Comment 2 for bug 794620

I don't think overrideAttr is bad for any particular test. I think it
has some bad emergent behaviour because it does make it easy to do
monkey patching: we don't add a cleaner interface than monkey patching
to set up or observe the thing we care about; also it tends to be done
inline in particular tests. In some cases (sorry, I forget which) the
first of those has seemed to make the tests overly tightly coupled to
the implementation.

I think generally it is at least worth thinking about whether the
overrideAttr ought to be extracted out; and if it is extracted whether
it should go somewhere more general than the particular test file.