Re: Proposal: revert behavior of IS NULL on row types

Peter Eisentraut <peter.eisentraut@2ndquadrant.com> writes:
> On 7/26/16 7:46 PM, Thomas Munro wrote:
>> By the way, our documentation says that NOT NULL constraints are
>> equivalent to CHECK (column_name IS NOT NULL).  That is what the SQL
>> standard says, but in fact our NOT NULL constraints are equivalent to
>> CHECK (column_name IS DISTINCT FROM NULL).  Should we update the
>> documentation with something like the attached?

> Couldn't we just fix that instead?  For NOT NULL constraints on
> composite type columns, create a full CHECK (column_name IS NOT NULL)
> constraint instead, foregoing the attnotnull optimization.

Maybe.  There's a patch floating around that expands attnotnull into
CHECK constraints, which'd provide the infrastructure needed to consider
changing this behavior.  But that's not going to be back-patchable, and
as I noted in <10682.1469566308@sss.pgh.pa.us>, we have a problem right
now that the planner's constraint exclusion logic gets these semantics
wrong.
        regards, tom lane