Re: BUG #15338: pg_restore --disable-triggers --data-only AND schema for table is not set.

Sergei Kornilov <sk(at)zsrv(dot)org> writes:
> Plain format looks good:
> /usr/lib/postgresql/10/bin/pg_dump bug15338 --disable-triggers --data-only # all ok!
> But custom and dir formats are wrong:
> /usr/lib/postgresql/10/bin/pg_dump bug15338 --disable-triggers --data-only -Fc --file bug15338.pgdump
> /usr/lib/postgresql/10/bin/pg_restore bug15338.pgdump --disable-triggers --data-only
> Produces wrong ALTER TABLE test DISABLE TRIGGER ALL, not testschema.test

That's weird [ digs ] ... oh, here's the problem:

ahprintf(AH, "ALTER TABLE %s DISABLE TRIGGER ALL;\n\n",
fmtQualifiedId(PQserverVersion(AH->connection),
te->namespace,
te->tag));

This code is fine as long as we're connected to a database (either
for dump or restore). But if we're not, AH->connection is NULL.
PQserverVersion doesn't crash, but silently returns 0 ... which
fmtQualifiedId interprets as "too old for schemas", so it doesn't
emit the schema part.

We could make a narrow fix right here (and in the matching ENABLE stanza);
but having seen this, it seems like a potentially generic problem.
I think we'd be smarter to try to fix this by-design.

In particular, since no current version of pg_dump even pretends either
to dump from pre-8.0 servers or to produce output that could be loaded
into a pre-7.3 server without adjustment, it doesn't seem like
fmtQualifiedId should consider server version at all. I was tempted
to remove its server-version argument in 64f3524e2, but refrained ---
now that seems like a mistake.

However, that doesn't work for pre-v10 since we still supported dumping
from pre-7.3 servers then, and some of the uses of fmtQualifiedId are
for commands to be sent to the source server rather than commands to
be emitted. In the back branches, probably the best thing is just to
assume that the output is for >= 7.3 when producing DISABLE/ENABLE
TRIGGERS commands.

regards, tom lane