Обсуждение: Suggest using boolean index with (bflag is true) condition for the query with (bflag = true) clause

This is not a bug report but rather a suggestion.

Suppose database contains an index on a boolean column like this:
      create table1(id int, amount float, best_record boolean);
      create index index1 on table1(best_record) where best_record is true;

And suppose we issue a query like this:
      select * from table1 where best_record = true;

The planner does not try to use an index for this query because an index condition (bflag is true) does not literally
matchthe query (bflag = true).  
But is that reasonable?
The index does not cover records where "best_record" is FALSE or NULL.
And similarly the query is not interested in any records where "best_record" is FALSE or NULL.
So why not use this index for this query?
Thank you.

On Sun, 2022-09-18 at 18:24 +0300, Евгений Плискин wrote:
> This is not a bug report but rather a suggestion.
> 
> Suppose database contains an index on a boolean column like this:
>       create table1(id int, amount float, best_record boolean);
>       create index index1 on table1(best_record) where best_record is true;
> 
> And suppose we issue a query like this:
>       select * from table1 where best_record = true;
> 
> The planner does not try to use an index for this query because an index condition (bflag is true) does not literally
matchthe query (bflag = true). 
 
> But is that reasonable?
> The index does not cover records where "best_record" is FALSE or NULL.
> And similarly the query is not interested in any records where "best_record" is FALSE or NULL.
> So why not use this index for this query?

Because the conditions are different:

SELECT NULL = TRUE, NULL IS TRUE;
 ?column? │ ?column? 
══════════╪══════════
          │ f
(1 row)

The first result is NULL.

Yours,
Laurenz Albe
-- 
Cybertec | https://www.cybertec-postgresql.com

Laurenz Albe <laurenz.albe@cybertec.at> writes:
> On Sun, 2022-09-18 at 18:24 +0300, Евгений Плискин wrote:
>> So why not use this index for this query?

> Because the conditions are different:

> SELECT NULL = TRUE, NULL IS TRUE;
>  ?column? │ ?column? 
> ══════════╪══════════
>           │ f
> (1 row)

> The first result is NULL.

Nonetheless, indxpath.c knows it can transform between "bool = true"
and "bool IS TRUE" for the purpose of making an index search
qualification, so it seems a bit odd that we fail to do the equivalent
transformation when attempting to prove an index predicate.

It'd be possible to improve this by adding some proof rules to
predicate_implied_by_simple_clause: I think both "x => x IS TRUE"
and the converse would be valid per the proof requirements, and
if you wanted to gild the lily it'd likely be possible to handle
some related cases like "x => x IS NOT FALSE".

Whether it's worth the cycles isn't too clear to me, but we could
argue about that if somebody submitted a patch.

            regards, tom lane