On 29 January 2016 at 23:59, Robert Haas <robertmhaas@gmail.com> wrote: > On Fri, Jan 29, 2016 at 5:24 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: >> Alexander Korotkov <a.korotkov@postgrespro.ru> writes: >>> On Fri, Jan 29, 2016 at 6:36 PM, Alvaro Herrera <alvherre@2ndquadrant.com> >>> wrote: >>>> I'm thinking we'd do CREATE ACCESS METHOD foobar TYPE INDEX or something >>>> like that. >> >>> I would prefer "CREATE {INDEX | SEQUENCE | ... } ACCESS METHOD name HANDLER >>> handler;", but I don't insist. >> >> I think that Alvaro's idea is less likely to risk future grammar >> conflicts. We've had enough headaches from CREATE [ UNIQUE ] INDEX >> [ CONCURRENTLY ] to make me really wary of variables in the statement-name >> part of the syntax. > > Strong +1. If we put the type of access method immediately after > CREATE, I'm almost positive we'll regret it for exactly that reason. >
Just as a note, CREATE SEQUENCE ACCESS METHOD already causes grammar conflict now, that's why my proposal was different, I didn't want to add more keywords. I think Alvaro's proposal is fine as well.
The other point is that we are creating ACCESS METHOD object so that's what should be after CREATE.
In any case this is slightly premature IMHO as DDL is somewhat unless until we have sequence access methods implementation we can agree on, or the generic WAL logging so that custom indexes can be crash safe.
I've changed syntax of CREATE ACCESS METHOD syntax in the thread about index access method extendability.
Now it is "CREATE ACCESS METHOD name TYPE INDEX HANDLER handler;". New column amtype of pg_am stores access method type.