Re: Why don't we have a small reserved OID range for patch revisions?

On Thu, Feb 28, 2019 at 5:36 PM Peter Geoghegan <pg@bowt.ie> wrote:
> I have attempted to institute some general guidelines for what the
> thicket of rules are by creating the "committing checklist" page. This
> is necessarily imperfect, because the rules are in many cases open to
> interpretation, often for good practical reasons. I don't have any
> sympathy for committers that find it hard to remember to do a
> catversion bump with any kind of regularity. That complexity seems
> inherent, not incidental, since it's often convenient to ignore
> catalog incompatibilities during development.

Bumping catversion is something of a special case, because one does it
so often that one gets used to remembering it.  The rules are usually
not too hard to remember, although they are trickier when you don't
directly change anything in src/include/catalog but just change the
definition of some node that may be serialized in a catalog someplace.
It would be neat if there were a tool you could run to somehow tell
you whether catversion needs to be changed for a given patch.

But you know there are a log of other version numbers floating around,
like XLOG_PAGE_MAGIC or the pg_dump archive version, and it is not
really that easy to know -- as a new contributor or sometimes even as
an experienced one -- whether your work requires any changes to that
stuff, or even that that stuff *exists*.  Indeed, XLOG_PAGE_MAGIC is a
particularly annoying case, both because the constant name doesn't
contain VERSION and because the comment just says /* can be used as
WAL version indicator */ which does not exactly make it clear that if
you fail to bump it when you touch the WAL format you will Make People
Unhappy.

Indeed, Simon got complaints a number of years ago (2010, it looks
like) when he had the temerity to change the magic number to some
other unrelated value instead of just incrementing it by one.
Although I think that the criticism was to a certain extent
well-founded -- why deviate from previous practice? -- there is at the
same time something a little crazy about somebody getting excited
about the particular value that has been chosen for a number that is
described in the very name of the constant as a MAGIC number.  And
especially because there is absolutely zip in the way of code comments
or a README that explain to you how to do it "right."

> > So, I suspect that for every
> > unit of work it saves somebody, it's probably going to generate about
> > one unit of extra work for somebody else.
>
> Maybe so. I think that you're jumping to conclusions, though.

I did say "I suspect..." which was intended as a concession that I
don't know for sure.

> But you seem to want to make the mechanism itself even more
> complicated, not less complicated (based on your remarks about making
> OID assignment happen during the build). In order to make the use of
> the mechanism easier. That seems worth considering, but ISTM that this
> is talking at cross purposes. There are far simpler ways of making it
> unlikely that a committer is going to miss this step. There is also a
> simple way of noticing that they do quickly (e.g. a simple buildfarm
> test).

Well, perhaps I'm proposing some additional code, but I don't think of
that as making the mechanism more complicated.  I want to make it
simpler for patch submitters and reviewers and committers to not make
mistakes that they have to run around and fix.  If there are fewer
kinds of things that qualify as mistakes, as in Tom's latest proposal,
then we are moving in the right direction IMO regardless of anything
else.

> > OK.  Well, I think that doing nothing is superior to this proposal,
> > for reasons similar to what Peter Eisentraut has already articulated.
> > And I think rather than blasting forward with your own preferred
> > alternative in the face of disagreement, you should be willing to
> > discuss other possible options.  But if you're not willing to do that,
> > I can't make you.
>
> Peter seemed to not want to do this on the grounds that it isn't
> necessary at all, whereas you think that it doesn't go far enough. If
> there is a consensus against what Tom has said, it's a cacophonous one
> that cannot really be said to be in favor of anything.

I think Peter and I are more agreeing than we are at the opposite ends
of a spectrum, but more importantly, I think it is worth having a
discussion first about what people like and dislike, and what goals
they have, and then only if necessary, counting the votes afterwards.
I don't like having the feeling that because I have a different view
of something and want to write an email about that, I am somehow an
impediment to progress.  I think if we reduce discussions to
you're-for-it-or-your-against-it, that's not that helpful.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company