Support for Scala 3.4.0

[Gedochao]

No description provided.

[Gedochao]

Loading Scala 3 binary from /Users/pchabelski/IdeaProjects/Ammonite2/out/amm/interp/api/3.4.0/compile.dest/classes/ammonite/$ivy/$.class. It should have been loaded from `.tasty` file. This `.tasty` file is missing. Try cleaning the project to fix this issue.

This seems 3.4.0-specific and pops up all over the place... 😕
Replicable for example with ammonite.integration.BasicTests.testIvySnapshotNoCache (and some other tests)

[Gedochao]

After some investigation, the failed tests for 3.4.0 seem to be caused by this:
scala/scala3#19702
As it seems identifiers containing $ will not be allowed long term, it may require fixes on Ammonite's side.

[Gedochao]

I don't know the codebase enough to attempt this, but I suppose we'd need to refactor Ammonite to get rid of identifiers containing the $ special character.
Otherwise, scala/scala3#19702 will keep haunting us.
@lefou @lolgab can you take a look and check if I'm maybe missing some simple workaround?

[lihaoyi]

I think the basic issue here is we allow import $ivy.blah, which desugars at runtime into import $ivy.$ and some other stuff. The "other stuff" is what's important, the import $ivy.$ is just a placeholder stub that does nothing

I think it should be possible to desugar import $ivy.blah into just whitespace . The import $ivy.$ (and similar) stubs don't actually do anything, so removing them should be ok

[Gedochao]

Corresponding Ammonite issue:

• Deprecate object $ to avoid unsoundess #1395

[Gedochao]

Hey, can someone help me with this?
Or should I extract 3.3.2 to a separate PR for now?
CI seems to be fine on 3.3.2, I guess, although I'd rather have it done in one go...
cc @lihaoyi @lefou

[lefou]

Yeah, I think splitting this up into a Scala 3.3.x update and an separate 3.4.x is the best way to make progress.

[Ichoran]

It's a little hard to tell what is still needed to get this working. All of my code depends on 3.4 features, and Jupyter needs Almond, and Almond depends on Ammonite, so...basically I can't use Jupyter notebooks until this gets fixed. I also depend on Java 21 (for Loom threads), and I notice that this PR stops testing even Java 17. Is this because the matrix is too big?

Anyway, I have time to help resolve the issue myself, but I can't tell right now what still needs doing. I can't see failing logs with the test, and I don't really want to dive in without knowing what I'm getting myself into.

[lihaoyi]

It seems the latest scala 3 added bakc support for $. Maybe it means we dont need any code changes and can just skip the problematic 3.4.x versions?

[Ichoran]

That might be the thing to do for now, but we can't rely on anything with a trailing $ for the reason Nicolas mentions in the ad-hoc patch (which is supposed to be temporary): scala/scala3#19705

But I think you're right that it's in. 3.4.2, I think, so maybe the thing to do is just go with that and skip 3.4.0 and 3.4.1.

(But at least the $-alone needs to be fixed eventually. I don't see why $ivy and such shouldn't work if the compiler people decide to permit it.)