Programming error when the first JSONField path segment is an integer overflowing int4's range on PostgreSQL

[Clifford Gama]

JSONField key lookup raises an unhandled ProgrammingError ("operator does not exist: jsonb -> bigint") instead of behaving like any other out-of-range/non-matching lookup, when a purely-numeric key segment decimal-parses to a value outside PostgreSQL's int4 range.

I had a JSONField whose keys are generated using uuid.uuid4().hex[:12]. When a key was all-digit, this ran into #29504, leading to flaky tests when I did something like .filter(**{key: "some value"}). That's because Django ​interprets an integer key segment as an array index rather than a text path -- but only when it's the lookup's sole segment. In this case, though, instead of getting no match, I got a ProgrammingError: the number overflowed int, and -> isn't overloaded for bigint.

operator does not exist: jsonb -> bigint
HINT:  No operator matches the given name and argument types. You
       might need to add explicit type casts.

Since Django is guessing if an all-number string is an array index, I suggest that we treat numbers outside of the allowed range as strings. An appropriate solution would be to make this match the same way as an out of range index, which means returning NULL.

See attached PR for a regression test.

[Clifford Gama]

​PR

[Clifford Gama]

Now that I think about it, returning no match would be the consistent with the behaviour of the other database backends.

[Jacob Walls]

Thanks for the report. Returning NULL seems to agree with the behavior for other missing keys/indices.

[Jacob Walls <jacobtylerwalls@…>]

In 66c40aa:

Fixed #37297 -- Prevented crash on JSONField key lookups when an integer key is outside PostgreSQL's int4 range.

[Jacob Walls <jacobtylerwalls@…>]

In 7f48e26:

Refs #37297 -- Prevented crash on JSONField isnull lookups on Oracle when a key is outside its integer range.

Oracle's JSON_EXISTS() rejects an array-subscript literal this long
with ORA-40598/JZN-00211.