{-# LANGUAGE NoImplicitPrelude #-}
{-# LANGUAGE CPP #-}
{-# LANGUAGE DeriveDataTypeable #-}
{-# LANGUAGE DeriveGeneric #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE ViewPatterns #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE OverloadedStrings #-}
{-# LANGUAGE PatternGuards #-}
{-# LANGUAGE ScopedTypeVariables #-}
#if MIN_VERSION_base(4,8,0)
#define OVERLAPS {-# OVERLAPPING #-}
#else
{-# LANGUAGE OverlappingInstances #-}
#define OVERLAPS
#endif
module Text.CSL.Style ( readCSLString
, writeCSLString
, Formatted(..)
, Style(..)
, Locale(..)
, mergeLocales
, CslTerm(..)
, newTerm
, findTerm
, findTerm'
, Abbreviations(..)
, MacroMap
, Citation(..)
, Bibliography(..)
, Option
, mergeOptions
, Layout(..)
, Element(..)
, IfThen(..)
, Condition(..)
, Delimiter
, Match(..)
, match
, DatePart(..)
, defaultDate
, Sort(..)
, Sorting(..)
, compare'
, Form(..)
, Gender(..)
, NumericForm(..)
, DateForm(..)
, Plural(..)
, Name(..)
, NameAttrs
, NamePart(..)
, isPlural
, isName
, isNames
, hasEtAl
, Formatting(..)
, emptyFormatting
, rmTitleCase
, rmTitleCase'
, Quote(..)
, mergeFM
, CSInfo(..)
, CSAuthor(..)
, CSCategory(..)
, CiteprocError(..)
, Output(..)
, Citations
, Cite(..)
, emptyCite
, CitationGroup(..)
, BiblioData(..)
, CiteData(..)
, NameData(..)
, isPunctuationInQuote
, object'
, Agent(..)
, emptyAgent
)
where
import Prelude
import Control.Applicative ((<|>), (<$>))
import Control.Arrow hiding (left, right)
import Control.Monad (mplus)
import Data.Aeson hiding (Number)
import qualified Data.Aeson as Aeson
import Data.Aeson.Types (Pair)
import Data.Char (isLetter, isPunctuation, isUpper, toLower, isDigit)
import qualified Data.Char as Char
import Data.Generics (Data, Typeable)
import Data.List (intercalate, intersperse, nubBy)
import Data.List.Split (wordsBy)
import qualified Data.Map as M
import Data.Maybe (listToMaybe, isNothing)
import Data.String
import Data.Text (Text)
import qualified Data.Text as T
import Data.Yaml.Builder (ToYaml (..))
import qualified Data.Yaml.Builder as Y
import GHC.Generics (Generic)
import Text.CSL.Compat.Pandoc (readHtml, writeMarkdown)
import Text.CSL.Util (headInline, initInline,
lastInline, mapping', mb, parseBool,
parseString, query, splitWhen,
splitStrWhen, tailInline, trimr,
(.#:), (.#?),
AddYaml(..), addSpaceAfterPeriod)
import qualified Text.Pandoc.Builder as B
import Text.Pandoc.Definition hiding (Citation, Cite)
import qualified Text.Pandoc.Walk as Walk
import Text.Pandoc.XML (fromEntities)
#ifdef UNICODE_COLLATION
import qualified Data.Text as T
import qualified Data.Text.ICU as T
#else
import Data.RFC5051 (compareUnicode)
#endif
import qualified Data.Vector as V
readCSLString :: Text -> [Inline]
readCSLString s = Walk.walk handleSmallCapsSpans
$ case readHtml (adjustScTags s) of
Pandoc _ [Plain ils] -> ils
Pandoc _ [Para ils] -> ils
Pandoc _ x -> Walk.query (:[]) x
where handleSmallCapsSpans (Span ("",[],[("style",sty)]) ils)
| T.filter (`notElem` (" \t;" :: String)) sty == "font-variant:small-caps" =
SmallCaps ils
handleSmallCapsSpans x = x
adjustScTags :: Text -> Text
adjustScTags zs
| Just xs <- T.stripPrefix "<sc>" zs =
"<span style=\"font-variant:small-caps;\">" <> adjustScTags xs
| Just xs <- T.stripPrefix "</sc>" zs =
"</span>" <> adjustScTags xs
| Just (x, xs) <- T.uncons zs =
T.cons x (adjustScTags xs)
| otherwise = ""
writeYAMLString :: [Inline] -> Text
writeYAMLString ils =
trimr $ writeMarkdown
$ Pandoc nullMeta
[Plain $ Walk.walk (concatMap (adjustCSL False)) ils]
writeCSLString :: [Inline] -> Text
writeCSLString ils =
trimr $ writeMarkdown
$ Pandoc nullMeta
[Plain $ Walk.walk (concatMap (adjustCSL True)) ils]
adjustCSL :: Bool -> Inline -> [Inline]
adjustCSL _ (Span ("",[],[]) xs) = xs
adjustCSL _ (Span ("",["citeproc-no-output"],[]) _) =
[Str "[CSL STYLE ERROR: reference with no printed form.]"]
adjustCSL True (SmallCaps xs) =
RawInline (Format "html") "<sc>" : xs
++ [RawInline (Format "html") "</sc>"]
adjustCSL True (Subscript xs) =
RawInline (Format "html") "<sub>" : xs
++ [RawInline (Format "html") "</sub>"]
adjustCSL True (Superscript xs) =
RawInline (Format "html") "<sup>" : xs
++ [RawInline (Format "html") "</sup>"]
adjustCSL True (Emph xs) =
RawInline (Format "html") "<i>" : xs
++ [RawInline (Format "html") "</i>"]
adjustCSL True (Strong xs) =
RawInline (Format "html") "<b>" : xs
++ [RawInline (Format "html") "</b>"]
adjustCSL _ x = [x]
newtype Formatted = Formatted { unFormatted :: [Inline] }
deriving ( Show, Read, Eq, Ord, Data, Typeable, Generic )
instance FromJSON Formatted where
parseJSON v@(Array _) =
Formatted <$> (parseJSON v
<|> ((query (:[]) :: [Block] -> [Inline]) <$> parseJSON v))
parseJSON v = (Formatted . readCSLString) Control.Applicative.<$> parseString v
instance ToJSON Formatted where
toJSON = toJSON . writeCSLString . unFormatted
instance ToYaml Formatted where
toYaml = Y.string . writeYAMLString . unFormatted
instance IsString Formatted where
fromString = Formatted . toStr
instance AddYaml Formatted where
x &= (Formatted y) =
\acc -> if null y
then acc
else (x Y..= Formatted y) : acc
instance Semigroup Formatted where
(<>) = appendWithPunct
instance Monoid Formatted where
mempty = Formatted []
mappend = (<>)
mconcat = foldr mappend mempty
instance Walk.Walkable Inline Formatted where
walk f = Formatted . Walk.walk f . unFormatted
walkM f = fmap Formatted . Walk.walkM f . unFormatted
query f = Walk.query f . unFormatted
instance Walk.Walkable Formatted Formatted where
walk f = f
walkM f = f
query f = f
toStr :: String -> [Inline]
toStr = intercalate [Str "\n"] .
map (B.toList . B.text . T.pack . tweak . T.unpack . fromEntities) .
splitWhen (=='\n') . T.pack
where
tweak ('«':' ':xs) = "«\8239" ++ tweak xs
tweak (' ':'»':xs) = "\8239»" ++ tweak xs
tweak (' ':';':xs) = "\8239;" ++ tweak xs
tweak (' ':':':xs) = "\8239:" ++ tweak xs
tweak (' ':'!':xs) = "\8239!" ++ tweak xs
tweak (' ':'?':xs) = "\8239?" ++ tweak xs
tweak ( x :xs ) = x : tweak xs
tweak [] = []
appendWithPunct :: Formatted -> Formatted -> Formatted
appendWithPunct (Formatted left) (Formatted right) =
Formatted $
case (,) <$> lastleft <*> firstright of
Just (' ', d) | d `elem` (",.:;" :: String) -> initInline left ++ right
Just (c,d) | c `elem` (" ,.:;" :: String), d == c -> left ++ tailInline right
Just (c,'.') | c `elem` (",.!:;?" :: String) -> left ++ tailInline right
Just (c,':') | c `elem` (",!:;?" :: String) -> left ++ tailInline right
Just (c,'!') | c `elem` (",.!:;?" :: String) -> left ++ tailInline right
Just (c,'?') | c `elem` (",.!:;?" :: String) -> left ++ tailInline right
Just (c,';') | c `elem` (",:;" :: String) -> left ++ tailInline right
Just (':',c) | c `elem` (",.!:;?" :: String) -> left ++ tailInline right
Just (';',c) | c `elem` (",.!:;?" :: String) -> left ++ tailInline right
_ -> left ++ right
where lastleft = lastInline left
firstright = headInline right
data Style
= Style
{ styleVersion :: Text
, styleClass :: Text
, styleInfo :: Maybe CSInfo
, styleDefaultLocale :: Text
, styleLocale :: [Locale]
, styleAbbrevs :: Abbreviations
, csOptions :: [Option]
, csMacros :: [MacroMap]
, citation :: Citation
, biblio :: Maybe Bibliography
} deriving ( Show, Read, Typeable, Data, Generic )
data Locale
= Locale
{ localeVersion :: Text
, localeLang :: Text
, localeOptions :: [Option]
, localeTerms :: [CslTerm]
, localeDate :: [Element]
} deriving ( Show, Read, Eq, Typeable, Data, Generic )
mergeLocales :: Text -> Locale -> [Locale] -> [Locale]
mergeLocales s l ls = doMerge list
where
list = filter ((==) s . localeLang) ls ++
filter ((\x -> x /= "" && x `T.isPrefixOf` s) . localeLang) ls ++
filter ((==) "" . localeLang) ls
doMerge x = return l { localeOptions = newOpt x
, localeTerms = newTerms x
, localeDate = newDate x
}
cht = cslTerm &&& termForm &&& termGenderForm
checkedLoc = if hasOrdinals ls then rmOrdinals (localeTerms l) else localeTerms l
newTerms x = nubBy (\a b -> cht a == cht b) (concatMap localeTerms x ++ checkedLoc)
newOpt x = nubBy (\a b -> fst a == fst b) (concatMap localeOptions x ++ localeOptions l)
newDate x = nubBy (\(Date _ a _ _ _ _)
(Date _ b _ _ _ _) -> a == b) (concatMap localeDate x ++ localeDate l)
data CslTerm
= CT
{ cslTerm :: Text
, termForm :: Form
, termGender :: Gender
, termGenderForm :: Gender
, termSingular :: Text
, termPlural :: Text
, termMatch :: Text
} deriving ( Show, Read, Eq, Typeable, Data, Generic )
newTerm :: CslTerm
newTerm = CT "" Long Neuter Neuter "" "" ""
findTerm :: Text -> Form -> [CslTerm] -> Maybe CslTerm
findTerm s f = findTerm'' s f Nothing
findTerm' :: Text -> Form -> Gender -> [CslTerm] -> Maybe CslTerm
findTerm' s f g = findTerm'' s f (Just g)
findTerm'' :: Text -> Form -> Maybe Gender -> [CslTerm] -> Maybe CslTerm
findTerm'' s f mbg ts
= listToMaybe [ t | t <- ts, cslTerm t == s, termForm t == f,
isNothing mbg || mbg == Just (termGenderForm t) ]
`mplus`
case f of
VerbShort -> findTerm'' s Verb Nothing ts
Symbol -> findTerm'' s Short Nothing ts
Verb -> findTerm'' s Long Nothing ts
Short -> findTerm'' s Long Nothing ts
_ -> Nothing
hasOrdinals :: [Locale] -> Bool
hasOrdinals = any (any hasOrd . localeTerms)
where
hasOrd o
| CT {cslTerm = t} <- o
, "ordinal" `T.isInfixOf` t = True
| otherwise = False
rmOrdinals :: [CslTerm] -> [CslTerm]
rmOrdinals [] = []
rmOrdinals (o:os)
| CT {cslTerm = t} <- o
, "ordinal" `T.isInfixOf` t = rmOrdinals os
| otherwise = o:rmOrdinals os
newtype Abbreviations = Abbreviations {
unAbbreviations :: M.Map Text (M.Map Text (M.Map Text Text))
} deriving ( Show, Read, Typeable, Data, Generic )
instance FromJSON Abbreviations where
parseJSON (Object v) = Abbreviations <$> parseJSON (Object v)
parseJSON (Bool False) = return $ Abbreviations M.empty
parseJSON _ = fail "Could not read Abbreviations"
type MacroMap
= (Text,[Element])
data Citation
= Citation
{ citOptions :: [Option]
, citSort :: [Sort]
, citLayout :: Layout
} deriving ( Show, Read, Typeable, Data, Generic )
data Bibliography
= Bibliography
{ bibOptions :: [Option]
, bibSort :: [Sort]
, bibLayout :: Layout
} deriving ( Show, Read, Typeable, Data, Generic )
type Option = (Text,Text)
mergeOptions :: [Option] -> [Option] -> [Option]
mergeOptions os = nubBy (\x y -> fst x == fst y) . (++) os
data Layout
= Layout
{ layFormat :: Formatting
, layDelim :: Delimiter
, elements :: [Element]
} deriving ( Show, Read, Typeable, Data, Generic )
data Element
= Choose IfThen [IfThen] [Element]
| Macro Text Formatting
| Const Text Formatting
| Variable [Text] Form Formatting Delimiter
| Term Text Form Formatting Bool
| Label Text Form Formatting Plural
| Number Text NumericForm Formatting
| Names [Text ] [Name] Formatting Delimiter [Element]
| Substitute [Element]
| Group Formatting Delimiter [Element]
| Date [Text ] DateForm Formatting Delimiter [DatePart] Text
deriving ( Show, Read, Eq, Typeable, Data, Generic )
data IfThen
= IfThen Condition Match [Element]
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data Condition
= Condition
{ isType :: [Text]
, isSet :: [Text]
, isNumeric :: [Text]
, isUncertainDate :: [Text]
, isPosition :: [Text]
, disambiguation :: [Text]
, isLocator :: [Text]
} deriving ( Eq, Show, Read, Typeable, Data, Generic )
type Delimiter = Text
data Match
= Any
| All
| None
deriving ( Show, Read, Eq, Typeable, Data, Generic )
match :: Match -> [Bool] -> Bool
match All = and
match Any = or
match None = all not
data DatePart
= DatePart
{ dpName :: Text
, dpForm :: Text
, dpRangeDelim :: Text
, dpFormatting :: Formatting
} deriving ( Show, Read, Eq, Typeable, Data, Generic )
defaultDate :: [DatePart]
defaultDate
= [ DatePart "year" "" "-" emptyFormatting
, DatePart "month" "" "-" emptyFormatting
, DatePart "day" "" "-" emptyFormatting]
data Sort
= SortVariable Text Sorting
| SortMacro Text Sorting Int Int Text
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data Sorting
= Ascending Text
| Descending Text
deriving ( Read, Show, Eq, Typeable, Data, Generic )
instance Ord Sorting where
compare (Ascending "") (Ascending "") = EQ
compare (Ascending "") (Ascending _) = GT
compare (Ascending _) (Ascending "") = LT
compare (Ascending a) (Ascending b) = compare' a b
compare (Descending "") (Descending "") = EQ
compare (Descending "") (Descending _) = GT
compare (Descending _) (Descending "") = LT
compare (Descending a) (Descending b) = compare' b a
compare _ _ = EQ
compare' :: Text -> Text -> Ordering
compare' x' y'
= case (x, y) of
('-':_,'-':_) -> comp (normalize y) (normalize x)
('-':_, _ ) -> LT
(_ ,'-':_) -> GT
_ -> comp (normalize x) (normalize y)
where
x = T.unpack x'
y = T.unpack y'
zeropad [] = []
zeropad xs = if all isDigit xs
then replicate (10 - length xs) '0' ++ xs
else xs
normalize = zeropad .
map (\c -> if c == ',' || c == '.' then ' ' else c) .
filter (\c -> c == ',' ||
not (isPunctuation c || Char.isSpace c
|| c == 'ʾ' || c == 'ʿ'
))
#ifdef UNICODE_COLLATION
comp a b = T.collate (T.collator T.Current) (T.pack a) (T.pack b)
#else
comp a b = compareUnicode (map toLower a) (map toLower b)
#endif
data Form
= Long
| Short
| Count
| Verb
| VerbShort
| Symbol
| NotSet
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data Gender
= Feminine
| Masculine
| Neuter
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data NumericForm
= Numeric
| Ordinal
| Roman
| LongOrdinal
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data DateForm
= TextDate
| NumericDate
| NoFormDate
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data Plural
= Contextual
| Always
| Never
deriving ( Eq, Show, Read, Typeable, Data, Generic )
data Name
= Name Form Formatting NameAttrs Delimiter [NamePart]
| NameLabel Form Formatting Plural
| EtAl Formatting Text
deriving ( Eq, Show, Read, Typeable, Data, Generic )
type NameAttrs = [(Text, Text)]
data NamePart
= NamePart Text Formatting
deriving ( Show, Read, Eq, Typeable, Data, Generic )
isPlural :: Plural -> Int -> Bool
isPlural p l
= case p of
Always -> True
Never -> False
Contextual -> l > 1
isName :: Name -> Bool
isName x = case x of Name {} -> True; _ -> False
isNames :: Element -> Bool
isNames x = case x of Names {} -> True; _ -> False
hasEtAl :: [Name] -> Bool
hasEtAl = any isEtAl
where isEtAl (EtAl _ _) = True
isEtAl _ = False
data Formatting
= Formatting
{ prefix :: Text
, suffix :: Text
, fontFamily :: Text
, fontStyle :: Text
, fontVariant :: Text
, fontWeight :: Text
, textDecoration :: Text
, verticalAlign :: Text
, textCase :: Text
, display :: Text
, quotes :: Quote
, stripPeriods :: Bool
, noCase :: Bool
, noDecor :: Bool
, hyperlink :: Text
} deriving ( Read, Eq, Ord, Typeable, Data, Generic )
instance Show Formatting where
show x
| x == emptyFormatting = "emptyFormatting"
| otherwise = "emptyFormatting{"
++ intercalate ", "
[ k ++ " = " ++ f x |
(k, f) <- [("prefix", show . prefix)
,("suffix", show . suffix)
,("fontFamily", show . fontFamily)
,("fontStyle", show . fontStyle)
,("fontVariant", show . fontVariant)
,("fontWeight", show . fontWeight)
,("textDecoration", show . textDecoration)
,("verticalAlign", show . verticalAlign)
,("textCase", show . textCase)
,("display", show . display)
,("quotes", show . quotes)
,("stripPeriods", show . stripPeriods)
,("noCase", show . noCase)
,("noDecor", show . noDecor)
,("hyperlink", show . hyperlink)],
f x /= f emptyFormatting ]
++ "}"
rmTitleCase :: Formatting -> Formatting
rmTitleCase f = f{ textCase = if textCase f == "title" then "" else textCase f }
rmTitleCase' :: Output -> Output
rmTitleCase' o =
case o of
OStr s f -> OStr s (rmTitleCase f)
Output os f -> Output os (rmTitleCase f)
_ -> o
data Quote
= NativeQuote
| ParsedQuote
| NoQuote
deriving ( Show, Read, Eq, Ord, Typeable, Data, Generic )
emptyFormatting :: Formatting
emptyFormatting
= Formatting "" "" "" "" "" "" "" "" "" "" NoQuote False False False ""
mergeFM :: Formatting -> Formatting -> Formatting
mergeFM (Formatting aa ab ac ad ae af ag ah ai aj ak al am an ahl)
(Formatting ba bb bc bd be bf bg bh bi bj bk bl bm bn bhl) =
Formatting (ba `orIfEmpty` aa)
(bb `orIfEmpty` ab)
(bc `orIfEmpty` ac)
(bd `orIfEmpty` ad)
(be `orIfEmpty` ae)
(bf `orIfEmpty` af)
(bg `orIfEmpty` ag)
(bh `orIfEmpty` ah)
(bi `orIfEmpty` ai)
(bj `orIfEmpty` aj)
(if bk == NoQuote then ak else bk)
(bl || al)
(bm || am)
(bn || an)
(bhl <> ahl)
where orIfEmpty :: Text -> Text -> Text
orIfEmpty "" fallback = fallback
orIfEmpty t _ = t
data CSInfo
= CSInfo
{ csiTitle :: Text
, csiAuthor :: CSAuthor
, csiCategories :: [CSCategory]
, csiId :: Text
, csiUpdated :: Text
} deriving ( Show, Read, Typeable, Data, Generic )
data CSAuthor = CSAuthor Text Text Text
deriving ( Show, Read, Eq, Typeable, Data, Generic )
data CSCategory = CSCategory Text Text Text
deriving ( Show, Read, Eq, Typeable, Data, Generic )
data CiteprocError
= NoOutput
| ReferenceNotFound Text
deriving ( Eq, Ord, Show, Typeable, Data, Generic )
data Output
= ONull
| OSpace
| OPan [Inline]
| OStatus [Inline]
| ODel Text
| OStr Text Formatting
| OErr CiteprocError
| OLabel Text Formatting
| ONum Int Formatting
| OCitNum Int Formatting
| OCitLabel Text Formatting
| ODate [Output]
| OYear Text Text Formatting
| OYearSuf Text Text [Output] Formatting
| OName Agent [Output] [[Output]] Formatting
| OContrib Text Text [Output] [Output] [[Output]]
| OLoc [Output] Formatting
| Output [Output] Formatting
deriving ( Eq, Ord, Show, Typeable, Data, Generic )
type Citations = [[Cite]]
data Cite
= Cite
{ citeId :: Text
, citePrefix :: Formatted
, citeSuffix :: Formatted
, citeLabel :: Text
, citeLocator :: Text
, citeNoteNumber :: Text
, citePosition :: Text
, nearNote :: Bool
, authorInText :: Bool
, suppressAuthor :: Bool
, citeHash :: Int
} deriving ( Show, Eq, Typeable, Data, Generic )
instance FromJSON Cite where
parseJSON (Object v) = Cite <$>
v .#: "id" <*>
v .:? "prefix" .!= mempty <*>
v .:? "suffix" .!= mempty <*>
v .#? "label" .!= "page" <*>
v .#? "locator" .!= "" <*>
v .#? "note-number" .!= "" <*>
v .#? "position" .!= "" <*>
(v .:? "near-note" >>= mb parseBool) .!= False <*>
(v .:? "author-in-text" >>= mb parseBool) .!= False <*>
(v .:? "suppress-author" >>= mb parseBool) .!= False <*>
v .:? "cite-hash" .!= 0
parseJSON _ = fail "Could not parse Cite"
instance OVERLAPS
FromJSON [[Cite]] where
parseJSON (Array v) = mapM parseJSON $ V.toList v
parseJSON _ = return []
emptyCite :: Cite
emptyCite = Cite "" mempty mempty "" "" "" "" False False False 0
data CitationGroup = CG [(Cite, Output)] Formatting Delimiter [(Cite, Output)] deriving ( Show, Eq, Typeable, Data, Generic )
data BiblioData
= BD
{ citations :: [Formatted]
, bibliography :: [Formatted]
, citationIds :: [Text]
} deriving ( Show, Typeable, Data, Generic )
data CiteData
= CD
{ key :: Text
, collision :: [Output]
, disambYS :: [Output]
, disambData :: [[Output]]
, disambed :: [Output]
, sameAs :: [Text]
, citYear :: Text
} deriving ( Show, Typeable, Data, Generic )
instance Eq CiteData where
(==) (CD ka ca _ _ _ _ _)
(CD kb cb _ _ _ _ _) = ka == kb && ca == cb
data NameData
= ND
{ nameKey :: Agent
, nameCollision :: [Output]
, nameDisambData :: [[Output]]
, nameDataSolved :: [Output]
} deriving ( Show, Typeable, Data, Generic )
instance Eq NameData where
(==) (ND ka ca _ _)
(ND kb cb _ _) = ka == kb && ca == cb
isPunctuationInQuote :: Style -> Bool
isPunctuationInQuote sty =
case styleLocale sty of
(l:_) -> ("punctuation-in-quote","true") `elem` localeOptions l
_ -> False
object' :: [Pair] -> Aeson.Value
object' = object . filter (not . isempty)
where isempty (_, Array v) = V.null v
isempty (_, String t) = T.null t
isempty ("first-reference-note-number", Aeson.Number n) = n == 0
isempty ("citation-number", Aeson.Number n) = n == 0
isempty (_, _) = False
data Agent
= Agent { givenName :: [Formatted]
, droppingPart :: Formatted
, nonDroppingPart :: Formatted
, familyName :: Formatted
, nameSuffix :: Formatted
, literal :: Formatted
, commaSuffix :: Bool
, parseNames :: Bool
}
deriving ( Show, Read, Eq, Ord, Typeable, Data, Generic )
emptyAgent :: Agent
emptyAgent = Agent [] mempty mempty mempty mempty mempty False False
removeQuoted :: Formatted -> Formatted
removeQuoted (Formatted ils) = Formatted (go ils)
where go [] = []
go (Quoted DoubleQuote ils' : xs) = ils' ++ go xs
go (x:xs) = x : go xs
instance FromJSON Agent where
parseJSON (Object v) = nameTransform <$> (Agent <$>
(v .: "given" <|> ((map Formatted . wordsBy isSpace .
addSpaceAfterPeriod . unFormatted) <$>
v .: "given") <|> pure []) <*>
v .:? "dropping-particle" .!= mempty <*>
v .:? "non-dropping-particle" .!= mempty <*>
(removeQuoted <$> (v .:? "family" .!= mempty)) <*>
v .:? "suffix" .!= mempty <*>
v .:? "literal" .!= mempty <*>
(v .:? "comma-suffix" >>= mb parseBool) .!= False <*>
(v .:? "parse-names" >>= mb parseBool) .!= False)
parseJSON _ = fail "Could not parse Agent"
instance ToYaml Agent where
toYaml ag = mapping' [ "family" &= familyName ag
, case givenName ag of
[] -> id
xs -> "given" &= Formatted
(intercalate [Space]
(map unFormatted xs))
, "non-dropping-particle" &= nonDroppingPart ag
, "dropping-particle" &= droppingPart ag
, "suffix" &= nameSuffix ag
, "literal" &= literal ag
, "comma-suffix" &= T.pack (if commaSuffix ag
then "true"
else "")
, "parse-names" &= T.pack (if parseNames ag
then "true"
else "")
]
nameTransform :: Agent -> Agent
nameTransform ag
| parseNames ag = nonDroppingPartTransform .
droppingPartTransform .
suffixTransform $ ag{ parseNames = False }
| otherwise = ag
nonDroppingPartTransform :: Agent -> Agent
nonDroppingPartTransform ag
| nonDroppingPart ag == mempty =
case break startWithCapital'
(splitStrWhen (\c -> isPunctuation c || isUpper c) $
unFormatted $ familyName ag) of
([], _) -> ag
(xs, ys)
| lastInline xs `elem` map Just (" -'’" :: String) -> ag {
nonDroppingPart = Formatted $ trimSpace xs,
familyName = Formatted ys }
| otherwise -> ag
| otherwise = ag
trimSpace :: [Inline] -> [Inline]
trimSpace = reverse . dropWhile isSpace . reverse . dropWhile isSpace
isSpace :: Inline -> Bool
isSpace Space = True
isSpace SoftBreak = True
isSpace _ = False
droppingPartTransform :: Agent -> Agent
droppingPartTransform ag
| droppingPart ag == mempty =
case break startWithCapital $ reverse $ givenName ag of
([],_) -> ag
(ys,zs) -> ag{ droppingPart = mconcat $
intersperse (Formatted [Space]) $
reverse ys
, givenName = reverse zs }
| otherwise = ag
startWithCapital' :: Inline -> Bool
startWithCapital' (Str (T.uncons -> Just (c,_))) = isUpper c && isLetter c
startWithCapital' _ = False
startWithCapital :: Formatted -> Bool
startWithCapital (Formatted (x:_)) = startWithCapital' x
startWithCapital _ = False
stripFinalComma :: Formatted -> (Text, Formatted)
stripFinalComma (Formatted ils) =
case reverse $ splitStrWhen isPunctuation ils of
Str ",":xs -> (",", Formatted $ reverse xs)
Str "!":Str ",":xs -> (",!", Formatted $ reverse xs)
_ -> ("", Formatted ils)
suffixTransform :: Agent -> Agent
suffixTransform ag
| nameSuffix ag == mempty = fst $ foldl go
(ag{ givenName = mempty
, nameSuffix = mempty
, commaSuffix = False }, False)
(givenName ag)
| otherwise = ag
where go (ag', False) n =
case stripFinalComma n of
("", _) -> (ag'{ givenName = givenName ag' ++ [n] }, False)
(",",n') -> (ag'{ givenName = givenName ag' ++ [n'] }, True)
(",!",n') -> (ag'{ givenName = givenName ag' ++ [n']
, commaSuffix = True }, True)
_ -> error "stripFinalComma returned unexpected value"
go (ag', True) n = (ag'{ nameSuffix = if nameSuffix ag' == mempty
then n
else nameSuffix ag' <>
Formatted [Space] <> n }, True)
instance ToJSON Agent where
toJSON agent = object' $ [
"given" .= Formatted (intercalate [Space] $ map unFormatted
$ givenName agent)
, "dropping-particle" .= droppingPart agent
, "non-dropping-particle" .= nonDroppingPart agent
, "family" .= familyName agent
, "suffix" .= nameSuffix agent
, "literal" .= literal agent
] ++ ["comma-suffix" .= commaSuffix agent | nameSuffix agent /= mempty]
++ ["parse-names" .= True | parseNames agent ]
instance OVERLAPS
FromJSON [Agent] where
parseJSON (Array xs) = mapM parseJSON $ V.toList xs
parseJSON (Object v) = (:[]) `fmap` parseJSON (Object v)
parseJSON _ = fail "Could not parse [Agent]"