Re: [BUG] pg_dump does not properly deal with BEGIN ATOMIC function

"Imseih (AWS), Sami" <simseih(at)amazon(dot)com> writes:
> With the attached repro, pg_restore fails with
> pg_restore: error: could not execute query: ERROR: constraint "a_pkey" for table "a" does not exist
> Command was: CREATE FUNCTION public.a_f(c1_in text, c2 integer DEFAULT 60) RETURNS void

Hmph. The other thing worth noticing is that pg_dump prints
a warning:

pg_dump: warning: could not resolve dependency loop among these items:

or with -v:

pg_dump: warning: could not resolve dependency loop among these items:
pg_dump: FUNCTION a_f (ID 218 OID 40664)
pg_dump: CONSTRAINT a_pkey (ID 4131 OID 40663)
pg_dump: POST-DATA BOUNDARY (ID 4281)
pg_dump: TABLE DATA a (ID 4278 OID 40657)
pg_dump: PRE-DATA BOUNDARY (ID 4280)

So it's lacking a rule to tell it what to do in this case, and the
default is the wrong way around. I think we need to fix it in
about the same way as the equivalent case for matviews, which
leads to the attached barely-tested patch.

BTW, now that I see a case the default printout here seems
completely ridiculous. I think we need to do

pg_log_warning("could not resolve dependency loop among these items:");
for (i = 0; i < nLoop; i++)
{
char buf[1024];

describeDumpableObject(loop[i], buf, sizeof(buf));
- pg_log_info(" %s", buf);
+ pg_log_warning(" %s", buf);
}

but I didn't actually change that in the attached.

regards, tom lane