如何修改Prolog的json_get谓词以支持数组索引查询？

Great question! Your current json_get works perfectly for grabbing top-level fields from JSON objects, but it doesn't know how to handle JSON arrays or nested paths (like digging into an array that's a value of an object field). Let's fix that by adding support for array indices and recursive path traversal.

First, let's rewrite the predicate to handle three key scenarios:

1. Fetching a field from a JSON object
2. Fetching an element from a JSON array by index
3. Recursively traversing a nested path of fields and indices

Here's the modified code:

% Grab a value from a JSON object using an atom field name
json_get(json_obj(JSON_obj), Field, Result) :-
    atom(Field),
    memberchk((Field, Result), JSON_obj).

% Grab an element from a JSON array using a 0-based integer index
json_get(json_array(JSON_arr), Index, Result) :-
    integer(Index),
    nth0(Index, JSON_arr, Result).

% Recursively walk through a nested path (mix of fields and indices)
json_get(JSON, [Step|Rest], Result) :-
    json_get(JSON, Step, Intermediate),
    json_get(Intermediate, Rest, Result).

% Optional: Empty path returns the original JSON (useful for edge cases)
json_get(JSON, [], JSON).

What Changed? #

• Array Support: We added a new clause that checks if we're dealing with a json_array and an integer index. We used nth0/3 for 0-based indexing (so index 0 gets the first element, 1 gets the second, etc.). If you prefer 1-based indexing (where 1 is the first element), just swap nth0 with nth1.
• Nested Paths: The recursive clause lets you pass a list of steps (like ["heads", 1]) to dig into nested structures. First it grabs the heads array from the object, then pulls the element at index 1 from that array.
• Avoid Ambiguity: We added checks like atom(Field) and integer(Index) to make sure the predicate knows whether it's dealing with an object field or an array index.

Testing It Out #

Let's run your example query:

json_parse('{"nome" : "Zaphod", "heads" : ["Head1", "Head2"]}', Z),
json_get(Z, ["heads", 1], R).

This should return R = "Head2" (since we're using 0-based indexing). If you wanted Head1 instead, just use index 0.

A quick note about your original working query: the behavior shifts a little with the modified code. Originally, json_get(O, ["nome"], R) would return R = ["Arthur"] (a list of results from maplist). Now it returns R = "Arthur" directly (since it treats ["nome"] as a nested path with one step). If you still want to grab multiple top-level fields as a list, just use maplist(json_get(O), ["nome", "cognome"], R) directly—it'll give you R = ["Arthur", "Dent"] just like before.

Bonus: Keep Both Behaviors #

If you want to retain the original ability to fetch multiple top-level fields with a list while adding nested path support, add this extra clause:

% Handle lists of top-level fields (original behavior)
json_get(JSON_obj, Fields, Results) :-
    is_list(Fields),
    forall(member(F, Fields), atom(F)), % Make sure all elements are field names
    maplist(json_get(JSON_obj), Fields, Results).

Now json_get(O, ["nome", "cognome"], R) returns R = ["Arthur", "Dent"], and json_get(Z, ["heads", 1], R) still gives you R = "Head2"—best of both worlds!

Content of the question originates from Stack Exchange, question author Giuseppe Giubaldo