Re: Improving test coverage of extensions with pg_dump

On Wed, Sep 16, 2015 at 8:00 PM, Michael Paquier wrote:
> Hm. OK. I didn't get your message correctly, sorry for that. Would you
> be fine then to have a pg_regress command using parallel_schedule + an
> extra schedule launching tests related to the extensions in
> src/test/modules/pg_dumprestore then? The choice of parallel_schedule
> is based on what Windows does, aka this schedule is used in the
> equivalent of make check in vcregress.pl. The TAP script then simply
> initializes the cluster, runs pg_regress, and does the dump/restore
> job. There is no real need to worry about setval as dump is not taken
> from a standby..

So, here we go.

I have found something quite interesting when playing with the patch
attached: dump does not guarantee the column ordering across databases
for some inherited tables, see that example from the main regression
test suite the following diff between a dump taken from a source
database and a target database where the source dump has been restored
in first:
-INSERT INTO b_star (class, aa, bb, a) VALUES ('b', 3, 'mumble', NULL);
-INSERT INTO b_star (class, aa, bb, a) VALUES ('b', 4, NULL, NULL);
-INSERT INTO b_star (class, aa, bb, a) VALUES ('b', NULL, 'bumble', NULL);
-INSERT INTO b_star (class, aa, bb, a) VALUES ('b', NULL, NULL, NULL);
+INSERT INTO b_star (class, aa, a, bb) VALUES ('b', 3, NULL, 'mumble');
+INSERT INTO b_star (class, aa, a, bb) VALUES ('b', 4, NULL, NULL);
+INSERT INTO b_star (class, aa, a, bb) VALUES ('b', NULL, NULL, 'bumble');
+INSERT INTO b_star (class, aa, a, bb) VALUES ('b', NULL, NULL, NULL);

Problem is similar with --column-inserts, --inserts and COPY. We could
use --exclude-table like in the patch attached when taking the dump
from source database but that's grotty, or we could improve pg_dump
itself, though it may not be worth it for just this purpose.
Thoughts?
--
Michael