Allow cancellation of computations forked with runAff

[sardonicpresence]

At present, in order to have a cancellable asynchronous computation it seems you forego both success and error callbacks in the call to forkAff, the only function to return a Canceler.

Particularly given that in the JavaScript runAff already returns a Canceler I saw no reason not to include a version returning it in PureScript.

In the process of adding a test for the above, I was inconvenienced by the inability to

do
  c <- runAff throwException (const (pure unit)) a
  launchAff $ cancel c (error msg)

due to the odd coersion of effect records in launchAff so I thought I'd propose a change there as well.

I'm assuming that the breaking nature of these changes will be contentious at least but thought I'd start here.

[jdegoes]

Why not modify this to return a canceller, to? User can always ignore a canceler if it's not necessary. Also this would allow you to delete the helper function runAff_.

[sardonicpresence]

No reason other than to minimise changes to the API. I certainly don't object to your suggestion.
Thanks for taking a look btw.

[jdegoes]

Since this will be included in a breaking release anyway (due to 0.9 updates), it's a good chance to clean up things like this. I'd rather a user void a return value they don't want, then not have the option (or multiply the surface area of the API).

[jdegoes]

@sardonicpresence Howdy there, waiting for your feedback above. Thanks for the PR, by the way!

[sardonicpresence]

Actioned review comments and re-based (hope that's reasonable etiquette).

[jdegoes]

These changes look great. This is the only one I'm concerned about. Technically, an EXCEPTION effect type in an Aff doesn't mean anything, since Aff doesn't throw exceptions in a way that can be caught via the same mechanism for Eff.

The intention is that you can lift any Eff with EXCEPTION into an Aff, and it drops the EXCEPTION; and that you can lower any Aff into an Eff and it adds EXCEPTION. That's not quite precise but I think it's probably the best we can do.

Is there some other way of addressing your need for this change?

[sardonicpresence]

To be frank I really struggle to see the utility of the effect record mechanism in Purescript generally.

I understand what you saying in regards to EXCEPTION not making sense in Aff, in fact this is the first instance I think I've encountered where rejecting certain effects makes sense like this. However runAff would still allow you to run an Aff with an EXCEPTION effect for example.

I think what you actually want is probably runAff to be able to declare that the effects in the Eff monads are a superset of those in the Aff monad but I don't believe that's currently possible in the language. I'm not sure what would be correct for the Canceler.

I hadn't thought of liftEff' as a kind of dual for launchAff; I'll have a think as to whether I can make use of it though I'm concerned by the use of Either as opposed to calling the error callback at a glance.

I'll have another stab at it tonight.

[sardonicpresence]

Alright I reverted the objectionable change, requiring further coercing in launchAff. If we could be more expressive about relationships between effect records I'd do differently but there we are.

I've also re-based once again.

[jdegoes]

@sardonicpresence Great work. Thanks! 🙏