Three valued logic (again)

In the interests of stirring up a good discussion here (why should I confine it to the site where the situation originated?): We have a SELECT COUNT(*) which includes the following term in a complex WHERE clause: product_code NOT IN (SELECT combination_code FROM components) (product_code and combination_code are both CHAR(7)s BTW) Running on 4.11 OL this gave a result of about 1700 and on 7.21 it gave about 300. Further investigation revealed that there were a couple of rows with null combination_code values in components. There are no NULLs amongst the set of product_code. Adding WHERE combination_code IS NOT NULL to the SELECT FROM components resulted in both versions giving the same result as 4.11 gave with the original version. The explanation seems to be that when product_code has a value which is in the non-null subset of combination_codes either both versions treat NOT IN as FALSE and the term fails. When product_code has a value which is not in the non-null subset the two versions treat it differently. 4.11 seems to be only comparing product_code against the non-null set so that NOT IN is FALSE and the term succeeds. 7.21 is comparing it against the full set of combination_code returned by the SELECT which includes NULL so that NOT IN is UNKNOWN and the term fails. I have my view of which is correct and so does the author of the code but these views do not coincide; fortunately we do not have a copy of the ANSI SQL spec (why ruin a good argument). Your views are invited..... Ian