[Zope3-dev] Interface field type

[Shane Hathaway]

On Sun, November 14, 2004 12:34 pm, Jim Fulton said:
> Shane Hathaway wrote:
>> Today I wrote a containerViews directive and accidentally pointed the
>> "for" attribute at a class rather than an interface.
>
> It should be legal to provide a class rather than an interface.
> This should be reported as a bug.

Oh.  The docstring describing the "for" attribute was pretty clear: "The
containerViews will be for all objects that implement this interface." 
It's a complete surprise to hear that this field should also accept a
class.

Since I thought the docstring and many similar docstrings were telling the
truth, this weekend I changed directives that want an interface to use the
new GlobalInterface type.  I wanted to be sure I (and other users new to
Zope 3) didn't repeat the mistake of using a class where everyone else is
using interfaces.

GlobalInterface is a simple subclass of GlobalObject.  It serves two
purposes: it's a shortcut for GlobalObject(value_type=InterfaceField()),
and it shows up in the API documentation, further clarifying which fields
require interfaces.

> Note again, however that a class should be allowed in this
> case.

Well, now I feel bad for adding excessive constraints.  I was certain I
understood your intent, since the documentation was clear.

On the other hand, I don't feel good about reverting my changes, because
the bug I found was quite painful to figure out.  I suggest we need to
relax the GlobalInterface constraints one by one, adding tests that verify
the code really works with classes in addition to interfaces.

Shane