Text.PrettyPrint bug or doc bug

[Benedikt Huber]

Hi,
I filed a bug (http://hackage.haskell.org/trac/ghc/ticket/3911)  
referencing this discussion,
and attached a patch fixing the problem.

On 02.03.2010, at 09:44, Marcus D. Gabriel wrote:
> No, I do not really have a use case.  I discovered it upon recompiling
> an older program whereupon the pretty printing behaviour changed which
> is how I discovered the inconsistency between the code and the
> documentation.
Oh, this seems like a reasonable use/test case, thank you.

cheers, benedikt


> Essentially, I just wanted to make columns, one line at a time.
> So,
>
> -- dwiw == do what i (basically) want
> dwiw :: [Doc] -> Doc
> dwiw = foldr ($$) empty
>
> f2 :: ([Doc] -> Doc) -> IO ()
> f1 =
>  \f -> (putStrLn $ render $
>  f [nest 0$text "abc", nest 10$text "def", nest 20$text "hig"])
>
> f2 :: ([Doc] -> Doc) -> IO ()
> f2 =
>  \f -> (putStrLn $ render $
>  f [nest 0$text "abc", nest 10$text "def4567890123456789", nest  
> 20$text
> "hig"])
>
> *> mapM_ f1 [vcat,dwiw]
> abc
>          def
>                    hig
> abc       def       hig
> *> mapM_ f1 [vcat,dwiw]
> abc
>          def4567890123456789
>                    hig
> abc       def4567890123456789
>                    hig
>
> Does that make visual sense?
>
> The logic is as follows:
> 1/ Figure out the size of your columns in advance to use nest;
> 2/ Require that the sum of the columns sizes is less than or
>   equal to the maximum length of line that you are willing to
>   tolerate; and
> 3/ dwiw splits the line if and when needed.
>
> Cheers,
> - Marcus
>
> P.S. I missed your previous post Ian; thanks for it.
>
> Benedikt Huber wrote:
>>
>> Ian Lynagh schrieb:
>>> On Wed, Feb 24, 2010 at 01:10:38PM +0100, Marcus D. Gabriel wrote:
>>>> The documentation for Text.PrettyPrint.HughesPJ states
>>>> that vcat is the "List version of $$", but it works as
>>>> the list version ($+$).
>>>>
>>>> So either the documentation needs to be changed or
>>>> the code.  It would be nice to have both versions.
>>>
>> Hi,
>> do you have a actual use case for the "List version of $$" ? It would
>> make a good test I suppose.
>>
>>> It looks like the behaviour changed in
>>>
>>> Tue Jun 24 12:37:15 BST 2008  benedikt.huber at gmail.com
>>>  * fillNB bug, lazy vcat
>>>
>>> i.e. the version that came with GHC 6.10. As it's had the current
>>> behaviour for some time, it would probably make sense to have a
>>> library submission to determine what the behaviour should be, and
>>> whether we want another function for the other behaviour:
>>>    http://www.haskell.org/haskellwiki/Library_submissions
>>>
>> This is indeed a bug; I've reported it back in December 2008, but
>> unfortunately did not file a bug report:
>>
>> http://www.haskell.org/pipermail/libraries/2008-December/011032.html
>> (at the end of the message, starting with [1])
>>
>> Personally I think the "List version of $+$" is more common, but
>> changing the implementation seems also fine.
>>
>> It would be nice to solve the other (trickier) issue discussed in the
>> mail referenced above as well.
>>
>> cheers, benedikt