Import selectors apply API methods allow to unsafely reference symbols via Strings

[jchyb]

Compiler version

main

Expectation

Import selectors apply API methods allow to reference symbols via their names, which is unsafe and incongruent to the rest of the API (where we reference things via Symbols. We should not pass Strings directly to those apply methods, instead we should rely on Symbols

See: #22457

[bishabosha]

yes this still makes it unclear what is the purpose of this feature - e.g. the import given wildcard - now you would need to pre implicit search those symbols in order to import them?

[jchyb]

@bishabosha Sorry about the late reply. The use case is being able to dynamically choose what gets imported for the delayed implicit search (like with summonInline), done after inlining. Also it might streamline even simple static additions to implicit scope, like this:

`{
  import something._
  summonInline[T]
}

when creating trees via reflection without having to suddenly switch to quoted code like the above. They will affect only the implicit searches done after inlining, so no additional handling from the compiler should be necessary. SimpleSelectors and RenameSelectors are useless though.

As for this issue itself, I think I panicked before, the current apply methods do match the previously added stable unapply ones. Also, we of course do use Strings directly at times, like in Symbol.unique, or even Symbol.classSymbol, etc.. With how ImportSelectors are implemented in the compiler (and exposed in the reflection api, with the prefix), it really would not make sense to pass symbols there. We might benefit from adding some checks for users whether a passed name exists, as currently incorrect passed String will just make the OmitSelector silently fail (so no crash or anything like that). I’m leaning towards closing this, but I’ ll leave it up for discussion for now.

[som-snytt]

I'm also skeptical about the proposal, but I understand the use case. Maybe the feature is "programmatic unused import warning".

[jchyb]

If that becomes an issue, we can fix it
Edit: by fixing I mean disabling it for the reflection generated Import expressions. I would not want to add more work for anyone of course (only now realized what the concern was).

[jchyb]

Closing the issue. As I mentioned, putting symbols into those apply method does not really make sense, and the current implementation/api seems fine. Those import selectors applies are experimental, so they can still be changed/removed if anything concrete comes up.