On Tue, May 31, 2022 at 02:30:32PM +0200, Alvaro Herrera wrote:
> I was thinking the opposite: REINDEX DATABASE with or without a database
> name should always process the user relations and skip system catalogs.
> If the user wants to do both, then they can use REINDEX SYSTEM in
> addition.
>
> The reason for doing it like this is that there is no way to process
> only user tables and skip catalogs. So this is better for
> composability.
No objections from me to keep this distinction at the end, as long as
the the database name in the command has no impact on the chosen
behavior. Could there be a point in having a REINDEX ALL though that
would process both the user relations and the catalogs, doing the same
thing as REINDEX DATABASE today?
By the way, the patch had better avoid putting a global REINDEX
command that would process everything. As far as I recall, we've
avoided such things on purpose because they are expensive, keeping
around only cases that generate errors or skip all the relations.
So having that in a TAP test would be better, I assume, for
isolation.
--
Michael