On 2004-10-06, Glynn Clements wrote:

I can't comment on nhc98, but the Haskell98 standard doesn't include any mechanism for binary I/O.

Ouch. That seems like a major oversight to me. Will there be any effort to fix that in the future? -- John

Malcolm Wallace wrote:

...

...

I can't comment on nhc98, but the Haskell98 standard doesn't include any mechanism for binary I/O.

Ouch. That seems like a major oversight to me. Will there be any effort to fix that in the future?

Note that, on Unix-like systems, there is no difference between text I/O and binary I/O on files. It is only Windows that requires a separation of the modes.

There are two issues here. The first is EOL conversion; as Malcom notes, this isn't an issue on Unix, but it is an issue on Windows. On Windows, there is no standard way to obtain the contents of a file such that \n and \r\n are distinct. The second is character encoding/decoding. The Haskell98 I/O functions all deal with Chars. When reading a file, the byte stream is converted to a list of characters using an *unspecified* encoding. AFAIK, all implementations are currently hardcoded to assume ISO-8859-1, so you can reliably obtain the original list of bytes using the ord function. However, nothing in the standard dictates that ISO-8859-1 is used, and there has been talk of using the locale's encoding instead. If that were to happen, it would be practically (as well as theoretically) impossible to perform binary I/O using the Haskell98 API, even on Unix. This issue has been beaten to death fairly recently, so I'm not going to repeat it here. See the thread entitled "Writing binary files" from Sep 11-18 for the details. -- Glynn Clements