Understandability. To a first approximation,
understandability is inversely proportional to the number
of distinct representation primitives necessary to use the
model. A word commonly used in the data modelling
literature is "naturalness", but this may be so undefinable
as to be useless.
2. Expressive Power. More sophisticated models
incorporate more of the real world's semantics and data
integrity constraints within the model framework. To the
extent this is achieved by the introduction of more
primitives, expressive power may conflict with
understandability.
4. Decomposition of Assumptions. In particular, a desirable
property is the separation of implementation from
specification (data independence).
5. Implementability. An abstraction that cannot be
implemented reasonably efficiently might better be
expressed in some other way.
There was less concern at the conference with user interfaces
and the view of data provided the end user (usually different from
the programmer's interface). The ultimate level of abstraction is
the one closest to the user, and it is at this level that all
abstractions must come together in an integrated way.
The variety of ways in which the participants combined
programming languages and database access was enlightening
for our work integrating Cedar database access into Mesa.
Some chose to equate programming language records with
relational tuples. Others chose to introduce new orthogonal
types and operators for database access, feeling that integration
with the language was not reasonable at this time. Another
dimension of choice is whether to provide aggregate operations
on tuples through (1) algebraic operations such as the relational
algebra integrated with the language, (2) run-time routines that
accept constructed queries (strings or data structures) passed as
arguments, or (3) control constructs or generators to iterate
through tuples in relations, indexes, and so on. The first of these
alternatives is probably the most desirable but also the most
difficult.