[tragomaskhalos]

I assume the author's alternative to null is to punt nullable values into their own table, where the presence of the row tells you that the value is in fact there. Thing is, if your table has three such columns then you're going to need three extra tables. The whole argument is reminiscent of the polemic against surrogate keys, viz that they are a violation of Codd. No-one is arguing that nulls in SQL aren't a mess, but an RDBMS is a tool, not a holy shrine, and most of us prefer to just use bloody things to get work done and ignore the purist bikeshedding.