The last seemingly incomplete part of the spec seems to be the
Dictionary type, which specifies m:n mappings.  While the interface
currently specified by the SRFI would work, it has a couple of
weaknesses:

  * It doesn't distinguish between dicts where values mapped
    by a key must be unique (ie a dict-set) or not (dict-bag).

  * Few people I've spoken to (including myself) know what
    actually makes a good interface for such dictionaries.

The former point argues somewhat to *three* map types, and both
say to me that more work (possibly a lot more) would be required for
an interface that is worthy of being called Schemely.

However, I'm of the opinion that the m:1 Maps currently defined are
pretty non-controversial and elegant, so I'm considering dropping
Dictionaries from this SRFI, but having a fairly strongly worded future
work statement in the SRFI about them, and also archiving the current
API on this list.  This would leave SRFI-44 very usable for 99% of
people (who really just want a hashtable), but not accidentally
standardize an M:N interface which frankly needs more expertise than is
currently assembled.

        Scott

--