Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Actually, I think this is a bug and the right thing is to make the code
> match the documentation not vice versa. ev_attr isn't being used for
> much at the moment, but if it were being used as an AttrNumber, -1 would
> not mean "whole row". It would be a reference to the system column
> with number -1 (ctid, if memory serves). Zero is the usual choice for a
> whole-row reference.
I figured that since the docs for all supported production versions
give incorrect information, I should backpatch this, which I did
before seeing your post.
I assume that this should be a 9.3 code fix, and a doc fix prior to
that, since it would require changing catalogs and might break
existing user queries? Should the docs mention the value used in
each version, or be changed to just be silent on the issue?
Such a change would require a catversion bump.
Such a change would require mention in the release notes because
existing user queries against pg_rewrite might fail unless
adjusted.
Is it worth doing that now, versus when and if the hypothetical
change to reference a column is made?
-Kevin