Horton

Re: Replication vs. float timestamps is a disaster

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>
Cc: Jim Nasby <Jim.Nasby@BlueTreble.com>, Andres Freund <andres@anarazel.de>, Petr Jelinek <petr.jelinek@2ndquadrant.com>, Robert Haas <robertmhaas@gmail.com>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2017-02-22T15:45:00Z
Lists: pgsql-hackers 
Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> On 02/22/2017 10:21 AM, Jim Nasby wrote:
>> Only in the catalog though, not the datums, right? I would think you
>> could just change the oid in the catalog the same as you would for a
>> table column.

> No, in the datums.

Yeah, I don't see any way that we could fob off float timestamps to an
extension that would be completely transparent at the pg_upgrade level.
And even a partial solution would be an enormous amount of fundamentally
dead-end work.

There has never been any project policy that promises everybody will be
able to pg_upgrade until the end of time.  I think it's not unreasonable
for us to say that anyone still using float timestamps has to go through
a dump and reload to get to v10.  The original discussion about getting
rid of them was ten years ago come May; that seems long enough.

			regards, tom lane

Commits

  1. Consistently declare timestamp variables as TimestampTz.

  2. Remove now-dead code for !HAVE_INT64_TIMESTAMP.

  3. Remove pg_control's enableIntTimes field.

  4. De-support floating-point timestamps.

  5. Make integer_datetimes the default for MSVC even if not mentioned in config.pl.

  6. Enable 64-bit integer datetimes by default, per previous discussion.