> On Thu, Oct 21, 2021 at 07:16:00PM -0700, Peter Geoghegan wrote:
>
> My general concern is that the skip scan patch may currently be
> structured in a way that paints us into a corner, MDAM-wise.
>
> Note that the MDAM paper treats skipping a prefix of columns as a case
> where the prefix is handled by pretending that there is a clause that
> looks like this: "WHERE date between -inf AND +inf" -- which is not so
> different from the original sales SQL query example that I have
> highlighted. We don't tend to think of queries like this (like my
> sales query) as in any way related to skip-scan, because we don't
> imagine that there is any skipping going on. But maybe we should
> recognize the similarities.
To avoid potential problems with extensibility in this sense, the
implementation needs to explicitly work with sets of disjoint intervals
of values instead of simple prefix size, one set of intervals per scan
key. An interesting idea, doesn't seem to be a big change in terms of
the patch itself.