Re: pg_dump bug in 9.4beta2 and HEAD

I have pushed this fix, except that instead of parsing the OID from the
dropStmt as in your patch, I used te->catalogId.oid, which is much
simpler.

I tested this by pg_restoring to 8.4 (which doesn't have
pg_largeobject_metadata); there is no error raised:

LOG: sentencia: SELECT CASE WHEN EXISTS(SELECT 1 FROM pg_catalog.pg_largeobject WHERE loid = '43748') THEN pg_catalog.lo_unlink('43748') END;

In 9.0 the command is the new style:

LOG: sentencia: SELECT pg_catalog.lo_unlink(oid) FROM pg_catalog.pg_largeobject_metadata WHERE oid = '43748';

So it's all fine. I guess it's fortunate that we already had the
DropBlobIfExists() function.

Now a further problem I notice is all the *other* commands for which we
inject the IF EXISTS clause; there is no support for those in older
servers, so they throw errors if --if-exists is given in the pg_restore
line. I think we can just say that --if-exists is not supported for
older servers; as Heikki said, we don't promise that pg_dump is
compatible with older servers anyway. In my test database, several
commands errored out when seeing the EXTENSION in CREATE EXTENSION IF EXISTS.
So we're okay now.

--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services