[IronPython] End of the World? No just end of the line.

[Sylvain Hellegouarch]

Hi folks,

Working with M. David Peterson trying to run the CherryPy HTTP server on
IP under Windows we ran into a fairly subtle bug that was actually not
so subtle at all.

When the server accepts a connection it maps the socket into a file
object so that it can read and write with a nice API.

When using IPCE socket module this achieves the same by doing this:

def makefile(self, mode='r', bufsize=-1):
    stream = NetworkStream(self.socket)
    return file(stream, mode)

This works great when you don't care about the way the end of the lines
is defined but if you do be ready for some hair-tearing hours.

Indeed with IronPython running on Windows with the .NET framework the
end of the line is defined as <LF> whereas for instance on Linux with
Mono it's <CR><LF>.

If I'm not mistaken CPython always ensures the end of line is <CR><LF>
so that people don't have to worry about the underlying OS behavior.

The problem I faced is that HTTP marks the end of the header lists via a
<CR><LF> but because CherryPy reads the socket as a file it only gets a
<LF> with IP under Windows and therefore fails miserably. I've added a
test for a single <LF> in my copy of CherryPy and it fixed the problem.
But this is a bit of a hassle.

I don't think there is a bug of any of the part here. However I believe
there should be either a very clear warning that this difference will
always exists and Python developer will have to check for that case, or
maybe IP should find a way to ensure that <CR><LF> is always returned
even under Windows (which I doubt is an easy thing to do).

In any case if your code depends on the line ending be aware of that
difference as you could end up swearing too loud for your own good.

- Sylvain