StrawmanTest.mainTest failing in Scala 2.13.0-M3

[SethTisue]

[collection-strawman] [error] Test strawman.collection.test.StrawmanTest.mainTest failed: java.lang.AssertionError: assertion failed, took 0.472 sec
[collection-strawman] [error]     at scala.Predef$.assert(Predef.scala:204)
[collection-strawman] [error]     at strawman.collection.test.StrawmanTest.lazyListOps(Test.scala:466)
[collection-strawman] [error]     at strawman.collection.test.StrawmanTest.mainTest(Test.scala:576)
[collection-strawman] [error]     ...

details at https://scala-ci.typesafe.com/view/scala-2.13.x/job/scala-2.13.x-integrate-community-build/856/consoleFull

doesn't seem to be intermittent (failed in both runs last night)

[SethTisue]

or, hmm, maybe it is intermittent, there is a run currently in progress where collection-strawman already went green: https://scala-ci.typesafe.com/view/scala-2.13.x/job/scala-2.13.x-integrate-community-build/857/consoleFull

we'll see how the next few runs go

[julienrf]

Did you change something about the evaluation of by-name right-associative operators?

[julienrf]

Yeah now I remember that one: scala/scala#5969

But the test should always fail with the new evaluation model.

[SethTisue]

I failed to notice earlier that run 857 was manually triggered by Miles against a different 2.13 build.

So yes, always failing.

[julienrf]

Hey @SethTisue! What should we do about that? Is this blocking you?

[SethTisue]

it's not blocking me.

I don't understand the issue, I don't know what you should do. but in general, this repo ought to compile, and tests pass, on whatever the latest 2.13 nightly is, since before long some descendant of that nightly will become 2.13.0-M3. (@szeiger?)

[julienrf]

The issue is that we have a test that checks the semantics of an operation (#:: on Stream LazyList) and that this semantics has changed because the language has changed. I would be in favor of changing the test so that it takes the new language semantics into account. However, the fact that the semantics of #:: will have changed between 2.12 and 2.13 might be a problem for existing users, though I doubt someone relies on the fact that this operator is not really lazy.

[SethTisue]

I would be in favor of changing the test so that it takes the new language semantics into account

👍

[julienrf]

@szeiger I tried to make #:: a proper method on LazyList (rather than an extension method), like you did in scala/scala#5969 but then I noticed that it was not anymore possible to define lazy lists via recursive lazy vals:

lazy val fibs: LazyList[Int] = 0 #:: 1 #:: fibs.zip(fibs.tail).map { case (prevPrev, prev) => prev + prevPrev }

We get an infinite recursion. Is that expected?

Note that on one hand, @odersky thinks that recursive lazy vals should not be supported (scala/scala3#1892 (comment)). On the other hand, being able to define the fibonacci sequence as shown above would be useful, in my opinion.

[SethTisue]

here's an example recursive lazy val Stream in my own code:

  implicit class RichStream[T](private val s: Stream[T]) extends AnyVal {
    // if we put the lazy val at the top level, we can't extend AnyVal,
    // so we make it local, like this:
    def circular: Stream[T] = {
      lazy val result: Stream[T] = s #::: result
      result
    }
}

the use of lazy val gives a true cycle (infinite if traversed, but occupying constant space in memory). still a niche use case, I guess, and wouldn't be horrible with a local var :-\