{-# LANGUAGE OverloadedStrings #-}
module Data.CSS.Syntax.StyleSheet (
parse, parse', parseForURL, TrivialStyleSheet(..),
StyleSheet(..), skipAtRule, scanBlock, skipSpace,
StyleRule(..),
-- For parsing at-rules, HTML "style" attribute, etc.
parseProperties, parseProperties',
-- for testing
scanValue
) where
import Data.CSS.Syntax.Tokens
import Data.CSS.Syntax.Selector
import Data.CSS.Syntax.StylishUtil
import Data.Text.Internal (Text(..))
import Data.Text (pack, unpack)
import Network.URI (parseRelativeReference, relativeTo, uriToString, URI(..))
--------
---- Output type class
--------
class StyleSheet s where
setPriority :: Int -> s -> s
setPriority _ = id
addRule :: s -> StyleRule -> s
addAtRule :: s -> Text -> [Token] -> (s, [Token])
addAtRule self _ tokens = (self, skipAtRule tokens)
addRules :: StyleSheet ss => ss -> ([Selector], ([(Text, [Token])], Text)) -> ss
addRules self (selector:selectors, val@(props, psuedoel)) = addRules self' (selectors, val)
where self' = addRule self $ StyleRule selector props psuedoel
addRules self ([], _) = self
data StyleRule = StyleRule Selector [(Text, [Token])] Text deriving (Show, Eq)
data TrivialStyleSheet = TrivialStyleSheet [StyleRule] deriving (Show, Eq)
instance StyleSheet TrivialStyleSheet where
addRule (TrivialStyleSheet self) rule = TrivialStyleSheet $ rule:self
--------
---- Basic parsing
--------
parse :: StyleSheet s => s -> Text -> s
parse stylesheet source = parse' stylesheet $ tokenize source
parseForURL :: StyleSheet s => s -> URI -> Text -> s
parseForURL stylesheet base source = parse' stylesheet $ rewriteURLs $ tokenize source
where
rewriteURLs (Url text:toks)
| Just url <- parseRelativeReference $ unpack text =
Url (pack $ uriToString id (relativeTo url base) "") : rewriteURLs toks
| otherwise = Function "url" : RightParen : rewriteURLs toks
rewriteURLs (tok:toks) = tok : rewriteURLs toks
rewriteURLs [] = []
parse' :: StyleSheet t => t -> [Token] -> t
-- Things to skip.
parse' stylesheet (Whitespace:tokens) = parse' stylesheet tokens
parse' stylesheet (CDO:tokens) = parse' stylesheet tokens
parse' stylesheet (CDC:tokens) = parse' stylesheet tokens
parse' stylesheet (Comma:tokens) = parse' stylesheet tokens -- TODO issue warnings.
parse' stylesheet [] = stylesheet
parse' stylesheet (AtKeyword kind:tokens) = parse' stylesheet' tokens'
where (stylesheet', tokens') = addAtRule stylesheet kind tokens
parse' stylesheet tokens = parse' (addRules stylesheet rule) tokens'
where (rule, tokens') = concatP (,) parseSelectors parseProperties tokens
--------
---- Property parsing
--------
parseProperties :: Parser ([(Text, [Token])], Text)
parseProperties (LeftCurlyBracket:tokens) = noPsuedoel $ parseProperties' tokens
parseProperties (Whitespace:tokens) = parseProperties tokens
parseProperties (Colon:Colon:Ident n:tokens) = ((val, n), tokens')
where ((val, _), tokens') = parseProperties tokens
-- This error recovery is a bit overly conservative, but it's simple.
parseProperties (_:tokens) = noPsuedoel ([], skipAtRule tokens)
parseProperties [] = noPsuedoel ([], [])
noPsuedoel :: (x, y) -> ((x, Text), y)
noPsuedoel (val, tokens) = ((val, ""), tokens)
parseProperties' :: Parser [(Text, [Token])]
parseProperties' (Whitespace:tokens) = parseProperties' tokens
parseProperties' (Ident name:tokens)
| Colon:tokens' <- skipSpace tokens =
concatP appendProp scanValue parseProperties' tokens'
where appendProp value props = (name, value):props
parseProperties' (RightCurlyBracket:tokens) = ([], tokens)
parseProperties' [] = ([], [])
parseProperties' tokens = parseProperties' (skipValue tokens)
--------
---- Skipping/Scanning utilities
--------
scanAtRule :: Parser [Token]
scanAtRule (Semicolon:tokens) = ([], tokens)
scanAtRule (LeftCurlyBracket:tokens) = scanInner tokens $ \rest -> ([], rest)
scanAtRule tokens@(LeftParen:_) = scanInner tokens scanValue
scanAtRule tokens@(Function _:_) = scanInner tokens scanValue
scanAtRule tokens@(LeftSquareBracket:_) = scanInner tokens scanValue
-- To ensure parens are balanced, should already be handled.
scanAtRule (RightCurlyBracket:tokens) = ([], RightCurlyBracket:tokens)
scanAtRule (RightParen:tokens) = ([], RightParen:tokens)
scanAtRule (RightSquareBracket:tokens) = ([], RightSquareBracket:tokens)
scanAtRule tokens = capture scanAtRule tokens
skipAtRule :: [Token] -> [Token]
skipAtRule tokens = snd $ scanAtRule tokens
scanValue :: Parser [Token]
scanValue (Semicolon:tokens) = ([], tokens)
scanValue (Whitespace:tokens) = scanValue tokens
scanValue tokens@(LeftCurlyBracket:_) = scanInner tokens scanValue
scanValue tokens@(LeftParen:_) = scanInner tokens scanValue
scanValue tokens@(Function _:_) = scanInner tokens scanValue
scanValue tokens@(LeftSquareBracket:_) = scanInner tokens scanValue
-- To ensure parens are balanced, should already be handled.
scanValue (RightCurlyBracket:tokens) = ([], RightCurlyBracket:tokens)
scanValue (RightParen:tokens) = ([], RightParen:tokens)
scanValue (RightSquareBracket:tokens) = ([], RightSquareBracket:tokens)
scanValue tokens = capture scanValue tokens
skipValue :: [Token] -> [Token]
skipValue tokens = snd $ scanValue tokens