Highlights:

* Unicode support for Handle I/O:

  ** Automatic encoding and decoding using a per-Handle encoding.

  ** The encoding defaults to the locale encoding (only on Unix 
     so far, perhaps Windows later).

  ** Built-in UTF-8, UTF-16 (BE/LE), and UTF-32 (BE/LE) codecs.

  ** iconv-based codec for other encodings on Unix

* Modularity: the low-level IO interface is exposed as a type class
  (GHC.IO.IODevice) so you can build your own low-level IO providers and
  make Handles from them.

* Newline translation: instead of being Windows-specific wired-in
  magic, the translation from \r\n -> \n and back again is available
  on all platforms and is configurable for reading/writing
  independently.


Unicode-aware Handles
~~~~~~~~~~~~~~~~~~~~~

This is a significant restructuring of the Handle implementation with
the primary goal of supporting Unicode character encodings.

The only change to the existing behaviour is that by default, text IO
is done in the prevailing locale encoding of the system (except on
Windows [1]).  

Handles created by openBinaryFile use the Latin-1 encoding, as do
Handles placed in binary mode using hSetBinaryMode.

We provide a way to change the encoding for an existing Handle:

   GHC.IO.Handle.hSetEncoding :: Handle -> TextEncoding -> IO ()

and various encodings (from GHC.IO.Encoding):

   latin1,
   utf8,
   utf16, utf16le, utf16be,
   utf32, utf32le, utf32be,
   localeEncoding,

and a way to lookup other encodings:

   GHC.IO.Encoding.mkTextEncoding :: String -> IO TextEncoding

(it's system-dependent whether the requested encoding will be
available).

We may want to export these from somewhere more permanent; that's a
topic for a future library proposal.

Thanks to suggestions from Duncan Coutts, it's possible to call
hSetEncoding even on buffered read Handles, and the right thing
happens.  So we can read from text streams that include multiple
encodings, such as an HTTP response or email message, without having
to turn buffering off (though there is a penalty for switching
encodings on a buffered Handle, as the IO system has to do some
re-decoding to figure out where it should start reading from again).

If there is a decoding error, it is reported when an attempt is made
to read the offending character from the Handle, as you would expect.

Performance varies.  For "hGetContents >>= putStr" I found the new
library was faster on my x86_64 machine, but slower on an x86.  On the
whole I'd expect things to be a bit slower due to the extra
decoding/encoding, but probabaly not noticeably.  If performance is
critical for your app, then you should be using bytestring and text
anyway.

[1] Note: locale encoding is not currently implemented on Windows due
to the built-in Win32 APIs for encoding/decoding not being sufficient
for our purposes.  Ask me for details.  Offers of help gratefully
accepted.


Newline Translation
~~~~~~~~~~~~~~~~~~~

In the old IO library, text-mode Handles on Windows had automatic
translation from \r\n -> \n on input, and the opposite on output.  It
was implemented using the underlying CRT functions, which meant that
there were certain odd restrictions, such as read/write text handles
needing to be unbuffered, and seeking not working at all on text
Handles.

In the rewrite, newline translation is now implemented in the upper
layers, as it needs to be since we have to perform Unicode decoding
before newline translation.  This means that it is now available on
all platforms, which can be quite handy for writing portable code.

For now, I have left the behaviour as it was, namely \r\n -> \n on
Windows, and no translation on Unix.  However, another reasonable
default (similar to what Python does) would be to do \r\n -> \n on
input, and convert to the platform-native representation (either \r\n
or \n) on output.  This is called universalNewlineMode (below).

The API is as follows.  (available from GHC.IO.Handle for now, again
this is something we will probably want to try to get into System.IO
at some point):

-- | The representation of a newline in the external file or stream.
data Newline = LF    -- ^ "\n"
             | CRLF  -- ^ "\r\n"
             deriving Eq

-- | Specifies the translation, if any, of newline characters between
-- internal Strings and the external file or stream.  Haskell Strings
-- are assumed to represent newlines with the '\n' character; the
-- newline mode specifies how to translate '\n' on output, and what to
-- translate into '\n' on input.
data NewlineMode 
  = NewlineMode { inputNL :: Newline,
                    -- ^ the representation of newlines on input
                  outputNL :: Newline
                    -- ^ the representation of newlines on output
                 }
             deriving Eq

-- | The native newline representation for the current platform
nativeNewline :: Newline

-- | Map "\r\n" into "\n" on input, and "\n" to the native newline
-- represetnation on output.  This mode can be used on any platform, and
-- works with text files using any newline convention.  The downside is
-- that @readFile a >>= writeFile b@ might yield a different file.
universalNewlineMode :: NewlineMode
universalNewlineMode  = NewlineMode { inputNL  = CRLF, 
                                      outputNL = nativeNewline }

-- | Use the native newline representation on both input and output
nativeNewlineMode    :: NewlineMode
nativeNewlineMode     = NewlineMode { inputNL  = nativeNewline, 
                                      outputNL = nativeNewline }

-- | Do no newline translation at all.
noNewlineTranslation :: NewlineMode
noNewlineTranslation  = NewlineMode { inputNL  = LF, outputNL = LF }


-- | Change the newline translation mode on the Handle.
hSetNewlineMode :: Handle -> NewlineMode -> IO ()



IO Devices
~~~~~~~~~~

The major change here is that the implementation of the Handle
operations is separated from the underlying IO device, using type
classes.  File descriptors are just one IO provider; I have also
implemented memory-mapped files (good for random-access read/write)
and a Handle that pipes output to a Chan (useful for testing code that
writes to a Handle).  New kinds of Handle can be implemented outside
the base package, for instance someone could write bytestringToHandle.
A Handle is made using mkFileHandle:

-- | makes a new 'Handle'
mkFileHandle :: (IODevice dev, BufferedIO dev, Typeable dev)
              => dev -- ^ the underlying IO device, which must support
                     -- 'IODevice', 'BufferedIO' and 'Typeable'
              -> FilePath
                     -- ^ a string describing the 'Handle', e.g. the file
                     -- path for a file.  Used in error messages.
              -> IOMode
                     -- ^ The mode in which the 'Handle' is to be used
              -> Maybe TextEncoding
                     -- ^ text encoding to use, if any
              -> NewlineMode
                     -- ^ newline translation mode
              -> IO Handle

This also means that someone can write a completely new IO
implementation on Windows based on native Win32 HANDLEs, and
distribute it as a separate package (I really hope somebody does
this!).

This restructuring isn't as radical as previous designs.  I haven't
made any attempt to make a separate binary I/O layer, for example
(although hGetBuf/hPutBuf do bypass the text encoding and newline
translation).  The main goal here was to get Unicode support in, and
to allow others to experiment with making new kinds of Handle.  We
could split up the layers further later.


API changes and Module structure
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

NB. GHC.IOBase and GHC.Handle are now DEPRECATED (they are still
present, but are just re-exporting things from other modules now).
For 6.12 we'll want to bump base to version 5 and add a base4-compat.
For now I'm using #if __GLASGOW_HASKEL__ >= 611 to avoid deprecated
warnings.

I split modules into smaller parts in many places.  For example, we
now have GHC.IORef, GHC.MVar and GHC.IOArray containing the
implementations of IORef, MVar and IOArray respectively.  This was
necessary for untangling dependencies, but it also makes things easier
to follow.

The new module structurue for the IO-relatied parts of the base
package is:

GHC.IO
   Implementation of the IO monad; unsafe*; throw/catch

GHC.IO.IOMode
   The IOMode type

GHC.IO.Buffer
   Buffers and operations on them

GHC.IO.Device
   The IODevice and RawIO classes.

GHC.IO.BufferedIO
   The BufferedIO class.

GHC.IO.FD
   The FD type, with instances of IODevice, RawIO and BufferedIO.

GHC.IO.Exception
   IO-related Exceptions

GHC.IO.Encoding
   The TextEncoding type; built-in TextEncodings; mkTextEncoding

GHC.IO.Encoding.Types
GHC.IO.Encoding.Iconv
GHC.IO.Encoding.Latin1
GHC.IO.Encoding.UTF8
GHC.IO.Encoding.UTF16
GHC.IO.Encoding.UTF32
   Implementation internals for GHC.IO.Encoding

GHC.IO.Handle
   The main API for GHC's Handle implementation, provides all the Handle
   operations + mkFileHandle + hSetEncoding.

GHC.IO.Handle.Types
GHC.IO.Handle.Internals
GHC.IO.Handle.Text
   Implementation of Handles and operations.

GHC.IO.Handle.FD
   Parts of the Handle API implemented by file-descriptors: openFile,
   stdin, stdout, stderr, fdToHandle etc.