Subscript performance in JSONObject

Question

Subscript performance in JSONObject

ivan-ushakov opened this issue 6 years ago · 7 comments

I faced with strange performance problem. When I use JSONObject directly with Data object, subscript access takes much more time than parsing.

4.16 s      45.0%   24.00 ms            JSONObject.subscript.getter
1.43 s      15.4%   5.00 ms             JSONObject.init(data:)
463.00 ms    5.0%   7.00 ms             JSONArrayIterator.next()

This is for ~ one million calls to JSONArray(data: ) and some hand transformation of returned object to structure. Looks like parsing is much cheaper than access to object field by name.

Answer 1 · 2019-05-13T18:31:25.000Z

Is there anything you can share about the data structure? It is a particularly big array or object, for example?

Answer 2 · 2019-05-14T21:15:03.000Z

This is a file with ~ 1 million JSON objects. I read each object to array and pass this array to JSONObject class. After that I work with object properties. Each object looks like this:

[
    {"company":"Рога и копыта",         "debt": 800,    "phones": [123, 234, 456]},
    {"company":"Первая коллекторская",  "debt": 1200,   "phones": [2128506, 456, 789]},
    {"company":"Святой престол",        "debt": "666",  "phones": 666},
    {"company": "Казачий спас",         "debt": 1500,   "phones": [234567, "345678"], "phone": 666},
    {"company": {"name": "Шестерочка"}, "debt": 2550,   "phones": 788, "phone": 789},
    {"company":"Рога и копыта",         "debt": 800,    "phones": [123, 234, 456]},
    {"company":"Первая коллекторская",  "debt": 1200,   "phones": [2128506, 456, 789]},
    {"company":"Святой престол",        "debt": "666",  "phones": 666},
    {"company": "Казачий спас",         "debt": 1500,   "phones": [234567, "345678"], "phone": 666}
]

As you see phones attribute could be Array or String, and Array could contain String or Number. This is the reason why I need to use subscript getter and as? operator.

In my current test it looks like getting object property by name takes more than parsing object.

Answer 3 · 2019-05-15T11:06:44.000Z

Thanks! I'll have a stab at this by generating a JSON for myself to test this upon. For clarity, what are the three numbers? 4.16s is the total time spent in that function I assume. And 24ms would be the average execution time of that function?

Answer 4 · 2019-05-15T14:30:30.000Z

Yes, you are right.

Answer 5 · 2019-05-16T04:11:10.000Z

Short update; Swift is adding a tonne of extra retains and releases in that code. These accumulate to be at least above 60% of the performance cost of a subscript. Thanks for highlighting this for me :)

Answer 6 · 2019-05-20T15:34:37.000Z

Technically everything you would need is present in the JSONDescription, but I'm struggling to create a good public API for it. If I could help you utilize the JSONDescription, it would allow you to read/search the JSON DataSet more directly, not using the JSONObject and JSONArray types.

In that case you'd have no overhead from the iterator and subscripts.

Answer 7 · 2022-01-28T14:29:59.000Z

If anyone has specific APIs they want for improved parsing performance, create a new issue.