On Mon, Mar 4, 2024 at 1:11 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> ... This morning I've been
> wondering about a highly targeted solution to that: could we recognise
> this condition cheaply, and then activate a special case that spills a
> *bucket* to disk? That is, a problem bucket that is indivisible by
> hashing because all the hash values are the same. Or something like
> that.
(One of the main problems being how to manage the match bits for outer
joins, a big problem also for idea #2. But maybe it's somehow more
tractable there, I dunno. I'd need vastly more coffee to have any
idea...)