optimize use int id

[asukaminato0721]

now in this engine, it use

str id -> node

since node_id is unique, we can give each str an id

then just nodes[int_id]

array visit is much faster than str hash.

this is called arena.

related context: https://www.fxzhihu.com/question/36756224/answer/3575810103

[WH-2099]

Given the complexity here, we'll treat this as a long-term optimization rather than a priority for the immediate future.

[QuantumGhost]

I highly suspect the str hash part would be the bottleneck. If you have a related profiling, please paste it in the issue.

[WH-2099]

https://bugs.python.org/issue27350

I actually don't think this is a bad proposal.

To be clear, I don't believe this is intended for performance in the conventional sense of access speed. The implementation proposed here is actually very similar to the Compact Dict introduced in Python 3.7.

In our specific scenario, I think the primary advantage is that it preserves the insertion order of the nodes. Admittedly, I am not entirely sure what the exact practical impact of that would be at this stage, but from a data structure perspective, it isn't a bad idea. We can certainly keep it on the radar for the long term.