Re: POC, WIP: OR-clause support for indexes

On Fri, Oct 4, 2024 at 7:45 AM Alexander Korotkov <aekorotkov(at)gmail(dot)com> wrote:
> Andrei, thank you for your opinion. Just for the record, I'm still
> exploring this and will reply later today or tomorrow.

The logic that allows this to work for the case of IN() lists appears
in transformAExprIn(), which is in parse_expr.c. I wonder if it would
be possible to do something similar at the point where the patch does
its conversion to a SAOP. What do you think?

The transformAExprIn() logic doesn't directly care about operator
families. It works by using coercions, which opfamily authors are
formally required to promise cannot affect sort order. According to
the sgml docs: "Another requirement for a multiple-data-type family is
that any implicit or binary-coercion casts that are defined between
data types included in the operator family must not change the
associated sort ordering".

This logic seems to always do the right thing for cases like my IN()
test case from today, which should have an array of the type of the
widest integer type from btree/integer_ops (so a bigint[] SAOP for
that specific test case). There won't ever be a "cannot coerce to
common array type" error because logic in select_common_type() aims to
choose a common array type that every individual expression can be
implicitly cast to. It can fail to identify a common type, but AFAICT
only in cases where that actually makes sense.

--
Peter Geoghegan