Andres Freund <andres@anarazel.de> writes:
> On Friday, March 16, 2012 10:52:55 PM Tom Lane wrote:
>> Something else I just came across is that there are assorted places that
>> are aware that ExplainStmt contains a Query, eg setrefs.c, plancache.c,
>> and those have got to treat CreateTableAsStmt similarly.
> Hm. Is that so? As implemented in my version the planner just sees a plain
> statement instead of a utility statement. Am I missing something?
Well, the work flow for EXPLAIN is:
parse analysis: recursively do parse analysis on contained queryplan: do nothingexecution: call planner on contained
query,then optionally run it
and the reason for doing it that way is explained by
transformExplainStmt:
* EXPLAIN is like other utility statements in that we emit it as a* CMD_UTILITY Query node; however, we must first
transformthe contained* query. We used to postpone that until execution, but it's really necessary* to do it during
thenormal parse analysis phase to ensure that side effects* of parser hooks happen at the expected time.
ISTM that argument applies just as much to CREATE TABLE AS, especially
in view of the fact that we're restructuring the SELECT INTO case, in
which parse analysis of the SELECT certainly does happen early. It's
also not clear to me why it wouldn't apply to COPY (SELECT ...).
I'm not going to touch the COPY (SELECT ...) issue right now, but
somebody ought to go back and check up on the exact user-visible bugs
that motivated moving EXPLAIN's parse analysis processing. (I suspect
it had to do with plpgsql variable processing, but too lazy to go look
right now.) If there's a plausible use case where similar bugs could
be exhibited in COPY, we're going to have to restructure that too.
regards, tom lane