"Pattern match(es) are overlapped" warning with a single view pattern (ghc 6.9.20080606)

[rwbarton]

Compiling the program

{-# LANGUAGE ViewPatterns #-}

main = print $ case 6 of
                 (even -> True) -> 4

yields the warning message

view.hs:3:15:
    Warning: Pattern match(es) are overlapped
             In a case alternative: ((even -> True)) -> ...

I haven't used view patterns before... but surely a single pattern can't overlap anything, can it?

[igloo]

This is due to this case of check' in deSugar/Check.lhs:

-- FIXME: hack to get view patterns through for now
   | otherwise    = ([([],[])],emptyUniqSet)

(the emptyUniqSet means "the pattern is never used").

[NeilMitchell]

If you use View Patterns, this is absolutely infuriating! In one project I use view patterns in 8 places, so the compilation output becomes so verbose I miss genuine errors. It's fine for view patterns to not participate in overlap/exhaustive checking, but it should do so by _not_ issuing loads of bogus warnings which cannot be worked around :-)

This bug seriously impacts on the usability of view patterns, which is a real shame because I'm finding them to be a very useful practical feature. I'd argue for this bug being "major" rather than "minor".

[Baughn]

This bug makes view patterns absolutely impossible to use.

I accept that fixing it could take some work, but can we at least suppress the warning in 6.10.4? Add an option for that, perhaps?

[igloo]

You can turn the warning off with -fno-warn-overlapping-patterns.

[Baughn]

I could, but then I lose all the overlap warnings, and I'd really like to keep them around. I don't always use view patterns.

[ajd]

Patch that fixes the problem by turning off the checker entirely for any equation that has view patterns

[ajd]

Regression test (not sure if I added the test correctly, but the meat is there)

[ajd]

I've attached a patch to GHC and a patch to the testsuite that ought to fix the problem by turning off the checker for any definition that includes view patterns.

[igloo]

Thanks, patches applied.

[simonpj]

I've improved the fix with

Mon Aug 10 15:11:58 BST 2009  simonpj@microsoft.com
  * Improve the recent changes to overlap-checking for view patters
  
  The previous patch simply gave up for view patterns; this version
  instead treats them like n+k patterns and gives signficantly better
  results.
  
  Less code, too.
  

    M ./compiler/deSugar/Check.lhs -85 +101
    M ./compiler/hsSyn/HsPat.lhs -29 +1

and added a test case

Simon