BTW I don't like "missing identifier", I think it's trying to say "unbound identifier" (I have a local patch I'll send at some point).

[liufengyun]

Thanks for reporting @Blaisorblade . This is related to #1424.

Unlike #1424, it seems the parser can do a better job here, as it's possible to know syntactically it's not a self-annotation, but a lambda def.

[Blaisorblade]

Thanks I missed that! Also Martin there claims that handling this in Scalac is very tricky, so maybe it shouldn't be fixed. Making sure the grammar rejects this might be enough.

[odersky]

BTW I don't like "missing identifier", I think it's trying to say "unbound identifier" (I have a local patch I'll send at some point).

I don't like that either. But "unbound" is probably too academic for some. How about

No definition for `a` was found.

?

[odersky]

I forgot to say: "Unbound Identifier" looks fine as a name for this class of error messages. But the explanation should say something else.

[odersky]

I think we should not try to overengineer fixing this. The scalac code doing this is quite impenetrable. I think "not a legal self type clause" is perfectly fine as an error message.

[Blaisorblade]

BTW I don't like "missing identifier", I think it's trying to say "unbound identifier" (I have a local patch I'll send at some point).

I don't like that either. But "unbound" is probably too academic for some. How about

No definition for a was found.
?

I was also concerned on unbound, I'll probably use "declaration" rather than "definition".

I think we should not try to overengineer fixing this. The scalac code doing this is quite impenetrable. I think "not a legal self type clause" is perfectly fine as an error message.

Agreed, especially since the grammar already excludes this (which I missed) and the current parser is already nontrivial. I'll close this as duplicate.