Hi,
On 04/28/15 21:50, Tom Lane wrote:
> Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
>> the attached trivial patch adds handling of RestrictInfo nodes into
>> expression_tree_walker().
>
> RestrictInfo is not a general expression node and support for it has
> been deliberately omitted from expression_tree_walker(). So I think
> what you are proposing is a bad idea and probably a band-aid for some
> other bad idea.
>
>> This is needed for example when calling
>> pull_varnos or (or other functions using the expression walker) in
>> clausesel.c, for example. An example of a query causing errors with
>> pull_varnos is
>
>> select * from t where (a >= 10 and a <= 20) or (b >= 15 and b <= 20);
>
> Really?
>
> regression=# create table t (a int, b int);
> CREATE TABLE
> regression=# select * from t where (a >= 10 and a <= 20) or (b >= 15 and b <= 20);
> a | b
> ---+---
> (0 rows)
That's not what I said, though. I said that calling pull_varnos() causes
the issue - apparently that does not happen in master, but I ran into
that when hacking on a patch.
For example try adding this
Relids tmp = pull_varnos(clause); elog(WARNING, "count = %d", bms_num_members(tmp));
into the or_clause branch in clause_selectivity(), and then running the
query will give you this:
db=# select * from t where (a >= 10 and a <= 20) or (b >= 15);
ERROR: unrecognized node type: 524
But as I said - maybe calls to pull_varnos are not supposed to happen in
this part of the code, for some reason, and it really is a bug in my patch.
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services