Warn about encoded pkg obj names

[som-snytt]

Extend "encoded package name" warning to file package object names.

Avoid warning about "casual" code in the empty package.

Add -Yreporter as a testing convenience.

Fixes #22670

[som-snytt]

Practicing closing PRs for reworking instead of leaving them as drafts.

[som-snytt]

Related issue #22866 was scaladoc receiving names via tasty, where the package prefix and class name must be encoded (apparently because scaladoc would split on dot?). I don't remember the details, but it sounds ad hoc and perhaps wrong.

The scaladoc issue was the filename a.b.scala with name a.b and suffix .scala. (The use case was a.test.scala.) Then the package object uses a.b; the original ticket for the warning says tasty supplies an encoded package name, but compiler unencoded. Then maybe the warning for package object a.b or a-b should be guided by whether there is a corresponding package (under either name), in order to avoid warning about snippets.

A package

a.`b.c`.d

must preserve segments; the name is a.b$dotc.d.

If that isn't handled uniformly (correctly), then warning is the best effort.

[som-snytt]

status quo for hyphen in package name but not in file package object name:

-- Warning: /tmp/macro.scala:2:8 -----------------------------------------------
2 |package `X-Y` // explicit package name gets a diagnostic
  |        ^^^^^
  |The package name `X-Y` will be encoded on the classpath, and can lead to undefined behaviour.
1 warning found
-- Error: /tmp/usage.scala:4:8 -------------------------------------------------
4 |val x = foo // warn
  |        ^^^
  |Failed to evaluate macro.
  |  Caused by class java.lang.ClassNotFoundException: X-Y.macro$package$

So it's sufficient to add the warning for file package object name. It's desirable not to warn for snippets.

[som-snytt]

Added -Yno-reporter for convenience of SemanticbdTests, to reduce noise in the test console. Not sure if it should just use StoreReporter instead of Reporter.SilentReporter.

Added -Wconf to TestConfiguration.defaultOptions so that the new warning doesn't pop up in various tests in files with hyphens in the name. (The tests for the warning must reconfigure -Wconf to witness it.)

The rule is not to warn for package objects in the empty package, which is deemed "casual code". It's not obvious whether test code ought to meet that criterion.

[Gedochao]

added the release-notes label so we don't forget about mentioning -Yno-reporter, as the PR name/main desc doesn't mention it, cc @WojciechMazur

[som-snytt]

@Gedochao I'm not sure "private / forking" option needs to be advertised in release notes. Also, I was unsure whether to introduce an option that names a Reporter class (I think Scala 2 has -Xreporter), so I tried to be as private or "under the radar" as possible.

I'm available to make an adjustment per review. -Yno-reporter is not generally useful. I could change it to
-Yreporter:com.acme.MyReporter
as a way to introduce it without committing to long-term support.

Edit: that's kind of obvious, so I'll push -Yreporter before review, "for your consideration".

[Gedochao]

I'm not sure "private / forking" option needs to be advertised in release notes.

@som-snytt I guess generally speaking, it depends, since we do many things under -Y* 😅
Still, perhaps this is a no-op then

[noti0na1]

// errorNumber: 222

[som-snytt]

I was too quick to innovate. I'll revert it, but I was thinking that it would be nice if the compiler added the line comment for me, as a code action or quickfix. That is future work; and also verify that compilation does not reorder the existing enum.

[noti0na1]

    i"""Tools may not handle directories whose names differ from their corresponding package names.
       |For example, `p-q` is encoded as `p$$minusq` when written to the file system.
       |
       |Package objects derive their names from the file names, so files such as
       |`myfile.test.scala` or `myfile-test.scala` can produce encoded names for the generated package objects.
       |
       |In this case, the name `$name` is encoded as `${name.encode}`."""

[som-snytt]

Original comment:

Original ticket for warning is #14448

Doesn't yet address Cyclic macro dependencies.

Also packageObjectName is "forced" early to avoid cycles, even if there is no top-level def.

Also we write foo-bar.scala for descriptive snippets with top-level defs. Unreasonable to warn about that.

[noti0na1]

LGTM

[som-snytt]

param appears unused