Re: How about to have relnamespace and relrole? - Mailing list pgsql-hackers

Hello, I attached the latest patches missing in the previous mail.

Thanks for pointing Jeevan.

0001-Add-regrole_v4.patch
0002-Add-regnamespace_v4.patch

Jim> BTW, I think the potential for MVCC issues should be mentioned in the
Jim> docs (http://www.postgresql.org/docs/devel/static/datatype-oid.html).

The first patch of the aboves contains doc patch which adds the
following note to html/datatype-oid.html. Does it make sense?

> Note: The OID alias types don't sctrictly comply the transaction
>   isolation rules so do not use them where exact transaction
>   isolation on the values of these types has a
>   significance. Likewise, since they look as simple constants to
>   planner so you might get slower plans than the queries joining
>   the system tables correnspond to the OID types.

regards,


At Mon, 2 Mar 2015 17:50:37 -0600, Jim Nasby <Jim.Nasby@BlueTreble.com> wrote in <54F4F74D.8000003@BlueTreble.com>
> On 3/2/15 3:56 PM, Andres Freund wrote:
> > On 2015-03-02 16:42:35 -0500, Robert Haas wrote:
> >> >On Tue, Feb 3, 2015 at 10:12 AM, Tom Lane<tgl@sss.pgh.pa.us>  wrote:
> >>> > >Two reasons this isn't terribly compelling are (1) it's creating a
> >>> > >join in a place where the planner can't possibly see it and optimize
> >>> > >it, and (2) you risk MVCC anomalies because the reg* output routines
> >>> > >would not be using the same snapshot as the calling query.
> >>> > >
> >>> > >We already have problem (2) with the existing reg* functions so I'm
> >>> > >not that excited about doubling down on the concept.
> >> >
> >> >I think I agree.  I mean, I agree that this notation is more
> >> >convenient, but I don't really want to add a whole new slough of types
> >> >--- these will certainly not be the only ones we want once we go down
> >> >this path --- to the default install just for notational convenience.
> >> >It's arguable, of course, but I guess I'm going to vote against this
> >> >patch.
> > That's a justifyable position. I don't think there are other catalogs
> > referenced as pervasively in the catalog though.
> >
> > There's one additional point: Using reg* types in the catalog tables
> > themselves can make them*much* easier to read. I personally do look at
> > the catalogs a awful lot, and seing names instead of oids makes it
> > much
> > easier. And adding regtype/role would allow to cover nearly all types
> > containing oids.
> 
> +1. Constantly joining catalog tables together is a royal PITA, and
> regnamespace is the biggest missing part of this (I typically don't
> look at roles too much, but I can certainly see it's importance).
> 
> If we had more user friendly views on the catalogs maybe this wouldn't
> be an issue... but that's a much larger project.
> 
> BTW, I think the potential for MVCC issues should be mentioned in the
> docs (http://www.postgresql.org/docs/devel/static/datatype-oid.html).

-- 
Kyotaro Horiguchi
NTT Open Source Software Center