Re: Online checksums verification in the backend

On Sun, Apr 05, 2020 at 06:08:06PM +0900, Masahiko Sawada wrote:
> 
> Why do we need two rows in the doc? For instance, replication slot
> functions have some optional arguments but there is only one row in
> the doc. So I think we don't need to change the doc from the previous
> version patch.
> 

I thought that if we document the function as pg_check_relation(regclass [,
fork]) users could think that the 2nd argument is optional, so that
pg_check_relation('something', NULL) could be a valid alias for the 1-argument
form, which it isn't.  After checking, I see that e.g. current_setting has the
same semantics and is documented the way you suggest, so fixed back to previous
version.

> And I think these are not necessary as we already defined in
> include/catalog/pg_proc.dat:
> 
> +CREATE OR REPLACE FUNCTION pg_check_relation(
> +  IN relation regclass,
> +  OUT relid oid, OUT forknum integer, OUT failed_blocknum bigint,
> +  OUT expected_checksum integer, OUT found_checksum integer)
> +  RETURNS SETOF record STRICT VOLATILE LANGUAGE internal AS 'pg_check_relation'
> +  PARALLEL RESTRICTED;
> +
> +CREATE OR REPLACE FUNCTION pg_check_relation(
> +  IN relation regclass, IN fork text,
> +  OUT relid oid, OUT forknum integer, OUT failed_blocknum bigint,
> +  OUT expected_checksum integer, OUT found_checksum integer)
> +  RETURNS SETOF record STRICT VOLATILE LANGUAGE internal
> +  AS 'pg_check_relation_fork'
> +  PARALLEL RESTRICTED;
> 

Oh right this isn't required since there's no default value anymore, fixed.

v9 attached.