diff --git a/changelog.md b/changelog.md
--- a/changelog.md
+++ b/changelog.md
@@ -1,5 +1,19 @@
# Changelog for commonmark-extensions
+## 0.2.5.2
+
+ * Improve autolinks extension (#147).
+ The autolinks extension was interacting badly with explicit links,
+ To fix this we had to make autolink parsing a bit stricter than
+ the GFM spec does. They allow unbalanced `)` except at the end
+ of a URL (which is defined as: followed by optional final punctuation
+ then whitespace or eof). With this change, we don't allow unbalanced
+ `)` at all in raw URLs. This should not be a big problem in practice.
+
+ * Protect against quadratic generated table size explosion (Michael Howell).
+ This commit adds a limit to the number of auto-completed cells
+ around 200,000. The result is, in these original samples:
+
## 0.2.5.1
* Add `test/alerts.md` to extra-source-files in cabal file.
diff --git a/commonmark-extensions.cabal b/commonmark-extensions.cabal
--- a/commonmark-extensions.cabal
+++ b/commonmark-extensions.cabal
@@ -1,5 +1,5 @@
name: commonmark-extensions
-version: 0.2.5.1
+version: 0.2.5.2
synopsis: Pure Haskell commonmark parser.
description:
This library provides some useful extensions to core commonmark
diff --git a/src/Commonmark/Extensions/Autolink.hs b/src/Commonmark/Extensions/Autolink.hs
--- a/src/Commonmark/Extensions/Autolink.hs
+++ b/src/Commonmark/Extensions/Autolink.hs
@@ -32,7 +32,7 @@
wwwAutolink = try $ do
lookAhead $ satisfyWord (== "www")
validDomain
- linkSuffix
+ linkPath 0
return "http://"
validDomain :: Monad m => InlineParser m ()
@@ -47,30 +47,48 @@
domainPart
skipMany1 $ try (symbol '.' >> domainPart)
-linkSuffix :: Monad m => InlineParser m ()
-linkSuffix = try $ do
- toks <- getInput
- let possibleSuffixTok (Tok (Symbol c) _ _) =
- c `notElem` ['<','>','{','}','|','\\','^','[',']','`']
- possibleSuffixTok (Tok WordChars _ _) = True
- possibleSuffixTok _ = False
- let isDroppable (Tok (Symbol c) _ _) =
- c `elem` ['?','!','.',',',':','*','_','~']
- isDroppable _ = False
- let numToks = case dropWhile isDroppable $
- reverse (takeWhile possibleSuffixTok toks) of
- (Tok (Symbol ')') _ _ : xs)
- | length [t | t@(Tok (Symbol '(') _ _) <- xs] <=
- length [t | t@(Tok (Symbol ')') _ _) <- xs]
- -> length xs
- (Tok (Symbol ';') _ _
- : Tok WordChars _ _
- : Tok (Symbol '&') _ _
- : xs) -> length xs
- xs -> length xs
- count numToks anyTok
- return ()
+linkPath :: Monad m => Int -> InlineParser m ()
+linkPath openParens =
+ try (symbol '&' *>
+ notFollowedBy
+ (try (satisfyWord (const True) *> symbol ';' *> linkEnd)) *>
+ linkPath openParens)
+ <|> (pathPunctuation *> linkPath openParens)
+ <|> (symbol '(' *> linkPath (openParens + 1))
+ <|> (guard (openParens > 0) *> symbol ')' *> linkPath (openParens - 1))
+ -- the following clause is needed to implement the GFM spec, which allows
+ -- unbalanced ) except at link end. However, leaving this in causes
+ -- problematic interaction with explicit link syntax in certain odd cases (see #147).
+ -- <|> (notFollowedBy linkEnd *> symbol ')' *> linkPath (openParens - 1))
+ <|> (satisfyTok (\t -> case tokType t of
+ LineEnd -> False
+ Spaces -> False
+ Symbol c -> not (isTrailingPunctuation c || c == '&' || c == ')')
+ _ -> True) *> linkPath openParens)
+ <|> pure ()
+linkEnd :: Monad m => InlineParser m ()
+linkEnd = try $ skipMany trailingPunctuation *> (void whitespace <|> eof)
+
+trailingPunctuation :: Monad m => InlineParser m ()
+trailingPunctuation = void $
+ satisfyTok (\t -> case tokType t of
+ Symbol c -> isTrailingPunctuation c
+ _ -> False)
+
+isTrailingPunctuation :: Char -> Bool
+isTrailingPunctuation =
+ (`elem` ['!', '"', '\'', ')', '*', ',', '.', ':', ';', '?', '_', '~', '<'])
+
+pathPunctuation :: Monad m => InlineParser m ()
+pathPunctuation = try $ do
+ satisfyTok (\t -> case tokType t of
+ Symbol c -> isTrailingPunctuation c && c /= ')' && c /= '<'
+ _ -> False)
+ void $ lookAhead (satisfyTok (\t -> case tokType t of
+ WordChars -> True
+ _ -> False))
+
urlAutolink :: Monad m => InlineParser m Text
urlAutolink = try $ do
satisfyWord (`elem` ["http", "https", "ftp"])
@@ -78,7 +96,7 @@
symbol '/'
symbol '/'
validDomain
- linkSuffix
+ linkPath 0
return ""
emailAutolink :: Monad m => InlineParser m Text
diff --git a/src/Commonmark/Extensions/PipeTable.hs b/src/Commonmark/Extensions/PipeTable.hs
--- a/src/Commonmark/Extensions/PipeTable.hs
+++ b/src/Commonmark/Extensions/PipeTable.hs
@@ -34,6 +34,9 @@
data PipeTableData = PipeTableData
{ pipeTableAlignments :: [ColAlignment]
+ , pipeTableColCount :: !Int
+ , pipeTableRowCount :: !Int
+ , pipeTableCellCount :: !Int
, pipeTableHeaders :: [[Tok]]
, pipeTableRows :: [[[Tok]]] -- in reverse order
} deriving (Show, Eq, Data, Typeable)
@@ -136,6 +139,14 @@
{ syntaxBlockSpecs = [pipeTableBlockSpec]
}
+getAutoCompletedCellCount :: PipeTableData -> Int
+getAutoCompletedCellCount tabledata =
+ (numrows * numcols) - numcells
+ where
+ numrows = pipeTableRowCount tabledata
+ numcols = pipeTableColCount tabledata
+ numcells = pipeTableCellCount tabledata
+
-- This parser is structured as a system that parses the *second* line first,
-- then parses the first line. That is, if it detects a delimiter row as the
-- second line of a paragraph, it converts the paragraph into a table. This seems
@@ -173,6 +184,9 @@
updateState $ \st' -> st'{ nodeStack = rest }
let tabledata = PipeTableData
{ pipeTableAlignments = aligns
+ , pipeTableColCount = length cells
+ , pipeTableRowCount = 0
+ , pipeTableCellCount = 0
, pipeTableHeaders = cells
, pipeTableRows = []
}
@@ -194,23 +208,45 @@
let tabledata = fromDyn
(blockData ndata)
PipeTableData{ pipeTableAlignments = []
+ , pipeTableColCount = 0
+ , pipeTableRowCount = 0
+ , pipeTableCellCount = 0
, pipeTableHeaders = []
, pipeTableRows = [] }
pos <- getPosition
cells <- pCells
- let tabledata' = tabledata{ pipeTableRows =
- cells : pipeTableRows tabledata }
+ let cells' = take (pipeTableColCount tabledata) cells
+ let tabledata' =
+ tabledata{ pipeTableRows = cells' : pipeTableRows tabledata
+ , pipeTableRowCount = 1 + pipeTableRowCount tabledata
+ , pipeTableCellCount = length cells' + pipeTableCellCount tabledata
+ }
+ -- Protect against quadratic output size explosion.
+ --
+ -- Because the table extension fills in missing table cells,
+ -- you can, theoretically, force the output to grows as the
+ -- square of the input by adding one column and one row.
+ -- This is a side-effect of the extension as specified in GFM,
+ -- and follows from the geometric definition of "squaring."
+ --
+ -- To prevent this, we track the number of Real Cells,
+ -- and if the number of autocompleted cells exceeds 200,000,
+ -- stop parsing.
+ guard $ getAutoCompletedCellCount tabledata <= 200000
return $! (pos, Node ndata{ blockData =
toDyn tabledata' } children)
, blockConstructor = \(Node ndata _) -> do
let tabledata = fromDyn
(blockData ndata)
PipeTableData{ pipeTableAlignments = []
+ , pipeTableColCount = 0
+ , pipeTableRowCount = 0
+ , pipeTableCellCount = 0
, pipeTableHeaders = []
, pipeTableRows = [] }
let aligns = pipeTableAlignments tabledata
headers <- mapM runInlineParser (pipeTableHeaders tabledata)
- let numcols = length headers
+ let numcols = pipeTableColCount tabledata
rows <- mapM (mapM runInlineParser . take numcols . (++ (repeat [])))
(reverse $ pipeTableRows tabledata)
return $! (pipeTable aligns headers rows)
diff --git a/test/autolinks.md b/test/autolinks.md
--- a/test/autolinks.md
+++ b/test/autolinks.md
@@ -49,11 +49,38 @@
Visit www.commonmark.org/~jm/foo/bar.pdf.
````````````````````````````````
-When an autolink ends in `)`, we scan the entire autolink for the total number
-of parentheses. If there is a greater number of closing parentheses than
-opening ones, we don't consider the last character part of the autolink, in
-order to facilitate including an autolink inside a parenthesis:
+((commonmark-hs: We depart from the GFM spec here. Alternative spec and tests
+can be found after these commented-out ones. For motivation for this departure from
+the GFM spec, see #147.))
+> When an autolink ends in `)`, we scan the entire autolink for the total number
+> of parentheses. If there is a greater number of closing parentheses than
+> opening ones, we don't consider the last character part of the autolink, in
+> order to facilitate including an autolink inside a parenthesis:
+>
+> ```````````````````````````````` example
+> www.google.com/search?q=Markup+(business)
+>
+> (www.google.com/search?q=Markup+(business))
+> .
+> www.google.com/search?q=Markup+(business)
+> (www.google.com/search?q=Markup+(business))
+> ````````````````````````````````
+>
+> This check is only done when the link ends in a closing parentheses `)`, so if
+> the only parentheses are in the interior of the autolink, no special rules are
+> applied:
+>
+> ```````````````````````````````` example
+> www.google.com/search?q=(business))+ok
+> .
+> www.google.com/search?q=(business))+ok
+> ````````````````````````````````
+
+Autolinks can contain balanced pairs of parentheses, or unbalanced `)`.
+We don't allow unbalanced `)`, in order to facilitate including
+an autolink inside a parenthesis:
+
```````````````````````````````` example
www.google.com/search?q=Markup+(business)
@@ -63,15 +90,14 @@
(www.google.com/search?q=Markup+(business))
````````````````````````````````
-This check is only done when the link ends in a closing parentheses `)`, so if
-the only parentheses are in the interior of the autolink, no special rules are
-applied:
-
+Issue #147:
```````````````````````````````` example
-www.google.com/search?q=(business))+ok
+[link](https://baidu.com)aaabbb
.
-www.google.com/search?q=(business))+ok
+linkaaabbb
````````````````````````````````
+
+((End of diverging section.))
If an autolink ends in a semicolon (`;`), we check to see if it appears to
resemble an [entity reference][entity references]; if the preceding text is `&`