On Tue, May 28, 2024 at 9:06 PM Hannu Krosing <han...@google.com> wrote: > We should definitely also fix pg_dump, likely just checking that the > role exists when generating REVOKE commands (may be a good practice > for other cases too so instead of casting to ::regrole do the actual > join) > > ## here is the fix for pg_dump > > While flying to Vancouver I looked around in pg_dump code, and it > looks like the easiest way to mitigate the dangling pg_init_priv > entries is to replace the query in pg_dump with one that filters out > invalid entries
+1 for this approach. I agree with Tom that fixing this in REVOKE is a bad plan; REVOKE is used by way too many things other than pg_dump, and the behavior change is not in general desirable. -- Robert Haas EDB: http://www.enterprisedb.com
