CI Test Errors FiniteDuration antisymmetry

[Slakah]

Some CI failures on Scala 3.

CI failure from #3727 :

Click to expand!

failing seed for order.antisymmetry eq is tJVxJ6pgwg7UrNv9sWAqwY4eokyiECpCf8K9kK8SQhC=
failing seed for order.antisymmetry eq is tJVxJ6pgwg7UrNv9sWAqwY4eokyiECpCf8K9kK8SQhC=
==> X cats.kernel.laws.Tests.Order[FiniteDuration]: order.antisymmetry eq  0.046s munit.FailException: /home/runner/work/cats/cats/kernel-laws/shared/src/test/scala/cats/kernel/laws/LawTests.scala:201
200:  checkAll("Order[Duration]", OrderTests[Duration].order)
201:  checkAll("Order[FiniteDuration]", OrderTests[FiniteDuration].order)
202:  checkAll("Order[UUID]", OrderTests[UUID].order)

CI failure from #3730 :

Click to expand!

failing seed for hash.antisymmetry eq is tJ9I1dkawrvX2DEc6NGtz5qNQw96EBLm1nV0-XvgFmA=
==> X cats.kernel.laws.Tests.Hash[FiniteDuration]: hash.antisymmetry eq  0.028s munit.FailException: /home/runner/work/cats/cats/kernel-laws/shared/src/test/scala/cats/kernel/laws/LawTests.scala:299
298:  checkAll("Hash[Duration]", HashTests[Duration].hash)
299:  checkAll("Hash[FiniteDuration]", HashTests[FiniteDuration].hash)
300:

Failing seed: 5MC5_eL2QrqZQkkWTTuITf7k9M0Wy1b5-RAvsNJZxHO=
You can reproduce this failure by adding the following override to your suite:

  override val scalaCheckInitialSeed = "5MC5_eL2QrqZQkkWTTuITf7k9M0Wy1b5-RAvsNJZxHO="

Falsified after 62 passed tests.
> Labels of failing property: 
Expected: true
Received: false
> ARG_0: -3888 hours
> ARG_1: -162 days
> ARG_2: org.scalacheck.GenArities$$Lambda$458/12131917@1f846122

[rossabaker]

Thanks for capturing these! Note that in both cases, the arguments are equal, but have different units:

@ 114.days === 2736.hours 
res12: Boolean = true

@ -162.days === -3888.hours 
res13: Boolean = true

That suggests a problem with the Cogen instance, whereupon we find #3470, and its reversion in #3694.

Does the plot thicken with Scala 3?

[larsrh]

@TimWSpence also observed this in #3762. I'm tempted to agree that either antiSymmetryEq should go, or the Cogen instance for FiniteDuration is borked.

[larsrh]

Fixed by #3772.

[TimWSpence]

@TimWSpence also observed this in #3762. I'm tempted to agree that either antiSymmetryEq should go, or the Cogen instance for FiniteDuration is borked.

Thanks for capturing these! Note that in both cases, the arguments are equal, but have different units:

@ 114.days === 2736.hours 
res12: Boolean = true

@ -162.days === -3888.hours 
res13: Boolean = true

That suggests a problem with the Cogen instance, whereupon we find #3470, and its reversion in #3694.

Does the plot thicken with Scala 3?

@rossabaker What difference does scala 3 make sorry? The law seems somewhat fundamentally broken to me - if you have a non-trivial Eq instance like FiniteDuration then you can write perfectly reasonable functions that aren't well defined wrt Eq (ie x == y does not imply f x == f y) eg by pattern-matching on the time unit like in the Cogen instance above