Object's member access is O(n) about rapidjson HOT 6 CLOSED

[deleted comment]

I don't think it is possible to use a faster algorithm without ordering the 
keys.

Ordering the keys and putting them in a map also has a overhead so it is does 
not make sense to implement it.

Users should access a value once, then store it and re-use the value without 
calling Document::operator []. That will be faster than anything.

With C++11 it is possible to use std::unordered_map see e.g. 
http://en.cppreference.com/w/cpp/container/unordered_map

Just my opinion…

AFAIK, hash-map overhead is known to be a lot heavier than sorted B-trees if 
key count is small. And JSON object usually expected to have a few of keys - 
less than 10 keys for example - because large number of keys means there's 
something wrong with data structure design.

So I doubt on using hash-map for performance. If your object has dozens of 
keys, it would make sense, but I never seen such JSON object.

IMO, binary search is the best approach for a JSON object, and C++11 std::map 
is usually intended to be a b-tree.

Added an analysis on this issue.
https://github.com/miloyip/rapidjson/issues/102