This is quite an hack, but at least it will start the ball rolling on this issue.

There are several problems, all related to CRLF. The first problem I had, and that the simpler patch I posted on issue 78 solved, was creating a diff on Linux to a file with CRLF endings.

The next problem, that this patch solves, is posting a patch created on windows (with svn diff or tortoisesvn). There were two problems:
* the parsing of the revision number, caused by the first part of the patch.
* after correcting revision parsing, another problem is caused by patch itself. If patch (in Unix, at least) gets a patch file with CRLF line endings, it assumes it was meant to have LF line endings and throws the CR away. This makes the patch fail because the file being patched has CRLF. 

This second problem is solved by converting the header lines to end with LF.
By the way, I think this is the cause for the issue mentioned by someone on the mailing list using tortoise svn to generate the diff.

The code is a hack, I just started looking at python code this week, and I don't really know what happens with review-board installed on Windows, but it solves my problem.

It works for patches to files containing CRLF, generated both on Windows and Linux, at least when the server is running on Linux