Language.Haskell.Parser and layout rules

[Simon Marlow]

On 18 February 2005 16:31, Mirko Rahn wrote:

>> The Report (section 9.3, especially notes 1 and 2) says "A nested
>> context must be further indented than the enclosing context", ...
> 
> Okay, I understand, but the behavoir of hugs vs ghc vs
> Language.Haskell.Parser is still strange: The sniplets
> 
> ** A:
> g xs = do ys <- workM xs
>            if null ys then return [] else do
>            zs <- workM ys
>            return zs
> 
> ** B:
> g xs = do ys <- workM xs
>            if null ys then return [] else do
>            zs <- workM ys
>             return zs
> 
> ** C:
> f xs = case xs of
>           y:ys -> case ys of
>           z:zs -> zs
> 
> ** D:
> f x = case x of
>           False -> do
>      { return x; }
> 
> are accepted by
>                                        A    B    C   D
> ghc                                   yes  no   no  yes
> hugs                                  yes  no   no  no
> Language.Haskell.Parser.parseModule   no   yes  yes no
> report                                no   no   no  no

Hugs is probably the most correct here.  A is intentionally parsed by
Hugs and GHC.  D is a known bug in GHC.

It looks like Language.Haskell.Parser may have bugs in this area.

Interestingly, Haddock is better than GHC, giving the same results as
Hugs.  Haddock's parser is based on Language.Haskell.Parser, so it looks
like a bug has been introduced in Language.Haskell.Parser at some point.

Cheers,
	Simon