The WSGI reference implementation does not provide any means for application code to distinguish between the following request lines:

GET /foo/bar HTTP/1.1

GET /foo%2Fbar HTTP/1.1

Now, the relevant RFC-1945 (https://tools.ietf.org/html/rfc1945#section-3.2) does not explicitly state how these should be handled by application code, but it does clearly distinguish encoded from unencoded forward-slashes in the BNF, which suggests that percent-encoded slashes should be considered part of a path segment, while unencoded slashes should be considere segment separators, and thus that the first URL is supposed to be interpreted as ['foo', 'bar'], but the second one as ['foo/bar']. However, the 'PATH_INFO' WSGI environ variable contains the same string, '/foo/bar', in both cases, making it impossible for application code to handle the difference. I believe the underlying issue is that percent-decoding (and decoding URLs into UTF-8) happens before interpreting the 'PATH_INFO', which is unavoidable because of the design decision to present PATH_INFO as a unicode string - if it were kept as a bytestring, then interpreting it would remain the sole responsibility of the application code; if it were a fully parsed list of unicode path segments, then the splitting could be implemented correctly.

Unfortunately, I cannot see a pleasant way of fixing this without breaking a whole lot of stuff, but maybe someone else does.

It's also very possible that I interpret the RFC incorrectly, in which case please enlighten me.