Prefer type of overridden member when inferring (under -Xsource:3)

[som-snytt]

Under -Xsource:3, adopt the new rule that the type of an override is inferred from the overridden member, and not from the right-hand side of the overriding element, which may narrow the type unexpectedly. The type can be narrowed by supplying an explicit type.

Fixes scala/bug#7212

[SethTisue]

CI likes it now. Is it ready for review? I'm in favor of the change, given that it's behind -Xsource:3.

[som-snytt]

I don't remember what I wasn't sure about. If it breaks something, we might finally get a sense of who is actually using -Xsource:3 or, as Belle says in the current movie, How can we find one person among 5 billion?

[SethTisue]

Mind supplying a nice user-facing PR description?

[SethTisue]

Final review by @lrytz, please — I just want to be sure you're comfortable with the change.

If merged, this increases the pressure on us to produce a really nice doc page that documents everything that -Xsource:3 changes.

[lrytz]

👍 LGTM

[SethTisue]

we might finally get a sense of who is actually using -Xsource:3

yup, that is now happening :-)

e.g. scala/bug#12645 , and also:

https://twitter.com/not_xuwei_k/status/1572987057621929984

hold up. given:

• you want to keep bincompat in both Scala 2 and 3
• forgot a type ascription
• already released using "-Xsource:3"

since Scala 2 and 3 are already published on different types, it would be super difficult to bump to Scala 2.13.9
(what am I gonna do with json4s?)

bumping to 2.13.9 without type ascriptions, mima yells at me for changing return types in Scala 2, and if I did add it mima yells at me for changing return types in Scala 3

there's non-zero chance that dropping "-Xsource:3" would work, but likely a checkmate for bincompat anyway

[SethTisue]

@xuwei-k perhaps you'll need to resort to version-specific sources. either duplicate the entire file, or if that's too much duplication, make the return type a type alias and define the type alias differently in version-specific sources

[som-snytt]

Minimally, it ought to warn under -Xmigration. The purpose of -Xsource;3 is to aid in migration to 3, as well as leverage Scala 3ish features. The other example was implicit lookup, a major feature, which was made opt-in (because it wasn't quite ready for prime time). Maybe there's another mechanism to make it easier to use.

The other footnote is that they say don't rely on inferred types for API.

[SethTisue]

I'm curious when/where this design decision was made on the Scala 3 side — does anyone know of tickets, discussions, etc?

[SethTisue]

there is some talk now on scala/bug#12671 that this may need to be refined further to match the Scala 3 behavior more closely

[hmemcpy]

Sorry to necromance this. I just tried to upgrade our 2.13.6 project straight to 2.13.10 (we're using -Xsource:3), and I have some errors relating to this issue, as well as (likely) #10012. I'm still trying to work around it. If those are bugs, I'll try to minimize and report them.

[som-snytt]

@hmemcpy Thanks for the upgrade effort. It's not necessary to announce on an old PR. Since twitter is moribund, there is discord or the contributors forum to open a topic about this upgrade challenge. A fresh ticket for bugs is welcome, of course.