Introduce -Xsource-features, for customizing the behavior of -Xsource:3 and -Xsource:3-cross

[som-snytt]

This feature is only recommended for projects that cross-build between Scala 2.13 and Scala 3. For migrating to Scala 3, use plain -Xsource:3. See also https://docs.scala-lang.org/scala3/guides/migration/tooling-scala2-xsource3.html.

This PR makes Scala 3 language features opt-in under an option "buffet" or "à la carte". To view the menu:

$> scalac -Xsource-features:help
Enable Scala 3 features under -Xsource:3.

Instead of `-Xsource-features:_`, it is recommended to enable specific features, for
example `-Xsource-features:v2.13.14,-case-companion-function` (-x to exclude x).
This way, new semantic changes in future Scala versions are not silently adopted;
new features can be enabled after auditing the corresponding migration warnings.

`-Xsource:3-cross` is a shorthand for `-Xsource:3 -Xsource-features:_`.

Features marked with [bin] affect the binary encoding. Enabling them in a project
with existing releases for Scala 2.13 can break binary compatibility.

Available features:

  case-apply-copy-access    Constructor modifiers are used for apply / copy methods of case classes. [bin]
  case-companion-function   Synthetic case companion objects no longer extend FunctionN. [bin]
  infer-override            Inferred type of member uses type of overridden member. [bin]
  any2stringadd             Implicit `any2stringadd` is never inferred.
  unicode-escapes-raw       Don't process unicode escapes in triple quoted strings and raw interpolations.
  string-context-scope      String interpolations always desugar to scala.StringContext.
  leading-infix             Leading infix operators continue the previous line.
  package-prefix-implicits  The package prefix p is no longer part of the implicit search scope for type p.A.
  implicit-resolution       Use Scala-3-style downwards comparisons for implicit search and overloading resolution (see github.com/scala/scala/pull/6037).
  v2.13.13                  case-apply-copy-access,case-companion-function,infer-override,any2stringadd,unicode-escapes-raw,string-context-scope,leading-infix,package-prefix-implicits.
  v2.13.14                  v2.13.13 plus implicit-resolution

-Xsource:3 -Xsource-features:_ (or -Xsource:3-cross for convenience to start a REPL) offers all "courses" as "prix fixe". It's not recommended in a build script, use v2.13.14 instead to get migration warnings as new features are adopted.

-Xmigration is no longer "overloaded" to mean -Wconf:cat=scala3-migration:w. That incantation must be written in full.

Fixes scala/bug#12961

[som-snytt]

To be squashed after review. Nominally (whatever that weaselly word means), I checked usages of isScala3... for reasonableness, but it's surprisingly difficult to verify. A clearer API would be more like hardMigration: either migrationWarning if isScala3, or discriminating progressiveWarning(at2 = old_behavior, at3 = warn, at3X = cross) where cross warns if verbose.

The -Ysource-options:case-access warning for apply is an unduly heavyweight mechanism for a niche feature or warning.

[SethTisue]

associated doc change PR scala/docs.scala-lang#2994

[lrytz]

Thank you for getting it rolling, I was going to pick it up this week.

I'm forming somewhat strong opinions about providing multiple ways to do the same thing. It takes time for people to learn / look up what all those variants mean.

• -Xsource:3 is what it is in 2.13.13, that's good
• -Ysource-options could work with -Xsource:3, we don't need -Xsource:3-compat for that
• -Xsource:3-cross is the same as -Ysource-options:_
• -Xsource:3-migration is the same as -Xsource:3 -Wconf:cat=scala3-migration:w
• I agree to stop the -Xmigration overload of course

Compiler flags are not code, they are only touched once in a while and don't need to be as succinct as possible.

The current proposal also doesn't really solve the auditing problem. If I'm using 3-compat and upgrade to a new Scala that adds some semantic change, how do I find out? I can go back to 3 but then the new warnings are mixed up with all the older warnings which I already addressed. So instead of 3-compat, I think it's better to add all semantic changes to -Ysource-options, not just the ones affecting binary encoding.

I'll give it a go today and push to this PR.

[som-snytt]

Happy to have you progress this experiment to your liking.

Here, you find out "what's new" by -Ysource-options:verbose (if you triggered it) or help (IIRC there was help text).

I think just having -Xnew-stuff:case-access etc is fine, in the xlint mode; I just wanted to see if we could avoid it, because it results in "fine-tuning". The experiment here was 3-compat allows for opt-in to dangerous options. So there was the conservative flag, the all-in flag, and the goldilocks flag.

Here, -Xsource:3-migration is just an alias for -Xsource:3, for clarity.

[lrytz]

I realize that apply insertion should be enabled by default, not even under -Xsource:3.

For a library built with -Xsoruce:3-cross, case class companions don't extend FunctionN. This should not break List(1).map(A) for users of the library that use the default compiler options.

@som-snytt wdyt?

[som-snytt]

The library would use 3-compat (or equivalent) to avoid incompatible changes.

The librarian, for the same reason, would not add object A companion that has the same effect; or would be sorry if they did.

But you may be right that this should be considered a "benign" feature change that everyone could benefit from. Fork all the language features! It is of general utility. The downside is your 2.12 build breaks; but who does that? I guess only middle-size fish who consume smaller fish and are consumed in turn.

[lrytz]

We might rename -Ysource-options

Also I could probably live with an alias Xsource:3-cross for -Ysource-options:_, it would be handy for testing stuff (not just for us, also for actual users of Ysource-options when they want to fire up a REPL).

[lrytz]

I squashed almost everything into one commit, only a few semantic changes as separate commits as good as possible.

[som-snytt]

extra strip (it would be interesting if that failed to compile)

[lrytz]

👍 IntelliJ adds it when typing sm"""<return>...

[som-snytt]

debug cruft. The expected needs .trim and assert could be

    import scala.tools.testkit.AssertUtil.assertSameElements
    assertSameElements(expected.linesIterator, Files.readAllLines(argfile).asScala)

The test is because -Vprint-args was broken; the -Xsource is not relevant.

I'll submit the tweak separately. :)

[som-snytt]

The buffet option makes it easier to see in the source where the features take effect.

It also simplifies the warnings under the rule of "no warning if they asked for a feature". (I just remembered that the point of verbose mode was to enable "audit", so you could see "this override was inferred differently", for example. Reverting to migration might show the future warning, but it might also just show an error?) (Probably no one would ever use verbose audit anyway.)

It took me two tries to correctly test moving the "case function" check from ident to adapt. Very nice!

[som-snytt]

LGTM. Lukas pulled that old github mind trick where he gets me to open the PR so that only he can approve it later. :)

I partially edited the OP comment, which is probably still inaccurate.

[som-snytt]

Your old trick of emitting the "how to suppress it" message would be helpful, especially when the category differs if not -Xsource:3.

https://discord.com/channels/632150470000902164/632150470000902166/1217725189018615848

[som-snytt]

travis claims

-t12919.scala:48: error: Implicit method myClassToSeq was found in a package prefix of the required type, which is not part of the implicit scope in Scala 3.

+t12919.scala:48: error: Implicit method myClassToSeq was found in a package prefix of the required type, which is not part of the implicit scope in Scala 3 (or with -Xsource-features:package-prefix-implicits).

[lrytz]

semantic merge conflict with #10706, rebased.

Your old trick of emitting the "how to suppress it" message would be helpful

Maybe somebody will eventually figure out how to do interactive verbose (not "re-run with ...").

➜ sandbox s -Wconf:any:wv
> scala repl -S 2.13 -cp . -Wconf:any:wv
Welcome to Scala 2.13.13 (OpenJDK 64-Bit Server VM, Java 21.0.1).
Type in expressions for evaluation. Or try :help.

scala> implicit val x = 41
                    ^
       warning: Implicit definition should have explicit type (inferred Int) [quickfixable]
       Applicable -Wconf / @nowarn filters for this warning: msg=<part of the message>, cat=other-implicit-type, site=x
val x: Int = 41

[SethTisue]

I'm noticing in the community build that despite this change:

-Xmigration is no longer "overloaded" to mean -Wconf:cat=scala3-migration:w. That incantation must be written in full.

I'm still seeing errors like:

[playframework] [error] Scala 3 migration messages are errors under -Xsource:3. Use -Wconf / @nowarn to filter them or add -Xmigration to demote them to warnings.

seems like an oversight that this advice wasn't updated?

[som-snytt]

I never read those messages all the way to the end.

Edit: until now. Tweak at #10718

[SethTisue]

a bit of good news: over in the community build, I removed -Xmigration from about a two dozen repos and only had to replace it with -Wconf:cat=scala3-migration:w in five:

proj/blaze.conf:    "appendScalacOptions -Wconf:cat=scala3-migration:w"
proj/json4s.conf:    "appendScalacOptions -Wconf:cat=scala3-migration:w"
proj/mima.conf:    "appendScalacOptions -Wconf:cat=scala3-migration:w"
proj/playframework.conf:    "appendScalacOptions -Wconf:cat=scala3-migration:w"
proj/scalafx.conf:    "appendScalacOptions -Wconf:cat=scala3-migration:w"