How does GHC/Haskell decide what character encoding it's going to decode/encode from/to?

Question

It seems that GHC is at least inconsistent in the character encoding it decides to decode from.

Consider a file, omatase-shimashita.txt, with the following content, encoded in UTF-8: お待たせしました

readFile seems to read this in properly...

Prelude> content <- readFile "/home/chris/Desktop/omatase-shimashita.txt"
Prelude> length content
8
Prelude> putStrLn content
お待たせしました

However, if I write a simple "echo" server, it does not decode with a default of UTF-8. Consider the following code that handles an incoming client:

handleClient handle = do
  line <- hGetLine handle
  putStrLn $ "Read following line: " ++ toString line
  handleClient handle

And the relevant client code, explicitly sending UTF-8:

Data.ByteString.hPutStrLn handle $ Codec.Binary.UTF8.Generic.fromString "お待たせしました"

Is this not inconsistent behavior? Is there any method to this madness? I am planning to rewrite my application(s) to explicitly use ByteString objects and explicitly encode and decode using Codec.Binary.UTF8, but it would be good to know what's going on here anyway... :o/

UPDATE: I am running on Ubuntu Linux, version 10.10, with a locale of en_US.UTF-8...

$ cat /etc/default/locale 
LANG="en_US.UTF-8"
$ echo $LANG 
en_US.UTF-8

Answer 1

Which version of GHC are you using? Older versions especially didn't do unicode I/O very well.

This section in the GHC documentation describes how to change input/output encodings:

http://haskell.org/ghc/docs/6.12.2/html/libraries/base-4.2.0.1/System-IO.html#23

Also, the documentation says this:

A text-mode Handle has an associated TextEncoding, which is used to decode bytes into Unicode characters when reading, and encode Unicode characters into bytes when writing.

The default TextEncoding is the same as the default encoding on your system, which is also available as localeEncoding. (GHC note: on Windows, we currently do not support double-byte encodings; if the console's code page is unsupported, then localeEncoding will be latin1.)

Encoding and decoding errors are always detected and reported, except during lazy I/O (hGetContents, getContents, and readFile), where a decoding error merely results in termination of the character stream, as with other I/O errors.

Maybe this has something to do with your problem? If GHC has defaulted to something other than utf-8 somewhere, or your handle has been manually set to use a different encoding, that might explain the problem. If you're just trying to echo text at the console, then probably some kind of console code-page funniness is going on. I know I've had similar problems in the past with other languages like Python and printing unicode in a windows console.

Try running hSetEncoding handle utf8 and see if it fixes your problem.

Answer 2

Your first example uses the standard IO library, System.IO. Operations in this library use the default system encoding (also known as localeEncoding) unless you specify otherwise. Presumably your system is set up to use UTF-8, so that is the encoding used by putStrLn, hGetContents and so on.

Your second example uses Data.ByteString. Since this library deals in sequences of bytes only, it does no encoding or decoding. So Data.ByteString.hGetLine converts the bytes in the file directly to a ByteString.

The best way to do text I/O in general is to use the text package.