Comments on: Constraints

By: Sokrates

Sokrates — Mon, 19 Sep 2011 18:31:45 +0000

Kamal,

yes, I did

By: Kamal

Kamal — Mon, 19 Sep 2011 18:07:45 +0000

Did Sokrates mean this?

http://en.wikipedia.org/wiki/Tautology_%28logic%29#Efficient_verification_and_the_Boolean_satisfiability_problem

Kind regards

–
Kamal

By: NULL « Oracle Scratchpad

NULL « Oracle Scratchpad — Mon, 19 Sep 2011 16:58:47 +0000

[...] As a follow on from this: a predicate applied to a dataset will return rows only when it evaluates to TRUE; a constraint will allow rows to be inserted into a table if if doesn’t evaluate to FALSE. [...]

By: Log Buffer #214, A Carnival of the Vanities for DBAs | The Pythian Blog

Log Buffer #214, A Carnival of the Vanities for DBAs | The Pythian Blog — Mon, 28 Feb 2011 14:04:19 +0000

[...] Jonathan Lewis reminds about a subtle difference between a predicate and the constraint. [...]

By: Jonathan Lewis

Jonathan Lewis — Thu, 24 Feb 2011 08:01:34 +0000

Interesting point – but arguably this does more to enforce the idea that the NOT NULL constraint is a special case that is inconsistent in more ways than one with all other uses of constraints.

So the purist may say it shouldn’t happen, while the pragmatist might suggest that more cases should take advantage of the optimisation. (Why am I generating undo and redo for a row when I have a constraint that tells me it’s going to be rejected ?)

By: joel garry

joel garry — Wed, 23 Feb 2011 22:30:39 +0000

Is it really a purist defect? The difference is in the Check Constraint link cited above; the check constraint is done after the statement runs, as a rollback. Perhaps it isn’t logically inconsistent to disable the integrity constraint, see if a transaction works, then roll it back with the check constraint. Whether Oracle can actually do that I haven’t considered.

By: Sokrates

Sokrates — Wed, 23 Feb 2011 12:15:19 +0000

The fact that some reductions to “is not null” may not be feasible doesn’t mean that an explicit “is not null” should be treated differently from a NOT NULL check, though.

ok, agree – like many others
see for example
http://asktom.oracle.com/pls/apex/f?p=100:11:0::::P11_QUESTION_ID:2665514800346181577#2666924900346321646

(has someone filed an enhancement request ?)

By: Jonathan Lewis

Jonathan Lewis — Wed, 23 Feb 2011 10:54:55 +0000

Okay, let me modify that:

“…if the optimizer were good enough it would be able to infer the “column is not null” condition from any logical equivalent that hadn’t been written by a mathematician with a warped sense of humour, and ought to optimise accordingly”
;)

I think your conjecture seems reasonable – an intuitive view (not that I like to trust my intuition) would be that the constraint problem is the tautology problem extended to three-valued logic, and the tautology problem couldn’t be NP-hard if the constraint problem weren’t.

The fact that some reductions to “is not null” may not be feasible doesn’t mean that an explicit “is not null” should be treated differently from a NOT NULL check, though.

By: Houri Mohamed

Houri Mohamed — Tue, 22 Feb 2011 14:38:45 +0000

The same thing can be said about the Foreign key integrity constraint

http://jonathanlewis.wordpress.com/2009/05/03/foreign-keys/#more-1101

Best Regards

Mohamed Houri

By: Sokrates

Sokrates — Tue, 22 Feb 2011 13:27:09 +0000

…if the optimizer were good enough it would be able to infer the “column is not null” condition from any logical equivalent, and ought to optimise accordingly…

I would conjecture – though I cannot prove it – that this ( “given an SQL-Expression expr(column), is it logically equivalent to ‘column is not null’ ?”) is as hard as the tautology problem (“given a Boolean formula, is it equivalent to ’0=0′ ?”), which is known to be NP-hard