Обсуждение: Mutable CHECK constraints?

Поиск
Список
Период
Сортировка

Mutable CHECK constraints?

От
Laurenz Albe
Дата:
We throw an error if the expression in a CREATE INDEX statement is not IMMUTABLE.
But while the documentation notes that expressions in CHECK constraints are not
to be immutable, we don't enforce that.  Why don't we call something like
CheckMutability inside cookConstraint?  Sure, that wouldn't catch all abuse,
but it would be better than nothing.

There is of course the worry of breaking upgrade for unsafe constraints, but is
there any other reason not to enforce immutability?

Yours,
Laurenz Albe



Re: Mutable CHECK constraints?

От
Tom Lane
Дата:
Laurenz Albe <laurenz.albe@cybertec.at> writes:
> We throw an error if the expression in a CREATE INDEX statement is not IMMUTABLE.
> But while the documentation notes that expressions in CHECK constraints are not
> to be immutable, we don't enforce that.  Why don't we call something like
> CheckMutability inside cookConstraint?  Sure, that wouldn't catch all abuse,
> but it would be better than nothing.

> There is of course the worry of breaking upgrade for unsafe constraints, but is
> there any other reason not to enforce immutability?

Yeah, that's exactly it, it's a historical exemption for compatibility
reasons.  There are discussions about this in the archives, if memory
serves ... but I'm too tired to go digging.

            regards, tom lane



Re: Mutable CHECK constraints?

От
Laurenz Albe
Дата:
On Tue, 2023-01-24 at 01:38 -0500, Tom Lane wrote:
> Laurenz Albe <laurenz.albe@cybertec.at> writes:
> > We throw an error if the expression in a CREATE INDEX statement is not IMMUTABLE.
> > But while the documentation notes that expressions in CHECK constraints are not
> > to be immutable, we don't enforce that.  Why don't we call something like
> > CheckMutability inside cookConstraint?  Sure, that wouldn't catch all abuse,
> > but it would be better than nothing.
>
> > There is of course the worry of breaking upgrade for unsafe constraints, but is
> > there any other reason not to enforce immutability?
>
> Yeah, that's exactly it, it's a historical exemption for compatibility
> reasons.  There are discussions about this in the archives, if memory
> serves ... but I'm too tired to go digging.

Thanks for the answer.  A search turned up
https://postgr.es/m/AANLkTikwFfvavEX9nDwcRD4_xJb_VAitMeP1IH4wpGIt%40mail.gmail.com

Yours,
Laurenz Albe