Re: [HACKERS] Re: Apparent bug in _make_subplan

Vadim Mikheev <vadim(at)krs(dot)ru> writes:
>> But the varlevelsup I pass in from the parser are relative to the
>> current level, not absolute.

> subselect.c takes it into account, computes absolute numbers
> and stores them in PlannerParamVar only...

Right, I eventually figured that out, and I see that it's probably the
best way. I have added the following documentation to subselect.c:

/*--------------------
* PlannerParamVar is a list of Var nodes, wherein the n'th entry
* (n counts from 0) corresponds to Param->paramid = n. The Var nodes
* are ordinary except for one thing: their varlevelsup field does NOT
* have the usual interpretation of "subplan levels out from current".
* Instead, it contains the absolute plan level, with the outermost
* plan being level 1 and nested plans having higher level numbers.
* This nonstandardness is useful because we don't have to run around
* and update the list elements when we enter or exit a subplan
* recursion level. But we must pay attention not to confuse this
* meaning with the normal meaning of varlevelsup.
*--------------------
*/

along with other changes that I will commit once I get subselects in
HAVING working right ...

regards, tom lane