[GHC] #13646: strict patterns with no bindings (e.g. `let !() = ...`) trigger -Wunused-pattern-binds

4 May 2017


      #13646: strict patterns with no bindings (e.g. `let !() = ...`) trigger -Wunused-
pattern-binds
-------------------------------------+-------------------------------------
           Reporter:  exphp          |             Owner:  (none)
               Type:  bug            |            Status:  new
           Priority:  normal         |         Milestone:
          Component:  Compiler       |           Version:  8.0.2
           Keywords:                 |  Operating System:  Unknown/Multiple
       Architecture:                 |   Type of failure:  None/Unknown
  Unknown/Multiple                   |
          Test Case:                 |        Blocked By:
           Blocking:                 |   Related Tickets:
Differential Rev(s):                 |         Wiki Page:
-------------------------------------+-------------------------------------
 In each of the following programs, a bang pattern is used to force
 evaluation of a bottom, the intent being to perform fail-fast error
 checking.  Unfortunately, however, they also generate `-Wunused-pattern-
 binds`.

 {{{#!hs
 {-# OPTIONS_GHC -Wall #-}
 {-# LANGUAGE BangPatterns #-}

 main :: IO ()
 main = do
     let !Nothing = Just ()
     pure ()
 }}}

 {{{#!hs
 {-# OPTIONS_GHC -Wall #-}
 {-# LANGUAGE BangPatterns #-}

 import Control.Exception

 main :: IO ()
 main = do
     let !() = assert False ()
     pure ()
 }}}
 {{{
 src/Lib.hs:6:9: warning: [-Wunused-pattern-binds]
     This pattern-binding binds no variables: !Nothing = Just ()
 Linking src/Lib ...
 Lib: src/Lib.hs:6:9-26: Irrefutable pattern failed for pattern Nothing

 --------------------

 src/Lib.hs:8:9: warning: [-Wunused-pattern-binds]
     This pattern-binding binds no variables: !() = assert False ()
 Linking src/Lib ...
 Lib: Assertion failed
 CallStack (from HasCallStack):
   assert, called at src/Lib.hs:8:15 in main:Main
 }}}
 For clarity, non-monadic `let ... in` patterns are also affected; I only
 gloss over them because there tends to be other equally ergonomic
 alternatives in such cases.

 I found this #9127 (ticket), where a patch was accepted to allow wildcards
 of the form `!_`.  However, `!_` is unsatisfactory and perhaps even
 dangerous for such usage, as it does not constrain the type in any manner.
 In particular, there is nothing to prevent somebody from writing
 {{{#!hs
 main = do
     let !_ = assert someCondition   -- (missing an argument)
     pure ()
 }}}
 which //is// in fact useless.

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13646
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler