Hi All,
Judging from a bit of searching, some databases seem to do
short circuit evaluation of predicates (which has been
discussed here before)
SQL92 doesn't say either way whether this is valid - my
reading of section 8.12 of the standard implies that it's
not (I would have expected a X in the truth table for true
OR), but I guess you can read it either way. (it certainly
allows for re-ordering of predicates, which I'm not arguing).
But if re-ordering + short circuiting is valid, then how
are run-time errors coped with?
select * From a where 1=0 and a1 < (select b1 from b)
(b has 2 rows -> scalar subquery cardinality error).
Section 7.11 of SQL92 mandates the cardinality error.
SQL server seems to ignore the run-time error, and return
0 rows, but I can't see this as being correct! Surely
an optimisation, in order to be correct, should not allow
the result (or lack of) of a problem set to be changed?!
Cheers,
Lee Benfield
Received on Wed Feb 19 2003 - 12:45:31 CET
