kardash(dot)a(dot)v(at)yandex(dot)ru writes:
> Try to feed sql-commands below, then do dump and you will see errors while
> restore!
Well, this example doesn't work as-is because the command order is already
wrong. But the issue is that CREATE MATERIALIZED VIEW ... WITH NO DATA
tries to plan and execute the query, not just store it, which entirely
defeats pg_dump's attempt to avoid hidden dependencies. We need a less
chintzy solution to handling this variant of CREATE MATERIALIZED VIEW.
(I seem to recall complaining about this before.)
regards, tom lane