JSON_OVERLAPS returns 0 for a pair that does overlap when the document breaks later, and cannot report failure at all

JSON_OVERLAPS returns a boolean and has no way to say it could not read
its arguments, so a failed read is returned as FALSE. Where the two
documents do overlap, that is a wrong answer.

SELECT JSON_OVERLAPS('{"a":1, "b":2 "c":3}', '{"a":1}') AS v;

v

0

Warnings:

Warning  4038  Syntax error in JSON text in argument 1

              to function 'json_overlaps' at position 15

"a":1 is common to both arguments, so the documented answer is true.
The 0 is a genuine FALSE and not a NULL:

SELECT JSON_OVERLAPS('{"a":1, "b":2 "c":3}', '{"a":1}') IS NULL AS is_it_null;

is_it_null

0

Every sibling answers NULL on a failed read. On the same document,
JSON_CONTAINS and JSON_EQUALS both return NULL with Warning 4038 at
position 15. For JSON_OVERLAPS alone, "they do not overlap" and "I
could not read that" are the same value.

How to repeat

-- an overlap that exists, reported as no overlap

SELECT JSON_OVERLAPS('{"a":1, "b":2 "c":3}', '{"a":1}') AS wrong_answer;

0

-- and the other direction: a break that is never reported at all

-- 1, silent

SELECT JSON_OVERLAPS('[1,2,3 4]', '[3]')                     AS v;

-- 1, silent

SELECT JSON_OVERLAPS('[{"a":1},{"b":2 "c":3}]', '[{"a":1}]') AS v;

Root cause

The `error:` label in `Item_func_json_overlaps::val_bool()` returns 0
without setting `null_value`, so the failure is indistinguishable from a
computed FALSE.

Contract, `scripts/fill_help_tables.sql:455`: *"returns true if they
have at least one common key-value pair between two objects, array
element common between two arrays, ..."*

Not this bug

SET @d32 = CONCAT(REPEAT('[', 32), '1', REPEAT(']', 32));

SELECT JSON_OVERLAPS(@d32, @d32);

crashes the server with signal 11 on 10.11. That is MDEV-40590, which
is already filed and fixed, and is not part of this report.

Affected versions

JSON_OVERLAPS was added in 10.9.1.

Verified on 10.11 only (10.11.19-MariaDB-debug, commit 1dab253482d).
Other branches were not tested.

Note

Found while auditing the JSON functions against their documented
contracts. Partial and it pins the defect:
func_json_invalid.result:262 records JSON_OVERLAPS('[1,','[1]') = 1.
The overlap-exists-but-the-answer-is-0 case is covered nowhere.