Re: jsonb subscripting vs SQL/JSON array accessor semantics (SQL:2023)

On Tue, Feb 10, 2026 at 5:16 AM Alexandra Wang
<alexandra(dot)wang(dot)oss(at)gmail(dot)com> wrote:
>
> The attached patches implement exactly this:
>
> 0001: Add numeric type support with truncation

Subject: [PATCH v1 1/2] Support numeric type for jsonb subscripting

Previously, using a numeric value as a jsonb subscript would error.
Now numeric subscripts are accepted and truncated toward zero to
produce an integer index. This matches the semantics of array access
in json_query() per the SQL/JSON standard.

Examples:
SELECT ('["a","b","c"]'::jsonb)[1.7]; -- returns "b" (truncates to 1)
SELECT ('["a","b","c"]'::jsonb)[-1.7]; -- returns "c" (truncates to

first thing come to my mind would be special numeric value +inf, -inf, NaN

SELECT ('{"NaN":"b"}'::jsonb)['inf'::numeric];
ERROR: cannot convert infinity to integer

SELECT ('{"NaN":"b"}'::jsonb)['NaN'::numeric];
ERROR: cannot convert NaN to integer

Is the above what we expected, or should just return NULL?
Anyway, obviously we need to test these special numeric values.

+select ('[1, "2", null]'::jsonb)[1.5::float8]; -- errors
+ERROR: subscript type double precision is not supported
+LINE 1: select ('[1, "2", null]'::jsonb)[1.5::float8];
+ ^
+HINT: jsonb subscript must be coercible to either numeric or text.

This errhint message appears to be incorrect?
given that 1.5::float8 is coercible to numeric.

--
jian
https://www.enterprisedb.com/