Hi,
On 2019-02-27 09:32:17 -0300, Alvaro Herrera wrote:
> On 2019-Feb-27, Dmitry Dolgov wrote:
> > But I hope there are no objections if I'll then submit the original
> > changes with more consistent null handling separately to make decision
> > about them more consciously.
>
> I think we should save such a patch for whenever we next update the
> archive version number, which could take a couple of years given past
> history. I'm inclined to add a comment near K_VERS_SELF to remind
> whoever next patches it.
The pluggable storage patchset contains exactly that... I've attached
the precursor patch (CREATE ACCESS METHOD ... TYPE TABLE), and the patch
for pg_dump support. They need a bit more cleanup, but it might be
useful information for this thread.
One thing I want to bring up here rather than in the pluggable storage
thread is that currently the pg_dump support for access methods deals
with table access methods in a manner similar to the way we deal with
tablespaces. Instead of specifying the AM on every table creation, we
set the default AM when needed. That makes it easier to adjust dumps.
But it does basically require breaking archive compatibility. I
personally am OK with that, but I thought it might be worth discussing.
I guess we could try avoid the compat issue by only increasing the
archive format if there actually are any non-default AMs, but to me that
doesn't seem like an improvement worthy of the necessary complications.
Greetings,
Andres Freund