On Sunday, March 22, 2015, David Steele <
david@pgmasters.net> wrote:
On 3/20/15 3:37 PM, Tom Lane wrote:
> Fabrízio de Royes Mello <fabriziomello@gmail.com> writes:
>> On Fri, Mar 20, 2015 at 4:19 PM, Peter Geoghegan <pg@heroku.com> wrote:
>>> On Fri, Mar 20, 2015 at 12:15 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>>>> We could fix it by, say, having CheckConstraintFetch() sort the
>>>> constraints by name after loading them.
>
>>> What not by OID, as with indexes? Are you suggesting that this would
>>> become documented behavior?
>
>> I think they should be executed in alphabetical order like triggers.
>
> Yeah. We already have a comparable, and documented, behavior for
> triggers, so if we're going to do anything about this I'd vote for
> sorting by name (or more specifically, by strcmp()).
>
> regards, tom lane
+1 for strcmp() ordering. Not only is this logical and consistent with
the way triggers are fired, names can be manipulated by the user to
force order when desired. Not sure if that's as important for check
constraints as it is for triggers but it might be useful, even if only
for things like unit tests.
--
- David Steele
david@pgmasters.net
It would be nice to know that, at scale, the added comparisons are negligible since it almost all cases all checks are run and pass and the order is irrelevant. Would it be possible for all check constraints to always be run and the resultant error outputs stored in an array which could then be sorted before it is printed? You get better and stable output for errors while minimally impacting good inserts.
David J.