This module uses the stream decoding functions from streaming-commons package to define decoding functions and lenses. The exported names conflict with names in Data.Text.Encoding but not with the Prelude

Encoding is of course simple. Given

  text :: Producer Text IO ()

we can encode it with Data.Text.Encoding.encodeUtf8

  TE.encodeUtf8 :: Text -> ByteString

and ordinary pipe operations:

  text >-> P.map TE.encodeUtf8 :: Producer.ByteString IO ()

or, equivalently

  for text (yield . TE.encodeUtf8)

But, using this module, we might use

  encodeUtf8 :: Text -> Producer ByteString m ()

to write

  for text encodeUtf8 :: Producer.ByteString IO ()

All of the above come to the same.

Given

  bytes :: Producer ByteString IO ()

we can apply a decoding function from this module:

  decodeUtf8 bytes :: Producer Text IO (Producer ByteString IO ())

The Text producer ends wherever decoding first fails. The un-decoded material is returned. If we are confident it is of no interest, we can write:

  void $ decodeUtf8 bytes :: Producer Text IO ()

Thus we can re-encode as uft8 as much of our byte stream as is decodeUtf16BE decodable, with, e.g.

  for (decodeUtf16BE bytes) encodeUtf8 :: Producer ByteString IO (Producer ByteString IO ())

The bytestring producer that is returned begins with where utf16BE decoding failed; if it didn't fail the producer is empty.

We get a bit more flexibility, particularly in the use of pipes-style "parsers", if we use a lens like utf8 or utf16BE that focusses on the text in an appropriately encoded byte stream.

  type Lens' a b = forall f . Functor f => (b -> f b) -> (a -> f a)

is just an alias for a Prelude type. We abbreviate this further, for our use case, as

  type Codec
    =  forall m r .  Monad m => Lens' (Producer ByteString m r) (Producer Text m (Producer ByteString m r))

and call the decoding lenses utf8, utf16BE "codecs", since they can re-encode what they have decoded. Thus you use any particular codec with the view / (^.) , zoom and over functions from the standard lens libraries; lens, lens-family, lens-simple, or one of the and microlens packages will all work the same, since we already have access to the types they require.

Each decoding lens looks into a byte stream that is supposed to contain text. The particular lenses are named in accordance with the expected encoding, utf8, utf16LE etc. To turn a such a lens or Codec into an ordinary function, use view / (^.) -- here also called decode:

  view utf8 :: Producer ByteString m r -> Producer Text m (Producer ByteString m r)
  decode utf8 Byte.stdin :: Producer Text IO (Producer ByteString IO r)
  Bytes.stdin ^. utf8 ::  Producer Text IO (Producer ByteString IO r)

Of course, we could always do this with the specialized decoding functions, e.g.

  decodeUtf8 ::  Producer ByteString m r -> Producer Text m (Producer ByteString m r)
  decodeUtf8 Byte.stdin :: Producer Text IO (Producer ByteString IO r)

As with these functions, the stream of text that a Codec 'sees' in the stream of bytes begins at its head. At any point of decoding failure, the stream of text ends and reverts to (returns) the original byte stream. Thus if the first bytes are already un-decodable, the whole ByteString producer will be returned, i.e.

  view utf8 bad_bytestream

will just come to the same as

  return bad_bytestream

Where there is no decoding failure, the return value of the text stream will be an empty byte stream followed by its own return value. In all cases you must deal with the fact that it is a ByteString producer that is returned, even if it can be thrown away with Control.Monad.void

  void (Bytes.stdin ^. utf8) :: Producer Text IO ()

The eof lens permits you to pattern match: if there is a Right value, it is the leftover bytestring producer, if there is a Right value, it is the return value of the original bytestring producer:

  Bytes.stdin ^. utf8 . eof :: Producer Text IO (Either (Producer ByteString IO IO) ())

Thus for the stream of un-decodable bytes mentioned above,

  view (utf8 . eof) bad_bytestream

will be the same as

  return (Left bad_bytestream)

zoom utf8 converts a Text parser into a ByteString parser:

  zoom utf8 drawChar :: Monad m => StateT (Producer ByteString m r) m (Maybe Char)

or, using the type synonymn from Pipes.Parse:

  zoom utf8 drawChar :: Monad m => Parser ByteString m (Maybe Char)

Thus we can define a ByteString parser (in the pipes-parse sense) like this:

  charPlusByte :: Parser ByteString m (Maybe Char, Maybe Word8)))
  charPlusByte = do char_ <- zoom utf8 Text.drawChar
                    byte_ <- Bytes.peekByte
                    return (char_, byte_)

Though charPlusByte is partly defined with a Text parser drawChar; but it is a ByteString parser; it will return the first valid utf8-encoded Char in a ByteString, whatever its byte-length, and the first byte following, if both exist. Because we 'draw' one and 'peek' at the other, the parser as a whole only advances one Char's length along the bytestring, whatever that length may be. See the slightly more complex example 'decode.hs' in the haskellforall blog discussion of this type of byte stream parsing.

These are functions with the simple type:

  decodeUtf8 :: Monad m => Producer ByteString m r -> Producer Text m (Producer ByteString m r)

Thus in general

    decodeUtf8 = view utf8
    decodeUtf16LE = view utf16LE

and so forth, but these forms may be more convenient (and give better type errors!) where lenses are not desired.

These are simply defined

     encodeUtf8 = yield . TE.encodeUtf8

They are intended for use with for

     for Text.stdin encodeUtf8 :: Producer ByteString IO ()

which would have the effect of

     Text.stdin >-> Pipes.Prelude.map (TE.encodeUtf8)

using the encoding functions from Data.Text.Encoding

ascii and latin encodings only use a small number of the characters Text recognizes; thus we cannot use the pipes Lens style to work with them. Rather we simply define functions each way.