Restrict implicit args to using

[som-snytt]

Typer#adaptNoArgsImplicitMethod populates implicit args when an arg list is missing. To remedy missing implicits, it tries a named application using args it did find. Then Applications#tryDefault supplies a default arg if available. A previous fix to allow tryDefault to supply implicit args for implicit params is now restricted to explicit using; typer now adds using for implicit when it needs to try defaults.

This commit restores propagatedFailure and the previous condition that default params are tried if there is an error that is not an ambiguity. An additional restriction is that default params must be useful: there must be a param which has a default arg to be added (because it's not a named arg).

Fixes #22439

[som-snytt]

Aspirationally,

Since issueErrors specifies endPos, the position used for
implicitArgTree by matchArgs is also tweaked to be the endPos
(of the appPos). That places a single caret in the familiar
spot following the existing application, instead of underlining.

That needs a mechanism for distinguishing user's using from typing for default args, where user expr requires underlining, but the latter case wants trailing caret.

[EugeneFlesselle]

I'm in my exam period until 05/02, so I won't be able to review before then. I'll be happy to take a look once I can though.

[som-snytt]

@mbovel you were in this area last spring, and the LOC I restored will be familiar to you. (Maybe over-familiar.) If you wouldn't mind revisiting it for a review? This commit tweaks Eugene's previous fix by tweaking your previous fix.

[noti0na1]

LGTM. Very nice cleaning!

[som-snytt]

just a rebase

[som-snytt]

Just realized that the new error is the fixed ticket: the code has spurious parens.

[error] 1575 |        val (types, values) = pats.map(evalPattern(scrutinee, _)).unzip()
[error]      |                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[error]      |missing argument for parameter asPair of method unzip in trait StrictOptimizedIterableOps: (implicit asPair: ((dotty.tools.dotc.core.Types.Type, Objects.this.Value)) => (
[error]      |  A1, A2)): (List[A1], List[A2])
[error] one error found

[som-snytt]

I appreciate the explanatory comment from January, but frankly it's not very meaningful to me right now in March.

OK, after reading it a couple of times, I kind of understand it.

Additionally, a reminder that it's always futile to update vulpix check files "by hand". The "div" line will be some width you did not anticipate (and not the width in your terminal window). It would be nice if there were a -Yvulpix mode to make it work.

[tgodzik]

Any idea if it's safely portable to LTS? Or is it even needed to be back ported?

[som-snytt]

@tgodzik I remember the code has history over the past 12 months, which suggests that it would be desirable on LTS. I can try it out and update this comment. (The delta is not trivial.)

[tgodzik]

I will try to backport it then if possible.

[som-snytt]

It cherry-picks, FWIW. The zip() error was introduced just a few days ago! Backporting is not fun: LTS doesn't have the LazyVals @nowarn to compile on JDK 23. Hats off to backporters.