> +     ;; Special-case UTF-16 surrogate pairs,
> +     ;; cf. https://tools.ietf.org/html/rfc7159#section-7.  Note that
> +     ;; this clause overlaps with the next one and therefore has to
> +     ;; come first.
> +     ((looking-at
> +       (rx (group (any "Dd") (any "89ABab") (= 2 (any "0-9A-Fa-f")))
> +           "\\u" (group (any "Dd") (any "C-Fc-f") (= 2 (any "0-9A-Fa-f")))))
> +      (json-advance 10)
> +      (json--decode-utf-16-surrogates
> +       (string-to-number (match-string 1) 16)
> +       (string-to-number (match-string 2) 16)))
>       ((looking-at "[0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f][0-9A-Fa-f]")

Wouldn't it be better to move the surrogate-handling into the
hex-handling branch?