Re: invalid search_path complaints

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Tue, Apr 10, 2012 at 9:37 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> Well, the other thing we could do is tweak the rules for when to print a
>> complaint. I notice that in check_temp_tablespaces we use the rule
>>
>> source == PGC_S_SESSION (ie, SET) -> error
>> source == PGC_S_TEST (testing value for ALTER SET) -> notice
>> else -> silently ignore bad name
>>
>> which seems like it could be applied to search_path without giving
>> anyone grounds for complaint. I'm still in favor of the previous patch
>> for HEAD, but maybe we could do this in 9.1.

> Would that amount to removing the WARNING that was added in 9.1? If
> so, I think I could sign on to that proposal.

It would remove the warning that occurs while applying ALTER ... SET
values. Another case that would change behavior is PGC_S_CLIENT;
I observe that 9.1 rejects bad settings there entirely:

$ PGOPTIONS="--search_path=foo" psql
psql: FATAL: invalid value for parameter "search_path": "foo"
DETAIL: schema "foo" does not exist

but this did not happen in 9.0 so that seems like an improvement too.
I believe that the other possible source values all correspond to cases
where check_search_path would be executed outside a transaction and so
would not do the check in question anyway. I've not tried to prove
that exhaustively though.

regards, tom lane