tomorrowdawn / gitcgsimulator Goto Github PK

因回调函数中需要模拟运行回调后的游戏状态，所以需要能够保存当前所有未运行的事件，因此必须使用事件队列形式。目前考虑手动维护事件栈，并对回调函数接口进行调整，添加事件队列快照参数。预期修改文件范围为EventManage.py和GameInstance.py，示例搜索算法中也将作出相应调整。

计划设计一个passive方法，游戏开始时或者角色复活时会调用该方法。该方法应当返回一个事件列表，事件列表会如同正常事件一样被执行。

另外可能需要设计一下若复活应当修改哪些部分。目前似乎只有Aura（需要将角色附着从died变成empty）受到影响。

Now each action takes nearly 200us (0.2ms) to execute, which is extremely slow. Profiling the code with cprofiler reveals a high computational cost during event handling. Therefore, there are two potential approaches for future work:

Improve the Python code to enhance performance. While this approach requires less coding, it may not significantly improve performance since the event queue is necessary, and further complexity reduction may be limited.

Rewrite the code in another language, such as C++. This approach would require more time and effort for coding, but it would allow us to benefit from the efficient nature of C++. By doing so, we can achieve much better performance (on average, C++ is 20x faster than Python in terms of execution cycles). And this would enable us to train a more powerful AI.

It's good if we could implement an engine in C++, but it's too much for single developer, because a lot of coding(like reflection, double dispatch,......) should be made to simulate python function.