Add test for t12237

[liufengyun]

Add test for t12237 to avoid regression

Original issue:
scala/bug#12237

[dwijnand]

Is this the right expected result? What does PathAndQuery./(_, _) mean? Because the first case is matching values that are PathAndQuery./. Similarly PathAndQuery.:&(_, PathAndQuery.===(_, _)) and PathAndQuery.+?(_, PathAndQuery.===(_, _)). I don't understand what the check file is asserting.

[liufengyun]

Yes, it's the correct expect file, PathAndQuery./(_, _) means that pattern is not exhaustive. The test is to defend against compiler hangs.

[dwijnand]

PathAndQuery./(_, _) means that pattern is not exhaustive.

I see. (Perhaps reporting PathAndQuery.:&(_, _) and PathAndQuery.+?(_, _) might be more uniform?)

I guess the Scala 3 checker doesn't try to give good counter-examples like Scala 2, which tries to do something like Root / "page" / x forSome x != "1"?

[liufengyun]

We would like to have refined counter-examples as in Scala 2. However, we could not figure out how to make it principled. For example, given the following example:

object Even {
  def unapply(x: Int): Option[Int] = ???
}

class Test {
  def foo(x: Option[Option[Int]]): Unit =
    x match {
      case Some(Some(5)) =>
      case Some(Some(Even(10))) =>
      case None =>
    }
}

It's unclear how to handle extractors in the warnings. Even if it's possible to be more principled, there's the worry that the message gets messy, thus difficult to read.

Meanwhile, even with refined counter-examples, users still need to check the existing patterns. Due to the considerations, we only show general counter examples.

[dwijnand]

Yeah, Scala 2's counter-example generation definitely doesn't handle extractors: it's all field based (for case classes) so it has no chance - it needs to be revisited. But it does a good job for case classes I say. It seems to be to be a good thing to generalise case class's unapply/extraction with the "product match" extractors, but it seems like discarding that knowledge for counter-example generation is lossy and thus a shame.

[nicolasstucki]

It is good as a regression test.

We still need to follow up on the discussion. @liufengyun should we open an issue for this?

[dwijnand]

The discussion is more by-the-by, not a blocker to landing this.

I'm not even sure if it's issue-worthy, as I'm not sure how valuable it ends up being to users to have the compiler construct precise counter-examples, like Scala 2 does.

[liufengyun]

I'm not even sure if it's issue-worthy, as I'm not sure how valuable it ends up being to users to have the compiler construct precise counter-examples, like Scala 2 does.

Thanks @dwijnand . Let's think about how to improve this together when there is a strong need for it.