I've been trying to get my head around the Wadler-Leijen pretty printing combinators for a while now and having some trouble. Specifically, I have trouble getting them to pick optimal line breaks. The existing combinators like 'sep' (and everything built from it) merge all elements with <$> and then 'group' the whole thing, with the result that they either all go on one line, or get one line each. This is quite ugly for large lists of small elements. The other alternative is 'fillSep', which does a separate 'group' on each element. Unfortunately, it then tends to make very bad line wrapping decisions, e.g. you get: Rec { hi = "there" } Rec { hi = "there", hi = "there" , hi = "there" } Rec { lab = "short", label = [ 0, 1, 2 , 3, 4, 5 , 6, 7, 8 , 9, 10 , 11, 12 ] } No matter how much fancy 'group's and 'nest's and whatnot I threw in, it was always a choice between forcing wrapping on every element and looking ugly for many small elements, or trying to fit more into one line and having it wrap in the wrong place and wind up scrunched up on the right margin. Here's my latest attempt: list = commas PP.lbracket PP.rbracket . map format commas :: Doc -> Doc -> [Doc] -> Doc commas left right xs = PP.group $ left <+> punctuate (\x -> PP.group (x <$$> PP.comma <> PP.space)) xs <$> right punctuate :: (Doc -> Doc) -> [Doc] -> Doc punctuate f [] = mempty punctuate f [x] = x punctuate f (x:xs) = f x <> punctuate f xs record :: Doc -> [(String, Doc)] -> Doc record title fields = PP.group $ PP.hang 2 $ title <$> (commas PP.lbrace PP.rbrace (map f fields)) where f (label, field) = PP.hang 2 $ PP.group $ PP.text label <+> PP.equals <$> field But the thing is, the HughesPJ-using Language.Haskell.Pretty in haskell-src gets the line wrapping just right. So I investigated how it works, and it's very simple, here's the reduced version: class Pretty a where format :: a -> Doc list :: (Pretty a) => [a] -> Doc list = bracket_list . PP.punctuate PP.comma . map format fsep' :: [Doc] -> Doc fsep' [] = PP.empty fsep' (d:ds) = PP.nest 2 (PP.fsep (PP.nest (-2) d:ds)) bracket_list :: [Doc] -> Doc bracket_list = PP.brackets . PP.fsep brace_list :: [Doc] -> Doc brace_list = PP.braces . PP.fsep record :: Doc -> [(String, Doc)] -> Doc record title fields = title <> (brace_list (map field fields)) where field (name, val) = fsep' [PP.text name, PP.equals, val] ---- This formats records like so: Rec{hi = "there"} Rec{hi = "there" hi = "there" hi = "there"} Rec{label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30] label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]} Rec{lab = "short" label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]} Much better, even if it's not my preferred style! Of course WL doesn't have fsep or the negative nest craziness (I don't even know what it's doing there), but it has the more general 'group'. However, no matter how complicated I got with WL it just never came out right. In contrast, a very simple HughesPJ implementation gets it right. The thing is, when trying to figure out which pretty print library to use, the consensus is that WL is just all around better, even though HughesPJ is somewhat standard (but there are 6 WL variants on hackage, and no (?) HughesPJ ones). So am I just using it wrong? If I translate the HughesPJ one over directly into LW, here's what I get: Rec{hi = "there"} Rec{hi = "there", hi = "there", hi = "there"} Rec{label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30], label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]} Rec{lab = "short", label = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]} So is WL really all it's cracked up to be? Am I using it wrong? I was going to suggest some consolidation in the pretty printing library packages, but now I'm not even sure which style should "win"...