Google Summer of Code 2021 Project Log

Report.md

Google Summer of Code 2021 Project Log

Introduction

• Name: Yash Pandey (@EmperorYP7)
• Organisation: Casbin
• Project link: https://summerofcode.withgoogle.com/projects/#6057909131149312
• Repository link: https://github.com/casbin/casbin-cpp
• Project Name: Benchmarking, Python bindings and Migrating to CTest for Casbin-CPP
• Project Summary: Casbin is an authorization library that extends its features to implement Access Control Lists, Role-Based Access Control, and Attribute-Based Access Control models in various programming languages to its clients. Casbin's Core Engine is written in GoLang. Casbin-CPP has obvious benefits of speed and efficiency compared to its implementation in other languages and thus, benchmarking is vital for the project to stand out from the rest. Python is the most versatile as well as the most used programming language and has huge community support. Casbin-CPP has the potential to support new PyCasbin to compound the benefits of both languages through language bindings and extension libraries. Currently, the project uses Microsoft Unit Testing Framework for C++ for testing and Microsoft’s Azure DevOps pipelines for CI. CTest is truly cross-platform and can be configured using GitHub Actions for consistent and better CI. Project Ideas are as follows:
  • Modernizing the project with C++17 standard and Google’s benchmarking tool
  • Implement Python bindings for Casbin-CPP using pybind11 library
  • Implement testing based on CTest and set up workflows for Continuous Integration through GitHub Actions

---

Contributions

Pre-GSoC

10/03/2021 - 13/04/2021 *

I've been contributing to Casbin-CPP prior to GSoC period. This is when I got introduced to the project. I tackled some good first issues and learnt about the codebase. Major contributions include the implementation of SyncedEnforcer along with the unit tests. Here are my contributions:

#	Title	Status
casbin-cpp#82	feat: Replaced pragma with include guards
casbin-cpp#84	Python bindings to facilitate PyCasbin-on-CPP
casbin-cpp#90	feat: Added async ticker
casbin-cpp#92	feat: Replaced travis with GitHub Actions
casbin-cpp#93	feat: Added translation unit for synced enforcer
casbin-cpp#94	feat: Added Tests for SyncedEnforcer
casbin-cpp#96	Upgrade to C++17
casbin-cpp#98	fix: Build on MS Visual Studio

Community Bonding Period

17/05/2021 - 06/06/2021 *

I utilized this phase to plan out and discuss about my projects with the mentors and also work on the plan ahead. I studied about pybind11 and various ways people implemented it in their projects. I studied about mosquitto, Glewlwyd, and envoy proxy to work out ways to integrate casbin with these projects as per casbin-cpp#100, casbin-cpp#79, and casbin-cpp#50. I planned to work on ABAC support for the project. I tried to clean up the project and get it ready for the bindings and C++17 standardization.

Here's my contributions during this period:

#	Title	Status
casdoor#65	feat: Added captcha
casbin-cpp#99	Configure CMake build system 🏗️
casbin-cpp#101	feat: Initiate CMake configuration
casbin-cpp#103	feat: Completion of Enforcer API
casbin-cpp#104	fix: Taking in parameters by const reference

Coding Period (07/06/2021 - Present *)

Week 0 (June 8 - June 15)

I tried to work on integrating Casbin-CPP with Glewlwyd. It required a bit of planning since Casbin-CPP used C++17 and Glewlwyd, C. The discussion can be found in casbin-cpp#100. The idea was to create a middleware which would let casbin operate on the config file and users given by Glewlwyd.

Meanwhile, I added ABAC support by creating object wrapper using C++17 features. There was some overhead involved with the approach I proposed. I am currently working on getting around it.

I also worked on configuring CTests for the project and changed the testing framework from MS Unit Testing to googletest. This includes setting up CI to test the builds on every platform.

Here are my contributions:

#	Title	Status
casbin-cpp#105	feat: Added ABAC entity wrapper
casbin-cpp#106	feat: Added implementation and tests for UpdatePolicy API

Week 1 (June 15 - June 22)

I successfully shifted the testing framework of casbin-cpp from MS Unit testing to Google Test through CTest. This includes configuring CI for build and test through GitHub Actions. The work is ready to be merged to master.

I discovered that Google Test does not provide thread safe tests and is not configured to handle multithreaded/distributed workflows which prevents us from testing casbin::SyncedEnforcer. What can be done instead is to join the threads before assertions.

To inculcate ABAC features into the existing project, I'll be using std::variant to store std::string or a shared pointer to casbin::ABACData object. This means a refactoring of the internal API and inspecting ways to integrate this within casbin::Scope through duktape.

Here are my contributions:

#	Title	Status
casbin-cpp#107	feat: Initiated CTest setup
casbin-cpp#108	test: Management API, Utility methods and more
casbin-cpp#109	test: Added RBAC, Role Manager and removed MS Unit Tests
casbin-cpp#110	feat: Implemented ABAC in Internal API

Discussions:

#	Title
casbin-cpp#102	can casbin-cpp support ABAC?
casbin-cpp#100	Integrate with Glewlwyd

Plan of action

In the upcoming week, I'll be:

• Investigating ways and implementing Code Coverage for the project
• Start configuration for Python bindings
• Find out more about Glewlwyd and integrating the project

Week 2 (June 22 - June 29)

This week marked the beginning of benchmarking for casbin-cpp. This includes configuring build system for finding Google benchmark tool and fetch it from the web if needed. This means that the project doesn't need the user to setup/install any dependency explicitly. Also, a CI workflow for benchmarking was set up.

Since the build system was shifted to CMake, this called for updated documentations.

I continued working on CTest and added all the tests implemented in previous MS Unit Tests. I also tried to come up with an approach to implement ABAC in the internal API. I need a bit of reference from GoLang's implementation for the same.

Here are the PRs I opened this week:

#	Title	Status
casbin-cpp#112	feat: Inherit std::exception
casbin-cpp#113	chore: Updated docs for build instructions
casbin-cpp#114	feat: Initiated benchmark workflow

Here are the ones I continued working on this week:

#	Title	Status
casbin-cpp#108	test: Management API, Utility methods and more (Continued)
casbin-cpp#109	test: Added RBAC, Role Manager and removed MS Unit Tests (Continued)
casbin-cpp#110	feat: Implemented ABAC in Internal API (Continued)

Here are the discussions I took part into:

#	Title
casbin-cpp#41	Setup code coverage badge
casbin-cpp#111	Exceptions should inherit std::exception
casbin-cpp#102	can casbin-cpp support ABAC?

Plan of Action

In the upcoming week I plan to:

• Complete benchmarking all functions within the library
• Setup code coverage for the project
• Implement ABAC in the internal API
• Cleaning up the code

Week 3 (June 29 - July 6)

I utilized this week to inspect about expression evaluators used in casbin-cpp as per the conversation in casbin-cpp#110. This will take me a while to analyse and implement ABAC in casbin-cpp. I'd also include casbin::ABACData abstraction into the internal and management APIs. Evaluators also need to recognise and parse sub_rules within policy and model files.

I benchmarked all the functions which are on par with GoLang's casbin. This revealed that there was a significant memory leak in the project. The benchmarks also bottlenecked at specific functions including Enforcer::AddPolicies() and Enforcer::AddGroupingPolicies(). The library performed significantly slower with RBAC workflows than GoLang's casbin.

I also tried to incorporate visual benchmarking within the project. However, it is not intended to be merged in casbin-cpp:master. Here's the commit: https://github.com/EmperorYP7/casbin-cpp/commit/e9923bf37a03eebc03655ccf601bc900488f724c

Here are the PRs I opened this week:

#	Title	Status
casbin-cpp#116	test: Benchmarks for CachedEnforcer
casbin-cpp#118	test: Benchmarks for ManagementAPI
casbin-cpp#120	test: Benchmarks for Model and RoleManager

Here are the issues I opened:

#	Title	Status
casbin-cpp#117	Error while using casbin::Enforcer::AddPolicies()
casbin-cpp#119	Memory leakage while using Enforcer::AddPolicy and RBAC workflows

Plan of Action

Next week I'll try to:

• Work on ABAC support for the project
• Set up code-coverage (This will require 3rd party apps from GitHub marketplace)
• Clean up code and fix memory leaks
• Improve the performance of the library.

Week 4 (July 6 - July 13)

This week I made modifications to the project to accomodate ABAC workflow into the API. My approach was to declare datatypes intrinsic to casbin. The datatypes declared are as follows:

• AttributeValue: The type of a value of an attribute
• Attribute: The type of an attribute containing the name and the value as a std::pair.
• AttributeVector: Vector of attributes.
• AttributeList: Initializer list of attributes.
• AttributeMap: Unordered map containing name and value of the attributes.
• Data: Type of Data for accomodating ABACData
• DataVector: Vector of casbin::Data
• DataList: Initializer list of casbin::Data
• DataMap: unordered map containing name as std::string, and casbin::Data

In case the data is an ABACData object, the API generates a casbin::Scope object named as per the corresponding token key and feeds in the attributes within that object.

Understanding intrinsic types might be confusing for the user/client at first. I made sure that the type names are intuitive and can easily be grasped by anyone having remote familiarity with ABAC.

I also invested some time in cleaning up the project, i.e. reducing the overall memory footprint by minimizing the number of copies and initializations through const reference, references, initializer lists etc.

Here are the PRs I opened this week:

#	Title	Status
casbin-cpp#121	chore: Cleanup
casbin-cpp#122	feat: Streamlined types

Here are the PRs I followed up on this week:

#	Title	Status
casbin-cpp#110	feat: Implemented ABAC in Internal API (continued)
casbin-cpp#116	test: Benchmarks for CachedEnforcer (continued)
casbin-cpp#118	test: Benchmarks for ManagementAPI (continued)
casbin-cpp#120	test: Benchmarks for Model and RoleManager (continued)

Discussions I took part in:

#	Title	Status
casbin-cpp#102	can casbin-cpp support ABAC and eval()?

Plan of Action

The next week I'll try to:

• Come up with an approach to implement eval()
• Initiate workflow for Python bindings
• Release the benchmark results on the official site

Week 5 (July 13 - July 20)

This week marked the beginning of python bindings. I configured the build system so that it finds pybind11 if it is installed within PATH. If it doesn't find it there, CMake fetches the zip file from Github and place the contents into build/_deps directory. This means that the project doesn't require any special dependency to be installed. I also added a mock python binding which compiled sucessfully.

However, I immediately faced a problem. Pybind11's build configuration enforces it to build with C++11, but the project is made in C++17. I tried to research about it and stumbled upon this documentation section - STL containers in pybind11. I'll try to work upon it in the upcoming week.

I also added the results of the benchmarks in casbin-cpp#118 to casbin's website. There was a discussion as to how the testbed should be selected to showcase the benchmarks. I used Apple M1 as the testbed which apparantly, isn't a popular server environemnt. Here is the discussion casbin-website#255.

The benchmarks revealed that the enforcer is a lot slower than expected. I am investigating into what is the cause of such an underperformance. Reference.

Here are the PRs I opened this week:

#	Title	Status
casbin-cpp#125	chore: Model Cleanup
casbin-website#256	feat: Added benchmarks for Casbin-CPP
casbin-cpp#123	feat: Initiated pybind11 Configuration
casbin-cpp#124	chore: Util cleanup

These are the discussions I participated in:

#	Title	Status
casbin-website#255	Add the C++ result to our benchmark page (between Go and Lua tabs)
casbin-cpp#120	test: Benchmarks for Model and RoleManager

Plan of Action

The next week I'll try to:

• Develop python bindings for the casbin API
• Fix and clean up casbin::Enforcer and the API for better performance
• Release the revised benchmark results on the official site

* Dates in DD/MM/YYYY format