Horton

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

Peter Geoghegan <pg@bowt.ie>

View thread
From: Peter Geoghegan <pg@bowt.ie>
To: Alexander Korotkov <aekorotkov@gmail.com>
Cc: Andrei Lepikhov <lepihov@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Alena Rybakina <a.rybakina@postgrespro.ru>, jian he <jian.universality@gmail.com>, Nikolay Shaplov <dhyan@nataraj.su>, pgsql-hackers@lists.postgresql.org, Robert Haas <robertmhaas@gmail.com>, pgsql-hackers@postgresql.org, Marcos Pegoraro <marcos@f10.com.br>, teodor@sigaev.ru, Peter Eisentraut <peter@eisentraut.org>, Ranier Vilela <ranier.vf@gmail.com>
Date: 2024-10-04T18:19:54Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Make group_similar_or_args() reorder clause list as little as possible

  2. Allow usage of match_orclause_to_indexcol() for joins

  3. Skip not SOAP-supported indexes while transforming an OR clause into SAOP

  4. Remove the wrong assertion from match_orclause_to_indexcol()

  5. Teach bitmap path generation about transforming OR-clauses to SAOP's

  6. Transform OR-clauses to SAOP's during index matching

  7. Fix the value of or_to_any_transform_limit in postgresql.conf.sample

  8. Transform OR clauses to ANY expression

  9. MergeAttributes code deduplication

  10. SEARCH and CYCLE clauses

  11. Improve estimation of OR clauses using extended statistics.

  12. Teach btree to handle ScalarArrayOpExpr quals natively.

  13. Revise collation derivation method and expression-tree representation.

  14. Instead of trying to force WHERE clauses into CNF or DNF normal form,

On Fri, Oct 4, 2024 at 2:00 PM Alexander Korotkov <aekorotkov@gmail.com> wrote:
> Yes, transformAExprIn() does the work to coerce all the expressions in
> the right part to the same type.  Similar logic could be implemented
> in match_orclause_to_indexcol().  What worries me is whether it's
> quite late stage for this kind of work.  transformAExprIn() works
> during parse stage, when we need to to resolve types, operators etc.
> And we do that once.

I agree that it would be a bit awkward. Especially having spent so
much time talking about doing this later on, not during parsing. That
doesn't mean that it's necessarily the wrong thing to do, though.

> If we replicate the same logic to
> match_orclause_to_indexcol(), then we may end up with index scan using
> one operator and sequential scan using another operator.

But that's already true today. For example, these two queries use
different operators at runtime, assuming both use a B-Tree index scan:

select * from tenk1 where four = any('{0,1}'::int[]) and four =
any('{1,2}'::bigint[]);

select * from tenk1 where four = any('{1,2}'::bigint[]) and four =
any('{0,1}'::int[]); -- flip the order of the arrays, change nothing
else

This isn't apparent from what EXPLAIN ANALYZE output shows, but the
fact is that only one operator (and one array) will be used at
runtime, after nbtree preprocessing completes. I'm not entirely sure
how this kind of difference might affect a sequential scan. I imagine
that it can use either or both operators unpredictably.

> Given we
> only use implicit casts for types coercion those are suppose to be
> strong equivalents.  And that's for sure true for builtin types and
> operators. But isn't it too much to assume the same for all
> extensions?

Anything is possible. But wouldn't that also mean that the extensions
were broken with the existing IN() list thing, in transformAExprIn()?
What's the difference, fundamentally?

-- 
Peter Geoghegan