BACKGROUND
Date & Darwen (1998, 2000), Date (1995), and Pascal (2000) are
unanimous in their conviction that 'null' must be prohibited in relational databases.
By 'null' they are specifically referring to the implementation in all versions of SQL
of a simple representation of a placeholder for missing data, but they have extended
this to a matter of principle: the prohibition of any missing data placeholder.
Their reasoning is that the use of such a device requires a three-valued logic,
alleging that any implementation would necessarily bring complexity and inconsistency.
According to this narrow view of the relational model, the prohibition of null is
equivalent to the use of two-valued logic (i.e. prohibition of an 'unknown' truth-value)
and is a direct result of the use of the 'closed world assumption' (CWA).
This is defined as "the presumption that what is not currently known to be true is false" - or
more commonly the "principle of the excluded middle".
The contrasting open world assumption (OWA)
"assumes that its knowledge of the world is incomplete.
If something cannot be proved to be true, then it doesn't automatically become false.
In the OWA, what is not stated is considered unknown, rather than wrong".
Although many data definition and modelling languages such as RDF and OWL assume the OWA implicitly, these
languages are not relational in any sense, and the present generation of relational
puritans exemplified by Date and Pascal consider the OWA incompatible with the relational database model,
preciesly because its acceptance of an 'unknown' truth-value allows the use of missing-data
placeholders, and "all that that entails".
However, Codd (1979, 1990) explicitly allowed the OWA to be used
in certain relations, and even allowed the mixing of CWA and OWA relations within
the same database.
In databases within the observational sciences, there are often a large number of
gaps - missing observations for which data might or might be supplied at some time
in the future. To insist that the design of the tables within the database be
dependent on the location and distribution of such missing-data 'holes' is clearly
unreasonable, and equally unreasonable would be the insistence that a table
containing such holes cannot be considered a 'relation' qualifying for full
membership of the database as long as some of its tuples remain unfilled.
For this reason, it is important that an OWA version of the relational database
model be developed, free of the straitjacket of 2VL fascism imposed by the CWA as
advocated by the relational puritans.
References
Codd, E.F., 1979. Extending the database relational model to capture more meaning. ACM Transactions on Database Systems 4(4), 397-434.
Codd, E.F., 1990. The relational model for database management: version 2, Addison Wesley, Reading, Mass., 538pp.
Date, C.J., 1995. Relational Database Writings, 1991-1994; Addison-Wesley, Reading, Mass., USA, 542pp.
Date, C.J. and Darwen, H. 1998. Foundation for object/relational databases: the third manifesto, Addison Wesley, Reading, Mass., 496pp.
Date, C.J. and Darwen, H. 2000. Foundation for Future Database Systems: The Third Manifesto, Addison Wesley, Reading, Mass., 608pp.
Pascal, F., 2000. Practical issues in database management: a reference for the thinking practitioner, Addison-Wesley, Boston, 256pp.
