Map aliases

[ioleo]

This proposition consists of two parts:

1. Adding the as and asError operators.

Similar to *> and <* operators, sometimes we just don't care about the value in the success/error channel and wish to replace it. With these operators, we can express that more clearly.

val answerToLive1 = getStrLn map (_ => "42")
// vs
val answerToLive2 = getStrLn as "42"

2. Adding symbolic aliases for map, mapError, as and asError.

The symbolic aliases are more readable when chaining lots of operations.

val result1 = foo flatMap bar flatMap baz map wrap
// vs
val result2 = foo >>= bar >>= baz +> wrap
// or even

val result3 = (
  foo 
    >>= bar
    >>= baz
    +> wrap
)

The choice of symbols is somewhat arbitrary. Among many alternatives I propose these (+>, *+>, #>, *#>) follwing this reasoning:

• the right arrow > indicates the direction of the flow, similar to *> operator
• the + and # indicates the channel being mapped (respectively success and error)
• the * prefix indicates the previous value is being ignored, similar to *> operator

I'm open to discuss alternatives.

[jdegoes]

I am not opposed to as. Note that it serves the same function as const, but const is inherited from Haskell; as reads better with OO syntax.

To make this change, however, we would need:

1. Deprecate all the const methods
2. Add asUnit, and deprecate unit

Anyone else have thoughts on this?

[jdegoes]

Clarification: I am opposed to the symbolic aliases. Developers outside FP complain loudly about these, and unlike, e.g., >>=, there is no precedent for new operators.

[ghostdogpr]

I quite liked .unit. .asUnit is as long as .as(()).

[adamgfraser]

I like as. I agree that asUnit seems a little wordy for a common and simple method, maybe skip or just leave as is? I am thumbs down on the symbolic aliases for the reasons discussed above. If you want to write in that style you could could always do fa *> succeed(b) or fa *> fail(e).

[iravid]

👍 on as
👎 on asUnit; prefer to leave unit as is
👎 on additional symbolics, especially ones that aren't established

[jdegoes]

All right, so it looks like we're:

• 👍 to as, but we need deprecations for const, and we need as added everywhere const is used (and that is a lot of places!)
• 👎 to asUnit, leaving as(()) with the name unit for brevity
• 👎 to more symbolic names

[ioleo]

Alright, thanks for the input! I will adjust the PR as soon as I have some spare time :)

[ioleo]

Also for consistency (principle of least surprise):

• deprecated ZSchedule.const
• added ZSchedule.as
• deprecated ZSink.const
• added ZSink.as

[ioleo]

@jdegoes Ready to review. ci/circleci: test_212_jdk8_jvm failure is in BlockingSpec, I believe its a flaky test and the failure is unrelated to this PR.

[ghostdogpr]

There are some other const functions defined in ZManaged, Gen, Assertion, Exit, Fiber, FunctionIO, Runtime and STM. I think we better change them all at once?

[ioleo]

@ghostdogpr Added as and asError (where mapError was defined) to ZManaged, Assertion, Exit, Fiber, FunctionIO, Runtime and STM.

Also, reordered combinators as you asked.

[ghostdogpr]

Looks good! Just one remark left in Exit.

[ioleo]

@ghostdogpr Ready

[ghostdogpr]

@ioleo thanks for the suggestion & implementation!