aboutsummaryrefslogtreecommitdiff
path: root/src/Text/Pandoc
diff options
context:
space:
mode:
authorJohn MacFarlane <jgm@berkeley.edu>2010-11-27 07:08:32 -0800
committerJohn MacFarlane <jgm@berkeley.edu>2010-11-27 07:08:32 -0800
commit283f1e60ccd2cdfe48a056687565aedade5ceb6d (patch)
tree56ed0532e03666a8c21b55b04bb2656b2a5a180a /src/Text/Pandoc
parent044a9a61574ff1414b44e2f92307996cba00a2e1 (diff)
downloadpandoc-283f1e60ccd2cdfe48a056687565aedade5ceb6d.tar.gz
Use parsec parsers to split locator.
This is easier to read and maintain. Also, formatting is now stripped from the locator prefix, so you can write e.g. '*p.* 33'.
Diffstat (limited to 'src/Text/Pandoc')
-rw-r--r--src/Text/Pandoc/Biblio.hs61
1 files changed, 34 insertions, 27 deletions
diff --git a/src/Text/Pandoc/Biblio.hs b/src/Text/Pandoc/Biblio.hs
index bf1624bb4..4a8cea4da 100644
--- a/src/Text/Pandoc/Biblio.hs
+++ b/src/Text/Pandoc/Biblio.hs
@@ -36,6 +36,9 @@ import qualified Data.Map as M
import Text.CSL hiding ( Cite(..), Citation(..) )
import qualified Text.CSL as CSL ( Cite(..) )
import Text.Pandoc.Definition
+import Text.Pandoc.Shared (stringify)
+import Text.ParserCombinators.Parsec
+import Control.Monad
-- | Process a 'Pandoc' document by adding citations formatted
-- according to a CSL style, using 'citeproc' from citeproc-hs.
@@ -152,7 +155,7 @@ setCitationNoteNum i = map $ \c -> c { citationNoteNum = i}
toCslCite :: Citation -> CSL.Cite
toCslCite c
= let (l, s) = locatorWords $ citationSuffix c
- (la,lo) = parseLocator $ unwords l
+ (la,lo) = parseLocator l
citMode = case citationMode c of
AuthorInText -> (True, False)
SuppressAuthor -> (False,True )
@@ -168,30 +171,34 @@ toCslCite c
, CSL.citeHash = citationHash c
}
-locatorWords :: [Inline] -> ([String], [Inline])
-locatorWords (Space : t) = locatorWords t
-locatorWords (Str "" : t) = locatorWords t
-locatorWords a@(Str (',' : s) : t)
- = if ws /= [] then (ws, t') else ([], a)
- where
- (ws, t') = locatorWords (Str s:t)
-locatorWords i
- = if any isDigit w then (w':ws, s'') else ([], i)
- where
- (w, s') = locatorWord i
- (ws, s'') = locatorWords s'
- w' = if ws == [] then w else w ++ ","
-
-locatorWord :: [Inline] -> (String, [Inline])
-locatorWord (Space : r) = (" " ++ ts, r')
- where
- (ts, r') = locatorWord r
-locatorWord (Str t : r)
- | t' /= "" = (w , Str t' : r)
- | otherwise = (t ++ ts, r' )
- where
- w = takeWhile (/= ',') t
- t' = dropWhile (/= ',') t
- (ts, r') = locatorWord r
-locatorWord i = ("", i)
+locatorWords :: [Inline] -> (String, [Inline])
+locatorWords inp =
+ case parse (liftM2 (,) pLocator getInput) "suffix" inp of
+ Right r -> r
+ Left _ -> ("",inp)
+
+pMatch :: (Inline -> Bool) -> GenParser Inline st Inline
+pMatch condition = try $ do
+ t <- anyToken
+ guard $ condition t
+ return t
+
+pSpace :: GenParser Inline st Inline
+pSpace = pMatch (== Space)
+
+pLocator :: GenParser Inline st String
+pLocator = try $ do
+ optional $ pMatch (== Str ",")
+ optional pSpace
+ f <- many1 (notFollowedBy pSpace >> anyToken)
+ gs <- many1 pWordWithDigits
+ return $ stringify f ++ (' ' : unwords gs)
+
+pWordWithDigits :: GenParser Inline st String
+pWordWithDigits = try $ do
+ pSpace
+ r <- many1 (notFollowedBy pSpace >> anyToken)
+ let s = stringify r
+ guard $ any isDigit s
+ return s