chore: Pycasbin benchmark

[sheny1xuan]

• Add benchmark for pycabin binding Benchmarking pycasbin's performance to compare it with Python's casbin implementation #181.
• Pycasbin binding result in my local machine, and the stddev is greater than mean, it's invalid.
  

pybind11 heavily relies on a template matching mechanism to convert parameters and return values that are constructed from STL data types such as vectors, linked lists, hash tables, etc. This even works in a recursive manner, for instance to deal with lists of hash maps of pairs of elementary and custom types, etc.
However, a fundamental limitation of this approach is that internal conversions between Python and C++ types involve a copy operation that prevents pass-by-reference semantics.

• Current pybind11 binding copy python list to a vector, it's will be inefficient. And I think wrap DataVector and DataList with python will be helpful. We can ref to https://pybind11.readthedocs.io/en/stable/advanced/cast/stl.html.
• And another thing, current casbin-cpp efficiency is smiliar with casbin-golang. Can we make it better? I think I need some help for the deep optimization. @EmperorYP7 @hsluoyz

[casbin-bot]

@EmperorYP7 @divy9881 @noob20000405 please review

[hsluoyz]

@EmperorYP7 plz review

[EmperorYP7]

Suggested change

	Pycasbin use `pytest` for benchmark.
	Pycasbin use `pytest` for benchmark.
	`pip3 install -U pytest`

[EmperorYP7]

Do this instead:

python3 -m pytest --benchmark-verbose --benchmark-columns=mean,stddev,iqr,ops,rounds casbin-cpp/pycasbin/benchmarks/benchmark_model.py

[EmperorYP7]

Does it crash often?

How about we set up a flag through an OS env variable for intensive testing?

[sheny1xuan]

It cost too much time in my machine, don't crash.

[EmperorYP7]

Current pybind11 binding copy python list to a vector, it's will be inefficient. And I think wrap DataVector and DataList with python will be helpful.

Flagging these data types as opaque might be one way to go.

And another thing, current casbin-cpp efficiency is smiliar with casbin-golang. Can we make it better? I think I need some help for the deep optimization.

Let's perform profiling on the current casbin-cpp and pycasbin setup. I suspect casbin-cpp's performance majorly depends on the performance of the underlying library used. (Exprtk)

There are scopes for micro-optimizations that I can think of. We'd need a thorough look into the code to create a significant difference in performance.

[hsluoyz]

@sheny1xuan

[ArashPartow]

@sheny1xuan in the first set of timing numbers, the stddev is greater than the mean.

This usually indicates invalid stats or the possibility of at least a bimodal (or more) distribution.

[hsluoyz]

@sheny1xuan

[sheny1xuan]

@sheny1xuan in the first set of timing numbers, the stddev is greater than the mean.

This usually indicates invalid stats or the possibility of at least a bimodal (or more) distribution.

Oh, I only paid attention to the mean, the stddev is also very important in the benchmark. And the above comparison of pycasbin and casbin-cpp binding is valid. The two set of benchmark data is all about casbin-cpp binding, I forget to change the package install environment.
And I rerun it in my local machine, the stddev is still greater than mean. Stddev is greater means the test data is unstable. I think there maybe something wrong in the code about binding or benmarking.