Uninline diagnostic position before passing it into xsbti.Reporter

[dos65]

Originally was found in scalameta/metals#2769 .
Without this fix, metals receives diagnostic that points into the wrong source file

[nicolasstucki]

Can we add a test in this project or is that something that needs to be added in metals?

[soronpo]

Actually, this not just affects metals, and I have a compiler test-case. I'll try to minimize it further.
I already tested and this PR fixes it.

[dos65]

@nicolasstucki
As @soronpo said it affects not only metals but also bloop (+ probably sbt-bsp).

Is there any reason why the position here shouldn't be nonInlined?
For example, ConsoleReporter does the same thing.

[soronpo]

(the examples assumes the latest update to the summonInline implementation)
Minimized:

trait Foo
inline def foo : Any =
  compiletime.summonInline[Foo]
val f = foo

Without the fix:

With the fix:

4 |val f = foo
  |        ^^^
  |        no implicit argument of type Foo was found
  | This location contains code that was inlined from Main.scala:3

[nicolasstucki]

Is there any reason why the position here shouldn't be nonInlined?

Not that I see right now.

The test would be good just to avoid regressions.

[dos65]

@nicolasstucki Oh. Yes, it's easy to cover by test in metals. Will do that.

[soronpo]

Actually, I was mistaken. My example error is shown in IntelliJ, so yeah bloop is affected but I have not been able to replicate it just as a compiler error.

[nicolasstucki]

Then let's add the test to metals.