On Wed, Nov 26, 2025 at 1:52 PM John Naylor <johncnaylorls@gmail.com> wrote:
> This made me think of something tangential: v4's common prefix
> skipping doesn't take into account that the upper 4 bytes can't
> matter. With a mix of positive and negative integers, I think it will
> do the radix sort on all 8 bytes of "datum1".
I accidentally edited out some context: The above is referring to SQL
"int" sort keys, i.e. 32-bit signed integers.
--
John Naylor
Amazon Web Services