Mention if missing class is on the class path

[som-snytt]

Hint if there is an object in scope, in a package, which has a class, and the class path finds it.

Fixes scala/bug#12289

[som-snytt]

java.io.IOException: No space left on device

[SethTisue]

I re-ran the Jenkins build, now that the community build runs that were using all the disk have finished.

I wonder sometimes if we should have a dedicated worker behemoth just for PR validation, separate from the behemoths used for the community build.

[som-snytt]

Needs refinement, as it does add verbiage for incremental compilation error.

[error] /home/amarki/snips/test-12289/src/main/scala/c.scala:4:17: not found: type T
[error] A definition on the class path may be hidden by a source artifact. (Companions must be defined together.)

because (as will surprise no one) the class is still there, i.e., previous artifacts are not cleaned first.

./target/scala-2.13/classes/p/Main.class
./target/scala-2.13/classes/p/T.class
./target/scala-2.13/classes/p/Main$.class
./target/scala-2.13/classes/p/C.class
./target/scala-2.13/classes/p/Main$delayedInit$body.class

[som-snytt]

The clever test is tweaked to allow -d.

The output check in typer may require tweaking: it uses single output only, and compares the path strings to mean "includes".

[lrytz]

It seems a helpful addition. Given

package p {
  object T
  class C extends T
}

and

package p {
  trait T
}

I assume IntelliJ would happily navigate from extends T to the definition of trait T, which makes it harder to figure why compilation fails.

[lrytz]

Maybe make it a DirectTest and use testOutput.path?

[lrytz]

The issue doesn't manifest the other way around, if name.isTermName?

[som-snytt]

I was too lazy to check after getting a first cut. Also, I considered it more likely to add an object as factory for an existing class.

[lrytz]

This I don't immediately understand; lookup for name failed (we're about to issue an error), but lookupSymbol(name.toTermName).owner.info.decls.lookupAll(name) gives a result. How is that?

[som-snytt]

The package is loaded from class path first, which creates symbols of existing class + its module & module class. Then compilation of source takes over the existing module symbol, which is entered into scope.

It's fishy that the package class has a set of members but not all are in scope. Previously I've read a comment about stale symbols from previous run; symbol history, like human history, is a story of wreckage and detritus. Somehow I had never fully considered the state of the environment during incremental compilation.

[lrytz]

Maybe we can find a message that's a bit more explicit, hands-on. Connecting the dots between a "definition hidden by a source artifact" and "companions" is not easy simple.