Allow `IO.onError` to be passed a total function

[Grogs]

#4121 deprecated IO.onError(f: Throwable => IO[Unit]) and added def onError(pf: PartialFunction[Throwable, IO[Unit]]) to be used instead.

Normally total functions can be passed in place of a partial function and Scala will automatically create a partial function wrapper for it. In other words, we should be able to call the partial function variant of onError with a total function. Here's a scastie snippet showing that expected behaviour of onError with Resource.

Unfortunately, when calling IO.onError with a total function in CE 3.6.0, there is deprecation warning. This is inconsistent with earlier versions or the behaviour of ApplicativeError.onError.

It'd be great to make IO.onError not generate a deprecation warning when passed a total function.

Proposed fix from discord is to mark the deprecated version of onError as package private.

[CJSmith-0141]

I think that this bug should get fixed, but for what it's worth, I used this deprecation as the motivation to start using NonFatal

effect.onError:
  case NonFatal(e) => someMetrics(e)

[armanbilge]

I used this deprecation as the motivation to start using NonFatal

@CJSmith-0141 IO already special-cases fatal errors internally (by shutting down the runtime as safely as possible), so you'll actually never observe a fatal error in onError (if you do, that's also a bug lol).

[CJSmith-0141]

I used this deprecation as the motivation to start using NonFatal

@CJSmith-0141 IO already special-cases fatal errors internally (by shutting down the runtime as safely as possible), so you'll actually never observe a fatal error in onError (if you do, that's also a bug lol).

So it's not needed, good to know. I have never observed this, and now I know why 🙂

[joroKr21]

Also for the people stuck on 2.12, having an overload means the argument type can't be inferred (which is much more inconvenient for the partial function overload). Hopefully making it package private will restore the type inference, but it would be good to check.

[CJSmith-0141]

Can this issue be closed now that 3.6.1 was cut?