WUnused: Ignore unused params of non-private methods that are members of non-final classlikes

[szymon-rd]

Compiler version

3.3.0-RC2

The bug

We should not report unused params in methods that may be overridden, as it's often for the sake of the implementers to use these arguments.

[odersky]

It's worse than that.

trait A:
  def f(x: Int): Int
final class B extends A:
  def f(x: Int) = 22
final class C extends A:
  def f(x: Int) = x

f in B cannot be overridden, but the parameter should still not be flagged. So the best one can do is to flag unused parameters in local methods or in private methods or in final methods that do not override or implement anything else. At this point it becomes dubious whether the warning makes sense at all, IMO.

[som-snytt]

➜  ~ scala -Wunused
Welcome to Scala 2.13.10 (OpenJDK 64-Bit Server VM, Java 19).
Type in expressions for evaluation. Or try :help.

scala> def f(x: Int) = 42
def f(x: Int): Int

scala> def f(x: Int) = println(42)
             ^
       warning: parameter value x in method f is never used
def f(x: Int): Unit

Scala 2 tries to guess whether the RHS is trivial, as a gauge of whether it qualifies as a serious implementation or a placeholder or a stub. There is an open ticket as to whether final is a useful factor; sealed is not considered, where you might say a parameter is used somewhere at least.

I've proposed a -Xlint:non-strict to enable these heuristics to reduce warnings; normal -Xlint would enable reduced warnings by default, but it could be turned off for "strict mode", as some people prefer (or profess to prefer).

Otherwise, as was originally judged when the feature was proposed, a lint for unused parameters is uselessly noisy. I think that was Jason's comment, if it needed to be articulated.

Ideally, just in those cases where it looks like a param is not used to calculate a result, the user must add @unused to communicate that fact. (They are communicating with the reader, not only the compiler.)

[odersky]

I think we should follow Jason and accept that some warnings simply don't make sense.

[szymon-rd]

For now (3.3.0 release), I think we should proceed with not reporting the unused parameters for non-private defs at all.