Re: [RFC] Removing "magic" oids

Hi,

On 2018-11-14 17:48:07 -0800, Andres Freund wrote:
> While clearly not ready yet, I don't think it's that far off.
>
> Missing:
> - docs polish
> - pg_upgrade early error
> - discussion of the pg_dump/restore behaviour when encountering tables
> or archives with oids. It currently warns. If we want to keep it that
> way - which I think is reasonable - a bit more code can be excised.

Attached is an updated version. It fixes all the FIXMEs/XXXs that were
left over. Most of these weren't particularly interesting, except for
the fact that GetSysCacheOid[1-4]? now has a new AttrNumber parameter
indicating the oid attribute.

I also:

- re-added printing of oids via pageinspect. As we can encounter tuples
with oids via pg_upgrade (even though the oid column must have been
removed), that seems appropriate
- merged John Naylor's genbki/reformat fixes
- performed docs polishing
- made the objectaddress.c changes more consistent
- corrected lots of formatting issues (too long lines)
- added necessary casts (mostly using ObjectIdGetDatum for oids)
- re-added AT_DropOids, so we can have a proper WITHOUT OIDS ALTER TABLE
option to ignore. I also considered supporting empty elements in the
AlterTableStmt->cmds, but that seemed uglier.
- removed #ifdefed out code
- improved error messages
- other small stuff
- re-added a few tests
- lots of small comment fixes

I'm pretty happy with the new state. Unless somebody announces they want
to do a review soon-ish, I'm planning to commit this soon. It's a
painful set to keep up2date, and it's blocking a few other patches. I'm
sure we'll find some things to adapt around the margins, but imo the
patch as a whole looks pretty reasonable.

Missing:
- nice and long commit message
- another detailed line-by-line read of the patch (last round took like
3h :()

341 files changed, 2263 insertions(+), 4249 deletions(-)

Greetings,

Andres Freund