From: Xavier Noria <fxn@...>
Date: 2003-09-11T06:25:55+09:00
Subject: Re: windows line termination

On Wednesday 10 September 2003 21:49, Chris Morris wrote:

> Still don't know if that quells my rant. I assume that the default \n
> is used in memory at all times, and Ruby will handle the appropriate
> translation for *nix/Windows/Mac, freeing me from worrying over that?
> And the case I stumbled upon today was a leaky abstraction?

When you read/write a text file Ruby (as C or Perl) lets you see a 
newline as "\n", a single character. Unless you do I/O in binmode that 
happens transparently behind the scenes, and in some platforms as Unix 
nothing needs to be done indeed. 

That way one can easily write text-oriented portable code, print "\n" 
will do the right thing anywhere.

The problem you faced has to do with the HTTP specs. HTTP headers, for 
instance, have to be separated by CRLF, that is \015\012 because that's 
required. So if you wrote an HTTP server written in Ruby with code like

    socket.print("Location: ${location}\n") # broken

that server wouldn't be portable. In such cases you have to manually 
ensure that the CRLF travels well, untouched. And that usually means 
you put the handle in binmode (which turns off that transalation) and 
print exactly what's needed.

The newlines for the body of an HTTP message with text (i.e., a normal 
HTML page) need not be CRLF, but in the kind of data you were uploading 
that's required (in my understanding of the RFC).

-- fxn