SQL and Relational Theory, 2nd Edition

Appendix B. SQL Departures from the Relational Model

In this appendix I summarize, mainly for purposes of reference and with little by way of additional commentary, some of the ways in which SQL—by which I mean, as always in this book, the standard version of that language except where otherwise noted—departs from the relational model. Now, I know there are those who will quibble over specific items in this list; it’s not easy to compile such a list, especially if it’s meant to be orthogonal (i.e., if an attempt is made to keep the various items all independent of one another). But I don’t think such quibbling is important. What’s important is the cumulative effect, which quite frankly I think is overwhelming.^[189]

• SQL fails to distinguish adequately between table values and table variables.

• SQL tables aren’t the same as relations (or relvars), because they either permit or require, as the case may be, (a) duplicate rows; (b) nulls; (c) left to right column ordering; (d) anonymous columns; (e) duplicate column names; (f) pointers; and—at least in some products, though not in the standard as such—(g) hidden columns. Note that most if not all of these differences constitute violations of The Assignment Principle.

• SQL has no proper table literals.

• SQL often seems to think views aren’t tables.

• SQL tables (views included!) must have at least one column.

• SQL has no explicit table assignment operator.

• SQL has no explicit multiple table assignment a fortiori (nor does it have an INSERT/DELETE ...