Comment 3 for bug 297563

On Tue, 2008-11-18 at 09:14 +0000, Glyph Lefkowitz wrote:
> My personal impression (take this with a grain of salt, I don't know too
> much about testresources yet) is that teaching the decorator to pass
> through unknown attribute lookups will fix this particular issue, but
> make the ultimate failure condition much more painful; especially given
> the proclivity of test-infrastructure code to make liberal use of
> potentially conflicting names like "name", "id", "count", etc.

No more than subclassing or decorating elsewhere in python; while there
is no super-robust answer in python core, I don't see that we need to
solve it any more in other code, except where problems are actually
occuring.

> The idea of somehow making the decorator less intrusive is a good one,
> though. If passing through attribute lookups could be combined with
> some mechanism for namespacing of additional attributes that might
> eliminate my concern...?

I think I need a restating of your concern, perhaps as a mini story, to
understand it. My attempt follows:

When code from source A looks for 'test.FOO' and a test from source B
has 'FOO', allowing A to find test.FOO will break if B's FOO is
different, as as such the twisted test decorator should prevent A from
accessing test.FOO.

I think this is bogus, A and B will either work well together, or will
not. twisted's implementation detail (that it decorates test cases)
neither makes the cooperation between A and B better nor worse (if it
passes attribute lookups to the decorated object). Of course, it can
make it worse if it prohibits such lookups.

-Rob
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.