> I thought I'd explained it in detail [...]

Well, so far, you didn't say that it was testUnicodeOpen that
failed. Now that I see that, I can understand the problem.

In your original message, I failed to see a problem. Yes, the repr
is inconsistent - but why is that a problem? (and then, why does
it actually cause real problems in the test suite on Vista - as
it turns out, the inconsistency causes no problems in the testsuite
at all)

> My patch proposes to 
> a) make sure that the repr(f) is a string containing the repr of the 
> filename (be it unicode, string or whatever)

I cannot understand how this contributes to solving the problem. -1.

> b) Fix the testsuite so that both this case and the one testing string 
> filenames assumes a repr of the filename.

As it is clearly the test suite that is broken, we should just fix the
test suite. If the two chunks in the test, only the second chunk is
relevant for fixing the problem.