Question : performance against python dict

[pfeatherstone]

Preface:
This library strikes me as a state-of-art STL-like container mimicking python dictionaries, that, in a way, provides compile time reflection and serialisation utils in one bag. It's got to the point where this library is so good at doing type -> dict -> serialised buf -> dict -> type (amongst many other things), that even if performance was poor (which it isn't) it would be irrelevant given how usable this library is. i'll carry on for a bit. This gives python users no more excuses when they say that C++ doesn't support dictionaries and therefore can't easily abstract away data types. (for example, they may want to implement **kwargs - like functionality in C++). Well now they can. I know you shouldn't replace struct/class types in C++ with the nholmann::json type, but if you really wanted to, you could.

Question:
Now that we are in agreement that the performance of this library isn't the priority, but rather usability is, lets ask the question anyway: how does the performance of this library compare to python dictionaries?

[FrancoisChabot]

That's really not an apples-to-apples comparison.

Python dictionaries are more akin to a std::unordered_map<std::string, std::any> than a json object. The flexibility to store any object, and not just a collection of properties is an important piece of the puzzle.

[nlohmann]

@pfeatherstone Is there anything we can do to bring this issue forward?

[pfeatherstone]

I think Francois is right. It's not exactly the same as Python dict, but i thought it might be interesting to find out if it exceeded the performance of dict if used in a JSON context. It's probably not worth doing, as it would only be a passing comment in the readme and a bit irrelevant. I'll close this.

[nlohmann]

Given dict is the most optimized data structure inside Python, I only wonder how much slower this library is...

[pfeatherstone]

I don't think this is something to worry about. My gut feeling was that it would be on par performance-wise. it was a silly issue to post really since python and C++ are so different. The only thing that might be worth looking into is key lookup performance vs python dict. So how fast is it to do something like :

j["a"]["b"]["c"][...]["z"]

where j is a nlohmann::json object. But even then, it works, so who cares.