[MERGE][bug #175524] Make http test servers 1.1 compliant

[Vincent Ladeuil]

>>>>> "Andrew" == Andrew Bennetts <andrew at canonical.com> writes:

    Andrew> Vincent Ladeuil wrote:
    >> HTTP/1.1 was only partially tested, this patch addresses that
    >> main point and several others found during the work:
    Andrew> [...]
    >> 
    >> On the whole, the tests are now more complete and adding an https
    >> additional parametrization becomes possible.
    >> 
    >> There was 236 tests for 'bzr selftest --no-plugins
    >> tests.test_http', we now have 463 tests, this is due to a
    >> previously incomplete coverage and some tests now raising N/A
    >> (where nothing was done before). This also reveals some pycurl
    >> limitations ;)
    >> 

    Andrew> bb:tweak

    Andrew> This is a very large diff,

Yeah I know ;-/ I tried hard to reduce the patch size and to not
include other smalls enhancements or fixes, but cruft had
accumulated there and this refactoring was really needed.

    Andrew> but none of the code in it seems particularly gnarly,
    Andrew> and it appears to have good test coverage.

That was the main point.

    Andrew> So even though I'm not sure I've comprehensively read
    Andrew> every last detail,

I would have been interested in comments on the test
parametrization and about the idea to push it down to the class
itself ? 

    Andrew> I'd be happy to have this merged in bzr.dev.

Thanks.

    Andrew> It might be good to get Alexander or another Windows
    Andrew> user to check it doesn't cause trouble for them
    Andrew> first, but I don't expect any serious problems there.

Aaron already mentioned (see 'Misc fixes for win32') that it
helped (with a tweak that will be included in my submission).

I also run 'bzr selftest tests.test_http' on winXP and the 302
tests are passing.

<snip/>

    Andrew> [...]
    >> -
    >> -     def test_empty_pass(self):
    >> -         raise tests.KnownFailure(
    >> -             'some versions of pycurl does not handle empty proxy passwords')

    Andrew> Elsewhere you check for the specific versions of
    Andrew> pycurl with this bug.  I think probably this should
    Andrew> be a Feature, so affected tests could check that.  I
    Andrew> also wonder if TestNotApplicable would be more
    Andrew> appropriate than KnownFailure for this situation?

Given that it *was* a bug in earlier versions of pycurl I think
the KnownFailure is appropriate, especially since it will be
triggered only for old versions, remembering people seeing it,
that, yes, there is a known failure in pycurl.

      Vincent