Summary of Changes

Hello @haochengxia, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

I've addressed a bug in the read_one_req function's Python binding to ensure it correctly handles non-zero return values, which signify read failures or the end of a trace. Concurrently, I've introduced a new, comprehensive Python benchmarking script for cache simulation. This script allows for performance comparisons between native C, C++-bound Python, and pure Python loop implementations. To support these changes, I've also updated the install script to facilitate building with optional dependencies and running associated tests.

Highlights

• Bug Fix in read_one_req Python Binding: I've modified the Python binding for read_one_req in src/export_reader.cpp. Previously, it would throw a runtime error if read_one_req returned a non-zero value. Now, it simply returns the non-zero value, allowing the Python side to correctly interpret read failures or the end of the trace without an exception.
• New Cache Simulation Benchmarking Script: I've introduced a new, comprehensive Python script at benchmark/simulation.py. This script is designed to benchmark cache simulation performance across different implementations: the native C binary, the Python c_process_trace method (which leverages C++ bindings), and a pure Python loop. It provides detailed statistics on execution time, memory usage, and miss ratios, and can generate visualizations and export results to CSV.
• Enhanced Install Script with Optional Dependencies: The scripts/install.sh has been updated to include a new --all option. This enhancement allows users to build Python wheels with optional cache eviction algorithms (LRB, GLCACHE, 3L_CACHE) enabled. Additionally, when this option is used, the script will now run specific 'optional' pytest tests relevant to these new algorithms.

Code Review

This pull request addresses an issue with the trace reader iterator and introduces a comprehensive benchmark script. The change in src/export_reader.cpp correctly modifies the read_one_req wrapper to return the status code, which is necessary for the Python iterator protocol. The new benchmark script in benchmark/simulation.py is a great addition, providing detailed performance analysis and visualizations. However, I've identified a few areas for improvement in the benchmark script, including an incorrect memory measurement methodology for the native C binary, a broad exception handler, and the use of global variables. Additionally, the scripts/install.sh changes have a misleading help message for the new option and a brittle conditional check. My review provides specific feedback and suggestions to enhance the robustness and clarity of these new additions.