2

I am trying to implement a straightforward associative array (no duplicates) with string keys and instances of my own "classes" as values in JavaScript. I am happy with expected O(1) behaviour of a hash set or the O(log n) access time of a set implemented as a balanced tree.

I am drawn to the simple use of an object to which I dynamically add new properties but I am on Node.js and have come across the V8 optimisation of hidden classes (see description of V8 hidden classes here).

If I do use properties on an object, will the V8 runtime notice the sheer number of properties and their transience and stop trying to JIT hidden classes behind the scenes?

I just cloned the V8 code, so pointers into it would be welcome alongside the top-line answer to my question.

Thanks for your help.

Follow-up 1

Thanks @vyacheslav-egorov, I see the guard code below in JSObject::AddFastProperty at your link. Without digging through more code it seems like a lot of overhead on every new property insertion. I think I see something like a per-object mode which causes JSObject::AddFastProperty to never be called any more. So JSObject::AddProperty calls straight through to JSObject::AddSlowProperty without much fuss. Do I need to do anything to push an object into that mode, or will the V8 runtime switch it over reliably using its own metrics? 

if ((!name->IsSymbol() && !IsIdentifier(isolate->unicode_cache(), name)
     && name != isolate->heap()->hidden_string()) ||
    (map()->unused_property_fields() == 0 &&
     TooManyFastProperties(properties()->length(), store_mode))) {
4

1 回答 1

2

是的，当 V8 注意到对象具有太多属性时，它会将对象属性切换为字典表示。

多少是“太多”取决于几个因素（对象是如何创建的，它最初有多少属性，属性是如何添加到对象中的）。例如，如果一个对象被创建为一个空对象字面量，“太多”将是大约 30 个属性。

于 2013-04-20T20:26:04.497 回答