Fix bug with registration of identical views with same 'accept'.

[dnouri]

In Kotti I register two views for HTTPForbidden:

config.add_view(
    forbidden_redirect,
    context=HTTPForbidden,
    accept='text/html',
    )

config.add_view(
    forbidden_view,
    context=HTTPForbidden,
    )

With this configuration, and due to the bug, it's not possible to override the first registration (the one with the accept). The following call will work (given autocommit=True or similar) but will fail to override the existing view; the result is that both views are registered in the MultiView but the first is sometimes preferred (both will have the same 'order'):

config.add_view(
    another_forbidden_view,
    context=HTTPForbidden,
    accept='text/html',
    )

Note: the fix looks very similar to the overwriting a couple lines above. However, I've added the check for order == s, since tests (TestMultiView.test_add) expect views with same 'accept' but different 'order' to live alongside each other.

    if phash == h and order == s:
        subset[i] = (order, view, phash)
        return

[mmerickel]

Out of curiosity does this solve your problem?

config.add_view(
    forbidden_redirect,
    context=HTTPForbidden,
    accept='text/html',
    )

config.add_view(
    forbidden_view,
    context=HTTPForbidden,
    accept='*/*'
    )

It seems to me the actual bug here is that for some reason you're able to add the view with the same accept twice without committing between (in test_add_views_with_accept_multiview_replaces_existing).

[dnouri]

Using accept='*/*' as you suggest produces a PredicateMismatch when I then try the view lookup.

Note that adding a view with the same accept twice is not possible when autocommit=False; it will correctly generate a configuration conflict. (That's contrary to your last statement, if I understood you correctly.)

[mmerickel]

Yeah misread the unit test, didn't notice the autocommit=True.

[dnouri]

I've updated the changeset to:

• remove the changes I originally made to TestMultiView.test_add, and only change that existing test so much to make it work
• add another test (test_add_with_phash_override_accept) that shows the bug more clearly than the previous one did

[mcdonc]

Thanks Daniel!