Re: memory context for tuplesort return values

Alvaro Herrera <alvherre(at)commandprompt(dot)com> writes:
> Crazy ideas: add a #warning or #error to the header file unless there is
> some special symbol previously defined, something like
> #define I_UNDERSTAND_MEM_ALLOC_IN_TUPLETABLESLOT
> which means the developer did update his code.

Well, if we wanted to go that far, the simple solution is to rename
tuplesort_gettuple to something else. Guaranteed code breakage then.
However, I think this is overkill for a problem that only *might*
bite people. I kinda doubt that people are using TupleTableSlots in
add-on code, because the design of the executor makes it hard for
anything except plan nodes to obtain slots. You'd have to posit a
piece of code that isn't structured as

while ((tuple = tuplesort_gettuple(...)) != NULL)
{
process tuple;
if (should_free)
pfree(tuple);
}

but instead hangs onto the last tuple past the sort shutdown, and
yet isn't a plan node.

regards, tom lane