On 09/02/2013 02:26 PM, Andres Freund wrote:
> On 2013-09-02 14:20:57 -0400, Andrew Dunstan wrote:
>> On 09/02/2013 01:30 PM, Tom Lane wrote:
>>> Andrew Dunstan <andrew@dunslane.net> writes:
>>>> Yes, possibly, but we can't do that now, but I would like to fix the
>>>> docs now.
>>> If you want this in 9.3.0 it needs to be committed in the next couple of
>>> hours.
>>>
>>> FWIW, the idea seemed generally sane to me, but I'd suggest not depending
>>> on reltoastrelid being zero when and only when there's no match.
>>> Why not test whether t.oid IS NULL, instead?
>>>
>>> Or actually, code it like this
>>>
>>> GREATEST(age(c.relfrozenxid), age(t.relfrozenxid))
>>>
>>> and be done, as well as not having an ugly direct use of int4larger.
>>>
>>>
>>
>> OK, I'll do it that way. Working on it now.
> I'd vote for c.relkind != 't' AND NOT c.relfrozenxid = 0; instead of
> relkind = 'r' for the main relation, that way you'd include materialized
> views and stuff.
>
See what was just committed - the matview case is included for 9.3+ (as
it was in fact in the original - I must have been looking at older docs
when saw it wasn't there.)
I'll be back in an hour or so if any final tweeks are needed.
cheers
andrew
