On Fri, Aug 30, 2024 at 12:18 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
Morris de Oryx <morrisdeoryx@gmail.com> writes: > From what I've seen in the wild, and can sort out from the source, I think > that ltree does *not* need to load rows from heap.
The comment in ltree_consistent is pretty definitive:
/* All cases served by this function are exact */ *recheck = false;
> I wonder because an ltree GiST index is "lossy" and this behavior is more > like a lossless strategy. I think that's either because I've misunderstood > what "lossy" means in this case, or it's because ltree GiST index *pages *are > based on a signature (lossy), while ltree GiST index *leaf entries* contain > the full tree/path (lossless.)
Yeah, the code is not terribly well commented but this bit in ltree.h appears to be saying that leaf entries contain the original ltree:
* type of index key for ltree. Tree are combined B-Tree and R-Tree * Storage: * Leaf pages * (len)(flag)(ltree) * Non-Leaf * (len)(flag)(sign)(left_ltree)(right_ltree) * ALLTRUE: (len)(flag)(left_ltree)(right_ltree)
and that seems consistent with the fact that ltree_consistent does different things at leaf and non-leaf levels.
