add JDK 21 Early Access to nightly CI

[SethTisue]

prompted by scala/bug#12783

[SethTisue]

I had hoped that this would be as easy adding 21-ea, but nooooooooooo, Error: Could not find satisfied version for SemVer '21'

[SethTisue]

okay, I got the CI config correct I think

but the new jobs are now failing because of scala/bug#12783 (which is the root cause of sbt/sbt#7235)

and note that it fails trying to even start sbt

so, eventually we do want to merge this PR, but we won't able to use this PR's setup to test the fix for 12783, because of the circularity involved

we'll need to make the fix, locally publish a Scala with the fix (2.12.18-M2, say), then test that it fixes sbt/sbt#7235 by starting sbt with that Scala version (will we need to build a new sbt, or can we tell the launcher directly to use 2.12.18-M2?). if that succeeds, then we're good

@eed3si9n if I have 2.12.18-M2 published locally, can I tell sbt to use it without having to rebuild sbt?

[eed3si9n]

@eed3si9n if I have 2.12.18-M2 published locally, can I tell sbt to use it without having to rebuild sbt?

You could try something like launch

sbt -Dsbt.scala.version=2.12.18-M2

with potentially needing project/plugins.sbt:

// Global / resolvers += "scala-integration" at "https://scala-ci.typesafe.com/artifactory/scala-integration/"
def scala212 = "2.12.18-M2"
scalaVersion := scala212
dependencyOverrides += "org.scala-lang" % "scala-library" % scala212
dependencyOverrides += "org.scala-lang" % "scala-compiler" % scala212

[SethTisue]

after #10397 goes through, this one can come next

[SethTisue]

in my local testing on top of #10397, I'm seeing some test failures:

!!  665 - run/repl-no-imports-no-predef.scala       [output differs]
!!  672 - run/repl-parens.scala                     [output differs]
!!  699 - run/sammy_after_implicit_view.scala       [output differs]
!!  711 - run/sammy_restrictions_LMF.scala          [output differs]
!! 1295 - run/t5535.scala                           [output differs]
!! 1338 - run/t5789.scala                           [output differs]
!! 1465 - run/t6434.scala                           [output differs]
!! 1643 - run/t7747-repl.scala                      [output differs]
!!  141 - neg/macro-exception                       [output differs]
!!  150 - neg/macro-invalidret                      [output differs]

a number of tests are sensitive to a toString change for lambdas, resulting in diffs like:

-res17: () => Int = <function>
+res17: () => Int = $Lambda/0x000000080177c430@52449bdb

the failing neg tests are because of test sensitivity to exact stack trace contents:

-   at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:104)
+   at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:103)

and run/sammy_restrictions_LMF fails with an exception, cause not immediately obvious. I'll look into it later

[lrytz]

toString change for lambdas

Probably needs an adjustment here:

scala/src/partest/scala/tools/partest/ReplTest.scala

Lines 78 to 79 in 418f76d

	private val lambdaless = """\$Lambda\$\d+/(?:0x[a-f0-9]{16}|\d+)(@[a-fA-F0-9]+)?""".r
	private def stripLambdaClassName(s: String): String = lambdaless.replaceAllIn(s, Regex.quoteReplacement("<function>"))

[som-snytt]

  private val lambdaless = """\$Lambda\$\d+/(?:0x[a-f0-9]{16}|\d+)(@[a-fA-F0-9]+)?""".r
  private val lambdaless = """\$Lambda(?:\$\d+)?/(?:0x[a-f0-9]{16}|\d+)(@[a-fA-F0-9]+)?""".r

optional bit after $Lambda, probably I should test it if I'm going to say it.

OK also make that last group noncapturing.

val lambdaless = """\$Lambda(?:\$\d+)?/(?:0x[a-f0-9]{16}|\d+)(?:@[a-fA-F0-9]+)?""".r

[som-snytt]

@SethTisue thanks for early action on EA. I pushed a tweak. I feel so nerdy pushing to 2.12.

[SethTisue]

rebased as #10400