Cause of Indeterminacy in SQL

One root cause of indeterminacy in SQL lies in its implementation of comparison for equality. For certain system-defined types it is possible for distinct values to compare equal (note the contradiction). One such type is CHARACTER. Like COBOL, SQL ignores trailing "pad characters" when comparing character strings. The pad character is normally the space obtained by depressing the space bar on a keyboard. Thus, for example, the comparison 'SQL' = 'SQL ' evaluates to TRUE, even though CHAR_LENGTH('SQL') = CHAR_LENGTH('SQL '), comparing the lengths of those two strings, 3 and 6, evaluates to FALSE. Now consider the relational projection of ENROLMENT over just its Name attribute: ENROLMENT{Name} has a counterpart of projection, but suppose the two rows for student S1 in the ENROLMENT table had 'Anne' and 'Anne ' for S1's name.

If both of those values were to appear in the result, that would be inconsistent with the fact that they compare equal in SQL. If just one of them appears, then which one? The SQL standard declares such an expression to be possibly non-deterministic and permits a conforming implementation to give any value that compares equal to 'Anne'-possibly one that doesn't even appear in the table-and does not require it to give the same value every time the expression is evaluated. As a consequence, there are several SQL operators whose use on character strings is not permitted to appear in constraint declarations. The SQL standard lists a multitude of conditions that cause an expression to be defined as possibly non-deterministic. Perhaps the most alarming is the assumption that equals comparison of values of user-defined types is assumed to suffer from the same problem as I have described for character strings: it is assumed that distinct values can compare equal, even if the type definition is such that this cannot possibly be the case.