Re: Is tuplesort_heap_siftup() a misnomer?

Peter Geoghegan <pg(at)heroku(dot)com> writes:
> Attached patch does it that way, then. I stuck with the reference to
> "shift down", though, since I think we all agree that that is
> unambiguous.

I dunno. What you've now got is

/*
* The tuple at state->memtuples[0] has been removed from the heap.
- * Decrement memtupcount, and sift up to maintain the heap invariant.
+ * Decrement memtupcount, and shift down to maintain the heap invariant.
*/
static void
-tuplesort_heap_siftup(Tuplesortstate *state, bool checkIndex)
+tuplesort_heap_delete_top(Tuplesortstate *state, bool checkIndex)

I don't find this clearer at all, because

(1) As noted in the comment, the heap top has *already* been removed,
we're just trying to re-establish the heap invariant. So maybe
"delete_top" isn't the best name after all.

(2) "shift down" seems a bit weird, since we're moving data closer to
the heap top, not further away from it; why isn't that direction "up"?

(3) "shift" makes it sound like it ought to be a simple memmove
kind of operation, which it surely is not.

regards, tom lane