Make NoAutoTupling work with Function1

[gabro]

See the discussion here and here.

The failing cases (rewrite not applied) are

def f: (((Int, String)) => ((String, List[Int])) => Int) = a => b => a._1
f(1, "")("a", Nil)

and

val f = (a: (Int, Boolean)) => a
f(2, true)

This is currently blocked by scalameta/scalameta#837

We can tackle this in two ways:

• by reading denotations Handle auto-tupling on functions #194 (currently blocked by Emit denotations for all used symbols scalameta/scalameta#930)
• by reading compiler messages (blocked by imprecise positions in errors emitted by Scalameta)

[olafurpg]

I just realized that this is still an issue when the Function1 is defined in a 3rd party library, a minimal repro is here 112a355

The problem happens when we don't have access to the denotation. We currently only store denotations for symbols compiled with the Scalahost compiler plugin enabled. I think we should keep this issue open. I believe a long-term solution to this issue would be to use reported compiler warnings, as proposed #194.

[gabro]

Right, I'll investigate further on the subject, but I think we'll need a scalameta "hack" to get the caret positions of those messages.

[olafurpg]

Yeah, we can accommodate on the Scalameta side to make the positions of the reported messages more accurate.

[olafurpg]

Seems scalameta/scalameta#930 is a go, which means we will store the denotations for all used symbols per .semanticdb file.

[gabro]

Cool! I still think compiler messages would be much more accurate than any name/denotation based approach, but that feature is certainly going to help in the future.

[olafurpg]

Is this fixed now that we use messages?

[gabro]

Yes!