Oh wait, why add a feature that is deprecated right away? 🤔

Not sure about knobs for enablement yet, but Scala 3 warns about apply insertion. It's strictly a migration mitigation.

blocked by #10573

Not sure about knobs for enablement yet, but Scala 3 warns about apply insertion. It's strictly a migration mitigation.

C.apply works both in Scala 2 and Scala 3 right?

@joroKr21 iirc the tupled case is not handled because scala 2 doesn't do parameter untupling adaptation. I'll try to nail it down while polishing.

The other missing parts are compose & andThen, so the adaptation here is incomplete in that sense. So really, we're just trying to make it easier for those slick users. I used slick once, do I have to try it again to test out the use case? 😿

@lrytz about to contribute a bit of doc for the incompatibility matrix as you pointed out.

I haven't tried it, but maybe it makes sense to support case class D(i: Int, j: Int); List(42->27).map(D) directly, by injection of tupled, where the syntax is the same as Scala 3 but the mechanism is slightly different.

It was interesting that the recommended idiom is what is implemented here. (Foo.apply _).curried(1)(true)

maybe it makes sense to support case class D(i: Int, j: Int); List(42->27).map(D) directly

Since this doesn't currently work on Scala 2, and only works in Scala 3 with a warning and is therefore discouraged, I don't think this is needed.

My case for tupled injection is cross-compilation. For the tupled case, it requires different sources because D.tupled is not supported by Scala 3. I haven't tried slick use case yet, but it's the tupled case they care about. I assume slick supports Scala 3.

Since this doesn't currently work on Scala 2, and only works in Scala 3 with a warning and is therefore discouraged, I don't think this is needed.

Also, as @joroKr21 pointed out, this is true of everything in this PR. It has limited shelf life, as was true for the Scala 3 feature when it was introduced recently.

OK, good point

Not sure I'll get around to tupled.

@lrytz added a commit for "insert apply with tupling adaptation". It's a bit tricky (subtle), so it is unsquashed.

I re-enabled the lrytz maneuver for your convenience. (Feel free to push here again.)

Perhaps there is better API usage for the simple questions in adaptApplyInsertion, such as "does it have an apply method and does its arity match the expected type or the tuple param", etc.

Slightly augmented tests. My takeaway vis-a-vis TDD is that for a code change of a certain complexity, discipline is required to add tests for intersecting cases. I don't evince that discipline here. A nice IDE tool would be a test generator that probes feature interactions, scalacheck + fuzzing + chatgpt. It's nice to have the community build, but we could use the corpus to generate targeted tests (because we know how people use case classes and tupled, for instance). Oh, I just thought of another test.

Edit: I checked that abstract case class (so there is no apply) shows the same error as arity mismatch, found C.type. In fact, without this commit it SOs. I checked out the parent commit and tried it standalone, it errors correctly.

Rebased and added support for case class E() (which is adapted by dotty).

The first typecheck of C.apply uses pt (so E works). On error, use wildcard (so tupled works).

Note: using wildcard first emits spurious warning about "empty application" of E(). Even though the result is discarded by us, the successful result makes silent emit the warnings. (Possibly you could typer.context.reporter.clearAll() before returning from the silent op, but it's clunky.)

Also added test for overloaded apply. Instead of erroring nicely, it just works because expected type is used first.

/rebuild

wasn't sure if /rebuild does older commits (the ones that the "combined" check is complaining about), but I see at https://scala-ci.typesafe.com/job/scala-2.13.x-validate-main/ that it does

@som-snytt can you do a round of making the PR description user-facing? I think it would be valuable to include an example, and to explicitly say what the behavior is on -Xsource:3 vs -Xsource:3-cross vs neither.

I'll try to nail it down while polishing.

The pun was 💅 .

9 months and noone noticed...