Use TestControl in tests

[danicheg]

This replaces custom deterministic runtime and uses new TestControl from cats-effect in tests.
@SystemFw your thoughts on this would be so much appreciated.

[mpilquist]

Looks like the CE3 upgrade has caused one of the io tests to fail on Scala.js /cc @armanbilge

[armanbilge]

I proposed a fix in #2651 (comment).

[danicheg]

I can apply the patch 843ff4a if we agree on it. Speaking for myself - I see no problem with it.

[mpilquist]

I don't think that's the same thing. I'm talking about the CI failure:

==> X fs2.io.IoPlatformSuite.toDuplexAndRead 0.09s munit.FailException: Failing seed: RUB4zd5KUZxR7R91LcjD6YarbRlLs9ray67RY2BBenL=
5861
You can reproduce this failure by adding the following override to your suite:
5862

5863
  override def scalaCheckInitialSeed = "RUB4zd5KUZxR7R91LcjD6YarbRlLs9ray67RY2BBenL="
5864

5865
Exception raised on property evaluation.
5866
> ARG_0: Stream(..)
5867
> ARG_1: Stream(..)
5868
> Exception: fs2.io.StreamDestroyedException: null

[armanbilge]

Woops, commented without looking :) I will investigate that.

[mpilquist]

@danicheg Sounds good, that patch looks good to me too.

[danicheg]

Unfortunately, the error on JS preserves. It seems some changes in main after df62117 are breaking with CE 3.3.0 (CI was green at df62117). /cc @armanbilge

[armanbilge]

Yeah, I worked on it a bit but confounded at the moment, I'll try again later. @danicheg if you have bandwidth to bisect CE3 to figure out what caused this here that would be ✨

[armanbilge]

I bisected down to typelevel/cats-effect#2411 about cancellation, which seems plausible since the error is StreamDestroyedException and the cancellation finalizer here destroys the Node.js stream. So now I get to do /cc @vasilmkd 😉 for any pointers about your PR, please and thank you! :)

[djspiewak]

@armanbilge That PR actually does quite a bit. Tldr there was a bug in IOFiber where cancelation recreated the runloop and failed to shift. This PR, together with an additional one, fixed that issue. I don't suppose it's possible to bisect further? The behavior as it stands should be correct, and was incorrect previously, but I can't imagine how you could have been depending on the broken behavior.

[armanbilge]

@djspiewak that was my first instinct to bisect further, but that PR is a single commit.

[djspiewak]

Since this is JS, is it at all possible that you were somehow relying on the fact that cancelation did not shift? Like, you could call cancel and it would remain on the current thread, even across multiple finalizers, if it didn't happen to hit an async boundary.

[armanbilge]

I must have? Certainly not intentionally, but seems plausible. I need to refresh myself with this code.

[armanbilge]

@danicheg I have another patch for you.

diff --git a/io/js/src/main/scala/fs2/io/ioplatform.scala b/io/js/src/main/scala/fs2/io/ioplatform.scala
index 57b2df3f..9dab4d6a 100644
--- a/io/js/src/main/scala/fs2/io/ioplatform.scala
+++ b/io/js/src/main/scala/fs2/io/ioplatform.scala
@@ -209,6 +209,7 @@ private[fs2] trait ioplatform {
           new streamMod.Duplex(
             streamMod
               .DuplexOptions()
+              .setAutoDestroy(false)
               .setRead { (duplex, _) =>
                 val readable = duplex.asInstanceOf[streamMod.Readable]
                 dispatcher.unsafeRunAndForget(

No idea why the CE3 change exposed this.

[danicheg]

@armanbilge Thanks! It seems failed test is passing well locally.