Use of Table Expressions - Expressing Constraint Conditions
With the exception of key constraints, the examples in the theory book all explicitly reference at least one relvar and thus involve invocations of relational operators or aggregate operators. Assuming support for CREATE ASSERTION, we can always derive SQL counterparts of these examples using table expressions and truth-valued operators, but when that assumption does not hold we need to look for alternative solutions using table constraints. In most cases these will entail the use of subqueries and even that technique is prohibited by many implementations. In some cases special syntactic constructs are available, as we shall see, but there are several for which no SQL solution is available unless the implementation supports CREATE ASSERTION or subqueries in table constraints.Now, the reason usually given for lack of support for subqueries in constraints is that in general such expressions can require the DBMS to examine the entire content of possibly very large tables.
If database updates are expected to occur frequently-and are perhaps required to occur very frequently indeed- then declaration of such constraints would give rise to an intolerable slowing down of the updating process. Of course this is an extremely valid concern and we have to admit that integrity might occasionally have to be compromised for performance reasons, but consider the user with a small database that is subject to comparatively infrequent updating but nevertheless has strong integrity requirements. Might not such a user feel unfairly treated by a system that prohibits the declaration of required constraints? Defenders of the status quo respond to this argument by holding that language constructs that can give rise to disappointment for performance reasons, to such an extent as to militate against their use in common practical situations, should be banned. But sometimes users resort to implementing constraints, as best they can, in application code when they wish to enforce a constraint that is not supported by the DBMS but nevertheless does not adversely impair performance. The DBMS could almost certainly enforce such constraints much more efficiently and much more reliably. We can also point to various other SQL constructs that might be subject to similar concerns but are supported nonetheless. For example, if tables T1, T2, and T3 each contain 100,000 rows, then SELECT * FROM T1, T2, T3, when evaluated, delivers a table containing a quadrillion rows.