"Timothy H. Keitt" wrote:
>
> If I were implementing this in C++, I would have the result object
> return a different generic STL iterator (forward, random access, etc.)
> depending on how I wanted to access the data. Perhaps you could emulate
> this in C. I generally don't like the one-interface-fits-all approach;
> you get a much cleaner and extensible interface if you introduce a type
> for each class of behavior being modeled.
If we want to relagate the current API to the status of "legacy", and
build something all-new and well thought out, then this could be done.
I'd certainly be willing to do this, but what is the consensus? If I
came up with something completely different but better would the rest of
the team be happy to make the current interface legacy? Or do we want a
compromise (like what Peter Eisentraut suggests perhaps), or do we want
something that slots into the current world view with minimum
disruption? (what I have suggested).