Re: [GHC] #12001: RFC: Add pattern synonyms to base

2 May 2016

      #12001: RFC: Add pattern synonyms to base
-------------------------------------+-------------------------------------
        Reporter:  Iceland_jack      |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  libraries/base    |              Version:  7.10.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by Iceland_jack):

 Great responses, there are obviously no best practices yet for this
 extension.

 As the dust settles, will we only accept ''constructor-like'' patterns?
 Are non-constructor-like patterns wholly undesirable or are there examples
 where they are worth it?
 I personally use all of these to good effect, let's wait until we
 understand them better.

 Let's identify and categorise some of these examples:

 === Literal synonyms ===
 Bidirectional patterns, nothing fancy

 {{{#!hs
 pattern I a  = Identity a
 pattern K a  = Const a
 pattern Less = LT
 }}}

 === ‘Getter’ patterns ===
 Unidirectional patterns, these are essentially ‘Getter’s

 {{{#!hs
 pattern Real      r <- r :+ _
 pattern Imaginary i <- _ :+ i
 }}}

 === Constructor-Like Patterns ===
 (Implicitly) bidirectional patterns that don't lose information

 {{{#!hs
 pattern Real, Imaginary :: (Num a, Eq a) => a -> Complex
 pattern Real      r = r :+ 0
 (Implicitly)
 pattern Imaginary i = 0 :+ i
 }}}

 In this case this is the identity function on complex numbers:

 {{{#!hs
 f :: (Num a, Eq a) => Complex a -> Complex a
 f (Real r) = Real r
 f complex  = complex
 }}}

 @ekmett argues for this property
 [https://github.com/ekmett/lens/issues/653 here].

 === Non-Constructor-Like Patterns ===
 Explicitly bidirectional patterns that do lose information

 {{{#!hs
 pattern Real, Imaginary :: Num a => a -> Complex a
 pattern Real r <- r :+ _
   where Real r =  r :+ 0

 pattern Imaginary i <- _ :+ i
   where Imaginary i =  0 :+ i
 }}}

 We lose the constructor property, but matching never fails so it's closer
 to the lens `_realPart`.

 ----

 We can keep this ticket open until the extension and understanding
 matures, same as #11349 where
 [https://github.com/RyanGlScott/proxied/blob/master/src/Data/Proxyless.hs
 Data.Proxyless] was created in response. It has been said of me that I
 embrace pattern synonyms to “almost an absurd level” so someone needs to
 show restraint `:--)` tickets such as #11977 show that this extension
 still isn't even fully understood and I look forward to seeing how it
 evolves

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/12001#comment:5
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler