Re: pg_stat_get_replication_slot() marked not strict, crashes

Hi,

On 2022-03-28 08:28:29 +0530, Amit Kapila wrote:
> I am not sure if for 14 we can make a catalog change as that would
> require catversion bump, so adding a code-level check as suggested by
> Andres seems like a better option. Andres/Tom, any better ideas for
> this?

I think we could do the catalog change too, so that future initdb's are marked
correctly. But we obviously do need the code-level check nevertheless.

> Thanks for the patch but for HEAD, we also need handling and test for
> pg_stat_get_subscription_stats. Considering this for HEAD, we can mark
> both pg_stat_get_replication_slot and pg_stat_get_subscription_stats
> as strict and in 14, we need to add a code-level check for
> pg_stat_get_replication_slot.

FWIW, I have a test for both, I was a bit "stuck" on where to put the
pg_stat_get_subscription_stats(NULL) test. I had put the
pg_stat_get_replication_slot(NULL) in contrib/test_decoding/sql/stats.sql
but pg_stat_get_subscription_stats() doesn't really fit there. I think I'm
coming down to putting a section of such tests into src/test/regress/sql/stats.sql
instead. In the hope of preventing future such occurrances by encouraging
people to copy the test...

Greetings,

Andres Freund