Hi

It would be nice if it was possible to assign sets to multilink properties as well as lists.  e.g.:

>>> ppl = set(db.issue.get(nodeid, 'nosy'))
>>> ppl.add(db.user.lookup('someone'))
>>> db.issue.set(nodeid, nosy=ppl)
instead of:
>>> db.issue.set(nodeid, nosy=list(ppl))

It may even make sense to return sets with the get() call?

Cheers
Floris

Right now the backends check during set_inner whether type(incoming_value) == type([]) for Multilinks. Really any finite interator would work; it looks like all roundup does is "for x in incoming_value". We could replace the type check with "hasattr(x, '__iter__') and let any iterator through. Then we wouldn't have to enumerate all the things that are allowed (set, frozenset, list, tuple, etc).

I don't know the hyperdb well enough yet to know if there is a good reason to limit it to just lists. Maybe Richard can comment. If he's okay with it, I'll change backend_anydbm.py, backend_metakit.py, rdbms_common.py.

As for returning sets, did you have any specific use cases where returning a set would be a big win over getting a list back?

I don't have a problem with set() taking any iterable.

No, get() should not return sets - Multilinks are ordered sequences.

Fix checked in CVS; multilinks can take any iterable. A unit test was added as well.

I agree that they shouldn't return sets with get(), not sure why I added that comment...

Making set() accept any iterable is cool, thanks!