On 2011-07-22 16:17, Hitoshi Harada wrote:
>
> :(
> I updated the patch. Could you try attached once more? The "issafe"
> switch seems wrong.
Works like a charm :-). However, now there is always a copyObject of a
subquery even when the subquery is not safe for qual pushdown. The
problem with the previous issafe was that it was only assigned for
rel->baserestrictinfo != NIL. If it is assigned before the if statement,
it still works. See attached patch that avoids subquery copy for unsafe
subqueries, and also exits best_inner_subqueryscan before palloc of
differenttypes in case of unsafe queries.
regards,
--
Yeb Havinga
http://www.mgrid.net/
Mastering Medical Data