In the case where a file in SVN had both unicode characters in it and

svn:keywords, trying to fetch the files would cause UnicodeDecodeErrors when

processing the keywords. I've fixed things up to do all the regex work in bytes

mode.

Ran unit tests. Before the changes to svn/base.py, the new test failed with the

same error as in the bug reports we've received about this. After, the test

runs fine.