I hereby agree to the terms of the CLA available at: https://docs.databend.com/dev/policies/cla/

Motivation

Fix unnecessary block reads for OR predicates on nullable columns. A comparison may be known to return only FALSE or NULL; neither passes WHERE. If every OR branch has that property, the whole filter cannot match.

The old constant folder still marks TRUE as possible, so range/bloom pruning keeps blocks that contain no matches. Query results remain correct, but pruning is lost.

Reproduction

Use default settings in a fresh test schema. No SET or special table option is needed.

CREATE TABLE bloom_test_nullable_or_t(c1 VARCHAR NULL, c2 VARCHAR NULL, c3 INT NULL);
INSERT INTO bloom_test_nullable_or_t VALUES
  ('a1','b1',1), (NULL,'b2',2), ('a3',NULL,NULL);
INSERT INTO bloom_test_nullable_or_t VALUES
  ('a4','b4',4), (NULL,NULL,5), ('a6','b6',NULL);

EXPLAIN SELECT * FROM bloom_test_nullable_or_t
WHERE c1 = 'a2' OR c2 = 'b5';

Each INSERT creates one block. Both searched strings are absent, but lie inside the min/max ranges, so bloom membership could exclude the blocks.

The SELECT returns no rows in both versions. The difference is that the fix avoids planning reads of both blocks.

Approach

Start domain combination with FALSE for OR and TRUE for AND, rather than with both values possible. Argument handling and ordinary SQL NULL semantics are unchanged.

Tests

• SQLLogicTest

bloom_filter.test covers impossible OR/IN predicates, nested AND, range pruning, matching branches that must retain a block, and actual returned rows where one OR branch is NULL and the other is TRUE. Comments explain the two-block layout and each expected pruning result. Rust test files are unchanged from main; the SQL coverage is not duplicated there.

The expected result checks retain NULL OR TRUE and TRUE OR NULL rows, while excluding a row where both branches are NULL. On the old implementation, the nullable-OR EXPLAIN reports the 6 rows / 2 scanned blocks above instead of zero.

Type of change

• Bug Fix (non-breaking change which fixes an issue)

AI assistance

• AI usage: AI coding agents investigated the domain fold, drafted the minimal initializer change and synthetic regressions, and assisted validation and privacy review.
• Responsible human: @dantengsky
• The responsible human has read every line of this diff and can explain each change

This change is