On Friday, April 3, 2026, Robert Treat <
rob@xzilla.net> wrote:
On Fri, Apr 3, 2026 at 3:35 PM Andreas Karlsson <andreas@proxel.se> wrote:
> On 4/3/26 9:27 PM, Andreas Karlsson wrote:
> > On 4/3/26 8:18 PM, David G. Johnston wrote:
> >> Per the discussion on -general [1] I propose that we stop using the
> >> generic label storage_parameter on the create table reference page and
> >> instead set up proper labels for table and index variants.
> >
It's sort of interesting that no one in the above discussion gave an
example like:
create table t (c int, constraint pk primary key (c) with (fillfactor
= 90)) with (fillfactor = 100);
and pointing out that where you put the parameter changes what it
effects, so I'm a little skeptical that this patch would help the
original discussion, but it certainly wouldn't hurt, so +1 from me.
Stretches my skills a bit but if the error message had been: (unrecognized index storage parameter “autovacuum_enabled”) the syntax placement issue may have been recognized working under the assumption the author knows their intent is to modify the table (and vice-versa).
As a quick footgun prevention attempt (not a huge fan though):
Note, specifying a primary key or unique constraint as the final component of create table makes assigning storage parameters to the wrong object more likely.
(Added to the with clause section)
David J.
