{-# LANGUAGE Safe #-} -- | -- Module : Data.Text.Lazy.Encoding.Base64.URL -- Copyright : (c) 2019-2020 Emily Pillmore -- License : BSD-style -- -- Maintainer : Emily Pillmore <emilypi@cohomolo.gy> -- Stability : stable -- Portability : non-portable -- -- This module contains 'Data.Text.Lazy.Text'-valued combinators for -- implementing the RFC 4648 specification of the Base64url -- encoding format. This includes strictly padded/unpadded and lenient decoding -- variants, as well as internal and external validation for canonicity. -- module Data.Text.Lazy.Encoding.Base64.URL ( -- * Encoding encodeBase64 , encodeBase64Unpadded -- * Decoding , decodeBase64 , decodeBase64With , decodeBase64Unpadded , decodeBase64UnpaddedWith , decodeBase64Padded , decodeBase64PaddedWith , decodeBase64Lenient -- * Validation , isBase64Url , isValidBase64Url ) where import Data.Bifunctor (first) import Data.ByteString.Lazy (ByteString) import qualified Data.ByteString.Lazy.Base64.URL as BL64U import qualified Data.Text as T import Data.Text.Encoding.Base64.Error import qualified Data.Text.Lazy as TL import qualified Data.Text.Lazy.Encoding as TL -- | Encode a 'TL.Text' value in Base64url with padding. -- -- See: <https://tools.ietf.org/html/rfc4648#section-5 RFC-4648 section 5> -- -- === __Examples__: -- -- >>> encodeBase64 "<<?>>" -- "PDw_Pj4=" -- encodeBase64 :: TL.Text -> TL.Text encodeBase64 :: Text -> Text encodeBase64 = ByteString -> Text BL64U.encodeBase64 forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE encodeBase64 #-} -- | Decode a padded Base64url-encoded 'TL.Text' value. If its length is not a multiple -- of 4, then padding chars will be added to fill out the input to a multiple of -- 4 for safe decoding as base64url encodings are optionally padded. -- -- For a decoder that fails on unpadded input, use 'decodeBase64Unpadded'. -- -- /Note:/ This function makes sure that decoding is total by deferring to -- 'T.decodeLatin1'. This will always round trip for any valid Base64-encoded -- text value, but it may not round trip for bad inputs. The onus is on the -- caller to make sure inputs are valid. If unsure, defer to `decodeBase64With` -- and pass in a custom decode function. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Examples__: -- -- >>> decodeBase64 "PDw_Pj4=" -- Right "<<?>>" -- -- >>> decodeBase64 "PDw_Pj4" -- Right "<<?>>" -- -- >>> decodeBase64 "PDw-Pg=" -- Left "Base64-encoded bytestring has invalid padding" -- -- >>> decodeBase64 "PDw-Pg" -- Right "<<>>" -- decodeBase64 :: TL.Text -> Either T.Text TL.Text decodeBase64 :: Text -> Either Text Text decodeBase64 = forall (f :: * -> *) a b. Functor f => (a -> b) -> f a -> f b fmap ByteString -> Text TL.decodeLatin1 forall b c a. (b -> c) -> (a -> b) -> a -> c . ByteString -> Either Text ByteString BL64U.decodeBase64 forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE decodeBase64 #-} -- | Attempt to decode a lazy 'ByteString' value as Base64url, converting from -- 'ByteString' to 'TL.Text' according to some encoding function. In practice, -- This is something like 'decodeUtf8'', which may produce an error. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Examples__: -- -- @ -- 'decodeBase64With' 'TL.decodeUtf8'' -- :: 'ByteString' -> 'Either' ('Base64Error' 'UnicodeException') 'TL.Text' -- @ -- decodeBase64With :: (ByteString -> Either err TL.Text) -- ^ convert a bytestring to text (e.g. 'TL.decodeUtf8'') -> ByteString -- ^ Input text to decode -> Either (Base64Error err) TL.Text decodeBase64With :: forall err. (ByteString -> Either err Text) -> ByteString -> Either (Base64Error err) Text decodeBase64With ByteString -> Either err Text f ByteString t = case ByteString -> Either Text ByteString BL64U.decodeBase64 ByteString t of Left Text de -> forall a b. a -> Either a b Left forall a b. (a -> b) -> a -> b $ forall e. Text -> Base64Error e DecodeError Text de Right ByteString a -> forall (p :: * -> * -> *) a b c. Bifunctor p => (a -> b) -> p a c -> p b c first forall e. e -> Base64Error e ConversionError (ByteString -> Either err Text f ByteString a) {-# INLINE decodeBase64With #-} -- | Encode a 'TL.Text' value in Base64url without padding. Note that for Base64url, -- padding is optional. If you call this function, you will simply be encoding -- as Base64url and stripping padding chars from the output. -- -- See: <https://tools.ietf.org/html/rfc4648#section-3.2 RFC-4648 section 3.2> -- -- === __Examples__: -- -- >>> encodeBase64Unpadded "<<?>>" -- "PDw_Pj4" -- encodeBase64Unpadded :: TL.Text -> TL.Text encodeBase64Unpadded :: Text -> Text encodeBase64Unpadded = ByteString -> Text BL64U.encodeBase64Unpadded forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE encodeBase64Unpadded #-} -- | Decode an unpadded Base64url encoded 'TL.Text' value. -- -- /Note:/ This function makes sure that decoding is total by deferring to -- 'T.decodeLatin1'. This will always round trip for any valid Base64-encoded -- text value, but it may not round trip for bad inputs. The onus is on the -- caller to make sure inputs are valid. If unsure, defer to `decodeBase64WUnpaddedWith` -- and pass in a custom decode function. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Examples__: -- -- >>> decodeBase64Unpadded "PDw_Pj4" -- Right "<<?>>" -- -- >>> decodeBase64Unpadded "PDw_Pj4=" -- Left "Base64-encoded bytestring has invalid padding" -- decodeBase64Unpadded :: TL.Text -> Either T.Text TL.Text decodeBase64Unpadded :: Text -> Either Text Text decodeBase64Unpadded = forall (f :: * -> *) a b. Functor f => (a -> b) -> f a -> f b fmap ByteString -> Text TL.decodeLatin1 forall b c a. (b -> c) -> (a -> b) -> a -> c . ByteString -> Either Text ByteString BL64U.decodeBase64Unpadded forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE decodeBase64Unpadded #-} -- | Attempt to decode an unpadded lazy 'ByteString' value as Base64url, converting from -- 'ByteString' to 'TL.Text' according to some encoding function. In practice, -- This is something like 'decodeUtf8'', which may produce an error. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Examples__: -- -- @ -- 'decodeBase64UnpaddedWith' 'TL.decodeUtf8'' -- :: 'ByteString' -> 'Either' ('Base64Error' 'UnicodeException') 'TL.Text' -- @ -- decodeBase64UnpaddedWith :: (ByteString -> Either err TL.Text) -- ^ convert a bytestring to text (e.g. 'TL.decodeUtf8'') -> ByteString -- ^ Input text to decode -> Either (Base64Error err) TL.Text decodeBase64UnpaddedWith :: forall err. (ByteString -> Either err Text) -> ByteString -> Either (Base64Error err) Text decodeBase64UnpaddedWith ByteString -> Either err Text f ByteString t = case ByteString -> Either Text ByteString BL64U.decodeBase64Unpadded ByteString t of Left Text de -> forall a b. a -> Either a b Left forall a b. (a -> b) -> a -> b $ forall e. Text -> Base64Error e DecodeError Text de Right ByteString a -> forall (p :: * -> * -> *) a b c. Bifunctor p => (a -> b) -> p a c -> p b c first forall e. e -> Base64Error e ConversionError (ByteString -> Either err Text f ByteString a) {-# INLINE decodeBase64UnpaddedWith #-} -- | Decode an padded Base64url encoded 'TL.Text' value -- -- /Note:/ This function makes sure that decoding is total by deferring to -- 'T.decodeLatin1'. This will always round trip for any valid Base64-encoded -- text value, but it may not round trip for bad inputs. The onus is on the -- caller to make sure inputs are valid. If unsure, defer to `decodeBase64PaddedWith` -- and pass in a custom decode function. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Examples__: -- -- >>> decodeBase64Padded "PDw_Pj4=" -- Right "<<?>>" -- -- >>> decodeBase64Padded "PDw_Pj4" -- Left "Base64-encoded bytestring requires padding" -- decodeBase64Padded :: TL.Text -> Either T.Text TL.Text decodeBase64Padded :: Text -> Either Text Text decodeBase64Padded = forall (f :: * -> *) a b. Functor f => (a -> b) -> f a -> f b fmap ByteString -> Text TL.decodeLatin1 forall b c a. (b -> c) -> (a -> b) -> a -> c . ByteString -> Either Text ByteString BL64U.decodeBase64Padded forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE decodeBase64Padded #-} -- | Attempt to decode a padded lazy 'ByteString' value as Base64url, converting from -- 'ByteString' to 'TL.Text' according to some encoding function. In practice, -- This is something like 'decodeUtf8'', which may produce an error. -- -- See: <https://tools.ietf.org/html/rfc4648#section-4 RFC-4648 section 4> -- -- === __Example__: -- -- @ -- 'decodeBase64PaddedWith' 'T.decodeUtf8'' -- :: 'ByteString' -> 'Either' ('Base64Error' 'UnicodeException') 'Text' -- @ -- decodeBase64PaddedWith :: (ByteString -> Either err TL.Text) -- ^ convert a bytestring to text (e.g. 'TL.decodeUtf8'') -> ByteString -- ^ Input text to decode -> Either (Base64Error err) TL.Text decodeBase64PaddedWith :: forall err. (ByteString -> Either err Text) -> ByteString -> Either (Base64Error err) Text decodeBase64PaddedWith ByteString -> Either err Text f ByteString t = case ByteString -> Either Text ByteString BL64U.decodeBase64Padded ByteString t of Left Text de -> forall a b. a -> Either a b Left forall a b. (a -> b) -> a -> b $ forall e. Text -> Base64Error e DecodeError Text de Right ByteString a -> forall (p :: * -> * -> *) a b c. Bifunctor p => (a -> b) -> p a c -> p b c first forall e. e -> Base64Error e ConversionError (ByteString -> Either err Text f ByteString a) {-# INLINE decodeBase64PaddedWith #-} -- | Leniently decode an unpadded Base64url-encoded 'TL.Text'. This function -- will not generate parse errors. If input data contains padding chars, -- then the input will be parsed up until the first pad character. -- -- __Note:__ This is not RFC 4648-compliant. -- -- === __Examples__: -- -- >>> decodeBase64Lenient "PDw_Pj4=" -- "<<?>>" -- -- >>> decodeBase64Lenient "PDw_%%%$}Pj4" -- "<<?>>" -- decodeBase64Lenient :: TL.Text -> TL.Text decodeBase64Lenient :: Text -> Text decodeBase64Lenient = ByteString -> Text TL.decodeLatin1 forall b c a. (b -> c) -> (a -> b) -> a -> c . ByteString -> ByteString BL64U.decodeBase64Lenient forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE decodeBase64Lenient #-} -- | Tell whether a 'TL.Text' value is Base64url-encoded -- -- === __Examples__: -- -- >>> isBase64Url "PDw_Pj4=" -- True -- -- >>> isBase64Url "PDw_Pj4" -- True -- -- >>> isBase64Url "PDw_Pj" -- False -- isBase64Url :: TL.Text -> Bool isBase64Url :: Text -> Bool isBase64Url = ByteString -> Bool BL64U.isBase64Url forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE isBase64Url #-} -- | Tell whether a 'TL.Text' value is a valid Base64url format. -- -- This will not tell you whether or not this is a correct Base64url representation, -- only that it conforms to the correct shape. To check whether it is a true -- Base64 encoded 'TL.Text' value, use 'isBase64Url'. -- -- === __Examples__: -- -- >>> isValidBase64Url "PDw_Pj4=" -- True -- -- >>> isValidBase64Url "PDw_Pj" -- True -- -- >>> isValidBase64Url "%" -- False -- isValidBase64Url :: TL.Text -> Bool isValidBase64Url :: Text -> Bool isValidBase64Url = ByteString -> Bool BL64U.isValidBase64Url forall b c a. (b -> c) -> (a -> b) -> a -> c . Text -> ByteString TL.encodeUtf8 {-# INLINE isValidBase64Url #-}