The biggest mistake in SQL was in using NULLs to mean three different things (no record found in an outer join, vs not applicable, vs not yet known). Fortunate

The biggest mistake in SQL was in using NULLs to mean three different things (no record found in an outer join, vs not applicable, vs not yet known). Fortunately with some types (varchar) on sane db's you can use dedicated empty values ('' for varchar) for not applicable, but you are still stuck with the other two as possibly conflicting. This is actually a big issue because your ability to have complex declarative constraints goes up with table width, so breaking off potentially nullable columns also removes them from being available for cross-column check constraints.