aboutsummaryrefslogtreecommitdiff
path: root/haddock-library/src
diff options
context:
space:
mode:
Diffstat (limited to 'haddock-library/src')
-rw-r--r--haddock-library/src/Documentation/Haddock/Parser.hs81
1 files changed, 46 insertions, 35 deletions
diff --git a/haddock-library/src/Documentation/Haddock/Parser.hs b/haddock-library/src/Documentation/Haddock/Parser.hs
index 6bb88035..ca9e9d8d 100644
--- a/haddock-library/src/Documentation/Haddock/Parser.hs
+++ b/haddock-library/src/Documentation/Haddock/Parser.hs
@@ -231,18 +231,20 @@ markdownImage = fromHyperlink <$> ("!" *> linkParser)
-- | Paragraph parser, called by 'parseParas'.
paragraph :: Parser (DocH mod Identifier)
-paragraph = examples <|> skipSpace *> (
- since
- <|> unorderedList
- <|> orderedList
- <|> birdtracks
- <|> codeblock
- <|> property
- <|> header
- <|> textParagraphThatStartsWithMarkdownLink
- <|> definitionList
- <|> docParagraph <$> textParagraph
- )
+paragraph = examples <|> do
+ indent <- takeIndent
+ choice
+ [ since
+ , unorderedList indent
+ , orderedList indent
+ , birdtracks
+ , codeblock
+ , property
+ , header
+ , textParagraphThatStartsWithMarkdownLink
+ , definitionList indent
+ , docParagraph <$> textParagraph
+ ]
since :: Parser (DocH mod a)
since = ("@since " *> version <* skipHorizontalSpace <* endOfLine) >>= setSince >> return DocEmpty
@@ -283,16 +285,16 @@ textParagraphThatStartsWithMarkdownLink = docParagraph <$> (docAppend <$> markdo
| otherwise = " "
-- | Parses unordered (bullet) lists.
-unorderedList :: Parser (DocH mod Identifier)
-unorderedList = DocUnorderedList <$> p
+unorderedList :: BS.ByteString -> Parser (DocH mod Identifier)
+unorderedList indent = DocUnorderedList <$> p
where
- p = ("*" <|> "-") *> innerList p
+ p = ("*" <|> "-") *> innerList indent p
-- | Parses ordered lists (numbered or dashed).
-orderedList :: Parser (DocH mod Identifier)
-orderedList = DocOrderedList <$> p
+orderedList :: BS.ByteString -> Parser (DocH mod Identifier)
+orderedList indent = DocOrderedList <$> p
where
- p = (paren <|> dot) *> innerList p
+ p = (paren <|> dot) *> innerList indent p
dot = (decimal :: Parser Int) <* "."
paren = "(" *> decimal <* ")"
@@ -301,23 +303,24 @@ orderedList = DocOrderedList <$> p
-- same paragraph. Usually used as
--
-- > someListFunction = listBeginning *> innerList someListFunction
-innerList :: Parser [DocH mod Identifier] -> Parser [DocH mod Identifier]
-innerList item = do
+innerList :: BS.ByteString -> Parser [DocH mod Identifier]
+ -> Parser [DocH mod Identifier]
+innerList indent item = do
c <- takeLine
- (cs, items) <- more item
+ (cs, items) <- more indent item
let contents = docParagraph . parseString . dropNLs . unlines $ c : cs
return $ case items of
Left p -> [contents `docAppend` p]
Right i -> contents : i
-- | Parses definition lists.
-definitionList :: Parser (DocH mod Identifier)
-definitionList = DocDefList <$> p
+definitionList :: BS.ByteString -> Parser (DocH mod Identifier)
+definitionList indent = DocDefList <$> p
where
p = do
label <- "[" *> (parseStringBS <$> takeWhile1 (`notElem` ("]\n" :: String))) <* ("]" <* optional ":")
c <- takeLine
- (cs, items) <- more p
+ (cs, items) <- more indent p
let contents = parseString . dropNLs . unlines $ c : cs
return $ case items of
Left x -> [(label, contents `docAppend` x)]
@@ -330,32 +333,40 @@ dropNLs = reverse . dropWhile (== '\n') . reverse
-- | Main worker for 'innerList' and 'definitionList'.
-- We need the 'Either' here to be able to tell in the respective functions
-- whether we're dealing with the next list or a nested paragraph.
-more :: Monoid a => Parser a
+more :: Monoid a => BS.ByteString -> Parser a
-> Parser ([String], Either (DocH mod Identifier) a)
-more item = innerParagraphs <|> moreListItems item
- <|> moreContent item <|> pure ([], Right mempty)
+more indent item = innerParagraphs indent
+ <|> moreListItems indent item
+ <|> moreContent indent item
+ <|> pure ([], Right mempty)
-- | Used by 'innerList' and 'definitionList' to parse any nested paragraphs.
-innerParagraphs :: Parser ([String], Either (DocH mod Identifier) a)
-innerParagraphs = (,) [] . Left <$> ("\n" *> indentedParagraphs)
+innerParagraphs :: BS.ByteString
+ -> Parser ([String], Either (DocH mod Identifier) a)
+innerParagraphs indent = (,) [] . Left <$> ("\n" *> indentedParagraphs indent)
-- | Attempts to fetch the next list if possibly. Used by 'innerList' and
-- 'definitionList' to recursively grab lists that aren't separated by a whole
-- paragraph.
-moreListItems :: Parser a
+moreListItems :: BS.ByteString -> Parser a
-> Parser ([String], Either (DocH mod Identifier) a)
-moreListItems item = (,) [] . Right <$> (skipSpace *> item)
+moreListItems indent item = (,) [] . Right <$> indentedItem
+ where
+ indentedItem = string indent *> skipSpace *> item
-- | Helper for 'innerList' and 'definitionList' which simply takes
-- a line of text and attempts to parse more list content with 'more'.
-moreContent :: Monoid a => Parser a
+moreContent :: Monoid a => BS.ByteString -> Parser a
-> Parser ([String], Either (DocH mod Identifier) a)
-moreContent item = first . (:) <$> nonEmptyLine <*> more item
+moreContent indent item = first . (:) <$> nonEmptyLine <*> more indent item
-- | Parses an indented paragraph.
-- The indentation is 4 spaces.
-indentedParagraphs :: Parser (DocH mod Identifier)
-indentedParagraphs = (concat <$> dropFrontOfPara " ") >>= parseParagraphs
+indentedParagraphs :: BS.ByteString -> Parser (DocH mod Identifier)
+indentedParagraphs indent =
+ (concat <$> dropFrontOfPara indent') >>= parseParagraphs
+ where
+ indent' = string $ BS.append indent " "
-- | Grab as many fully indented paragraphs as we can.
dropFrontOfPara :: Parser BS.ByteString -> Parser [String]