On Mon, Aug 5, 2019 at 2:35 PM Stephen Frost <sfrost@snowman.net> wrote:
> I dare say that if we had some documentation around what to expect and how to deal with it, for our own tools as well
asexternal ones, then maybe we wouldn’t be in this situation in the first place. Clearly ALTER SYSTEM and the
pg_basebackupmodifications had different understands and expectations.
But that's not the problem. The problem is that ALTER SYSTEM modifies
the first match instead of the last one, when it's a well-established
principle that when reading from a PostgreSQL configuration file, the
system adopts the value from the last match, not the first one. I
admit that if somebody had thought to document what ALTER SYSTEM was
doing, that person would probably have also realized that they had
made a mistake in the code, and then they would have fixed the bug,
and that would be great.
But we have exactly zero need to invent a new set of principles
explaining how to deal with postgresql.auto.conf. We just need to
make the ALTER SYSTEM code conform to the same general rule that has
been well-established for many years. The only reason why we're still
carrying on about this 95 messages later is because you're trying to
make an argument that postgresql.auto.conf is a different kind of
thing from postgresql.conf and therefore can have its own novel set of
rules which consequently need to be debated. IMHO, it's not, it
shouldn't, and they don't.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company