>>>>> "Aleksandr" == Aleksandr Parfenov <a.parfe...@postgrespro.ru> writes:
Aleksandr> The idea of the fix for this situation is to check is a Aleksandr> result of the function constant or not during the planning Aleksandr> of the query. Attached patch does this by processing Var Aleksandr> entries at planner stage and replace them with constant Aleksandr> value if it is possible. Plans after applying a patch Aleksandr> (SeqScan query for comparison): >From an implementation point of view your patch is obviously broken in many ways (starting with not checking varattno anywhere, and not actually checking anywhere if the expression is volatile). But more importantly the plans that you showed seem _worse_ in that you've created extra calls to to_tsquery even though the query has been written to try and avoid that. I think you need to take a step back and explain more precisely what you think you're trying to do and what the benefit (and cost) is. Specific coding style points (not exhaustive): Aleksandr> pointedNode->functions->length == 1 list_length() Aleksandr> pointedNode->functions->head->data.ptr_value linitial() or linitial_node() Aleksandr> if (result->type == T_FuncExpr) IsA() (what if the result is the inline expansion of a volatile function?) Aleksandr> pfree(result); result is a whole tree of nodes, pfree is pointless here (don't bother trying to fix the above points in this patch, that won't address the fundamental flaws) -- Andrew (irc:RhodiumToad)
