Unknown `-language` option yields an error

[WojciechMazur]

Before 3.5 passing -language:higherKinds was simply ignored without any mesage, now, it's yielding a compilation error. This conflicts with handling of other unknown scalac options, which emit warnings.
The new behaviour is problematic when project use utilities to set common scalacOptions like https://github.com/typelevel/sbt-tpolecat - it requires to modify the upstream project and release it or to introduce a special exclusion rules in the build.

Compiler version

Last good release: 3.5.0-RC1-bin-20240512-99c4c00-NIGHTLY
First bad release: 3.5.0-RC1-bin-20240514-7c9aae3-NIGHTLY
Bisect points to: f6345c6

Minimized code

//> using options -language:higherKinds

@main def Test = ()

Output

[error] invalid choice(s) for -language: higherKinds
[info]   scalac -help  gives more information

Expectation

Unknown variant of option should yield warning instead of error.

[noti0na1]

-language is a choice flag now, so it errors out any invalid choice like other flags, for example, -source.

We should make a decision what behaviour we want for this kind of flags.

[som-snytt]

As a warning, bad option '-junk' was ignored is not very opinionated. -Yrangepos is the one I remember being ignored.

sbt-tpolecat is the unproblematic case: its job is to present correct options.

[Gedochao]

We should make a decision what behaviour we want for this kind of flags.

@noti0na1 agreed, we'll discuss this behaviour on the next core meeting.

[WojciechMazur]

In the last OpenCB there are 434 failures, seems like all of this failures are caused by -language:higherKinds or other no longer viable options https://virtuslab.github.io/community-build3/2024/05/18/3.5.0-rc1-bin-20240516-c608177-nightly.html
We can adapt OpenCB to try filter out the no longer valid options, but still, it seems like large amount of ecosystem was using no-op flag for a long time.

[Gedochao]

@noti0na1
Some notes after core meeting:

• we should accept options from Scala 2, including -language:higherKinds.
• for unknown options, it is correct to fail.

[noti0na1]

@Gedochao Do we also accept _ with no effect? Also, for Scala 2 language features, we just accept them without any warning?

[Gedochao]

@Gedochao Do we also accept _ with no effect?

@noti0na1 Nice catch.
No, we don't want to accept _.