`-Yexplicit-nulls` no longer allows `eq` and `ne` comparsion, needs better errors message

[WojciechMazur]

Based on the OpenCB failure in stg-tud/bismuth - build logs

Compiler version

Last good release: 3.8.0-RC1-bin-20250817-8c3f1a6-NIGHTLY
First bad release: 3.8.0-RC1-bin-20250818-aaa39c5-NIGHTLY
Bisect points to aaa39c5 (seems unrelated)

Minimized code

//> using options -Yexplicit-nulls

trait Key[T]
trait Bucket[T]:
  def key: Key[T]
  
def Test = 
  val bucket: Option[Bucket[String]] = ???
  val key = bucket.fold[Key[String] | Null](null)(_.key)
  if (key ne null) then 
    println("not null")
        

Output

[error] ./test.scala:10:7
[error] expression does not take parameters
[error]   if (key ne null) then 
[error]       ^^^^^^^^^^^

Expectation

Should either compile or the error message should suggest using == / != operators

[noti0na1]

Em... The error is from inline phase, but doesn't make sense to me.

I checked the diff between the two releases: 8c3f1a6...aaa39c5

Nothing seems related...

[olhotak]

Em... The error is from inline phase, but doesn't make sense to me.

It could be related to the ne extension method in Predef, which is inline.

[noti0na1]

They are defined as inline from beginning. The recent change we made is marking them as infix (#23255), but that was in May.

[hamzaremmal]

I'm aware of this issue and it had to do with the fact that the extension methods in Predef will not be chosen anymore because the stdlib for now is not compiled with explicit nulls.

[noti0na1]

At least after re-enabling explicit nulls tests in #23566, all tests pass.

[jchyb]

I've made a non-regressive stdlib-agnostic (to a degree) version of this issue so we can still track it after we start shipping stdlib nightlies with -Yexplicit-nulls: #24150