Document cancelation boundaries

[TimWSpence]

No description provided.

[TimWSpence]

I was trying to justify 3 (boundary after poll is not a cancelation boundary) from the laws but failed to do so. What have I missed?

Also, I wasn't sure if this belonged in scaladoc or mdoc docs or both?

[vasilmkd]

I like it a lot.

[vasilmkd]

I think we'll definitely need some more on the website but this is a great start.

[TimWSpence]

I think we'll definitely need some more on the website but this is a great start.

Yeah, do you think I should directly copy this stuff to the website? Or is it too much detail for that? Still find that balance hard to judge

[djspiewak]

I quite like this, though one downside to it is the text in question is somewhat IO-specific. Other implementations of the typeclasses (notably, Monix) are likely to have different semantics. Would it be better to push this to the website or the IO scaladoc rather than putting it in MonadCancel? Otherwise looks very good!

[TimWSpence]

I quite like this, though one downside to it is the text in question is somewhat IO-specific. Other implementations of the typeclasses (notably, Monix) are likely to have different semantics. Would it be better to push this to the website or the IO scaladoc rather than putting it in MonadCancel? Otherwise looks very good!

Yeah I was worried about this as well. I'll see if I can find a place to move it to. The uncancelable boundary at the end of uncancelable is guaranteed by the laws though, right? Is the uncancelable boudnary at the end of poll(foo (assuming cancelation isn't observed during foo) also guaranteed by law? I didn't manage to prove it

[djspiewak]

The uncancelable boundary at the end of uncancelable is guaranteed by the laws though, right?

Meaning that uncancelable is guaranteed to not check cancelation at the termination? That is in fact guaranteed by laws:

  def onCancelAssociatesOverUncancelableBoundary[A](fa: F[A], fin: F[Unit]) =
    F.uncancelable(_ => F.onCancel(fa, fin)) <-> F.onCancel(F.uncancelable(_ => fa), fin)

Is the uncancelable boudnary at the end of poll(foo (assuming cancelation isn't observed during foo) also guaranteed by law?

This one is a bit trickier. I think this is guaranteed by the above law composed with the following:

  def uncancelablePollIsIdentity[A](fa: F[A]) =
    F.uncancelable(_(fa)) <-> fa

It's tricky though. Both are definitely pervasively assumed properties in polymorphic code.