{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE OverloadedStrings #-}
module Text.Pandoc.Citeproc.Util
( splitStrWhen
, toIETF )
where
import qualified Data.Text as T
import Data.Text (Text)
import Text.Pandoc.Definition
splitStrWhen :: (Char -> Bool) -> [Inline] -> [Inline]
splitStrWhen :: (Char -> Bool) -> [Inline] -> [Inline]
splitStrWhen Char -> Bool
p = (Inline -> [Inline] -> [Inline])
-> [Inline] -> [Inline] -> [Inline]
forall a b. (a -> b -> b) -> b -> [a] -> b
forall (t :: * -> *) a b.
Foldable t =>
(a -> b -> b) -> b -> t a -> b
foldr Inline -> [Inline] -> [Inline]
go []
where
go :: Inline -> [Inline] -> [Inline]
go (Str Text
t) = ((Text -> Inline) -> [Text] -> [Inline]
forall a b. (a -> b) -> [a] -> [b]
map Text -> Inline
Str ((Char -> Char -> Bool) -> Text -> [Text]
T.groupBy Char -> Char -> Bool
goesTogether Text
t) [Inline] -> [Inline] -> [Inline]
forall a. [a] -> [a] -> [a]
++)
go Inline
x = (Inline
x Inline -> [Inline] -> [Inline]
forall a. a -> [a] -> [a]
:)
goesTogether :: Char -> Char -> Bool
goesTogether Char
c Char
d = Bool -> Bool
not (Char -> Bool
p Char
c Bool -> Bool -> Bool
|| Char -> Bool
p Char
d)
toIETF :: Text -> Text
toIETF :: Text -> Text
toIETF Text
"english" = Text
"en-US"
toIETF Text
"usenglish" = Text
"en-US"
toIETF Text
"american" = Text
"en-US"
toIETF Text
"british" = Text
"en-GB"
toIETF Text
"ukenglish" = Text
"en-GB"
toIETF Text
"canadian" = Text
"en-US"
toIETF Text
"australian" = Text
"en-GB"
toIETF Text
"newzealand" = Text
"en-GB"
toIETF Text
"afrikaans" = Text
"af-ZA"
toIETF Text
"arabic" = Text
"ar"
toIETF Text
"basque" = Text
"eu"
toIETF Text
"bulgarian" = Text
"bg-BG"
toIETF Text
"catalan" = Text
"ca-AD"
toIETF Text
"croatian" = Text
"hr-HR"
toIETF Text
"czech" = Text
"cs-CZ"
toIETF Text
"danish" = Text
"da-DK"
toIETF Text
"dutch" = Text
"nl-NL"
toIETF Text
"estonian" = Text
"et-EE"
toIETF Text
"finnish" = Text
"fi-FI"
toIETF Text
"canadien" = Text
"fr-CA"
toIETF Text
"acadian" = Text
"fr-CA"
toIETF Text
"french" = Text
"fr-FR"
toIETF Text
"francais" = Text
"fr-FR"
toIETF Text
"austrian" = Text
"de-AT"
toIETF Text
"naustrian" = Text
"de-AT"
toIETF Text
"german" = Text
"de-DE"
toIETF Text
"germanb" = Text
"de-DE"
toIETF Text
"ngerman" = Text
"de-DE"
toIETF Text
"greek" = Text
"el-GR"
toIETF Text
"polutonikogreek" = Text
"el-GR"
toIETF Text
"hebrew" = Text
"he-IL"
toIETF Text
"hungarian" = Text
"hu-HU"
toIETF Text
"icelandic" = Text
"is-IS"
toIETF Text
"italian" = Text
"it-IT"
toIETF Text
"japanese" = Text
"ja-JP"
toIETF Text
"latvian" = Text
"lv-LV"
toIETF Text
"lithuanian" = Text
"lt-LT"
toIETF Text
"magyar" = Text
"hu-HU"
toIETF Text
"mongolian" = Text
"mn-MN"
toIETF Text
"norsk" = Text
"nb-NO"
toIETF Text
"nynorsk" = Text
"nn-NO"
toIETF Text
"farsi" = Text
"fa-IR"
toIETF Text
"polish" = Text
"pl-PL"
toIETF Text
"brazil" = Text
"pt-BR"
toIETF Text
"brazilian" = Text
"pt-BR"
toIETF Text
"portugues" = Text
"pt-PT"
toIETF Text
"portuguese" = Text
"pt-PT"
toIETF Text
"romanian" = Text
"ro-RO"
toIETF Text
"russian" = Text
"ru-RU"
toIETF Text
"serbian" = Text
"sr-RS"
toIETF Text
"serbianc" = Text
"sr-RS"
toIETF Text
"slovak" = Text
"sk-SK"
toIETF Text
"slovene" = Text
"sl-SL"
toIETF Text
"spanish" = Text
"es-ES"
toIETF Text
"swedish" = Text
"sv-SE"
toIETF Text
"thai" = Text
"th-TH"
toIETF Text
"turkish" = Text
"tr-TR"
toIETF Text
"ukrainian" = Text
"uk-UA"
toIETF Text
"vietnamese" = Text
"vi-VN"
toIETF Text
"latin" = Text
"la"
toIETF Text
x = Text
x