The meeting discussed updates on the graph library, specifically Graph v3, which uses descriptors to improve abstraction and eliminate memory references. Key features include vertex and edge descriptors, forward iterators, and support for non-integral vertex IDs. The team also discussed the need for a data frame and CSV file reading capabilities to enhance C++'s AI infrastructure. Michael Wong presented a strategic document on C++'s role in AI, emphasizing the importance of governance, ecosystem improvement, and strategic priorities like data frames and MD arrays. The team agreed to address feedback on special value detection and insufficient values in future proposals.

Action Items

• MW Write and submit the 'C++ directions for AI' paper for the directions group, prepare it for the February mailing (deadline ~Feb 23), and share it with the group for feedback before publication
• PR Implement the remaining graph algorithms (level propagation, articulation points, clustering coefficient, and any other outstanding algorithms), roughly ~1 hour per algorithm, then submit for group review
• PR Run benchmarks comparing graph v3 results to graph v2 and Boost Graph (performance and correctness) and publish the benchmark results in the repository for review

Outline

Introduction to Graph Library v3

• PR explains the background of the graph library, mentioning its evolution from Boost Graph based on C++98 to a modern graph library based on C++21.
• The new version, Graph v3, uses descriptors to improve abstraction and eliminate the need for memory references.
• PR details the changes in Graph v3, including the use of descriptors for vertices and edges, and the benefits of this approach.
• The discussion includes the implementation of forward iterators and the impact on the library's design and algorithms.

Detailed Explanation of Graph v3 Features

• PR provides a detailed explanation of the descriptor record for vertices, including storage types and member functions.
• The discussion covers the edge descriptor, which includes the source vertex descriptor and the significance of this feature.
• PR mentions the use of agile coding to accelerate development and the minimal refactoring required for the new version.
• The improvements in the library, including the reduction of concepts in the undirected adjacency list, are highlighted.

Future Plans and Documentation

• PR outlines the remaining work, including implementing additional algorithms and updating the documentation.
• The importance of having the documentation ready for the next meeting is emphasized.
• Michael Wong discusses the need for a public repository and the potential for benchmarking results.
• The conversation shifts to the use of GitHub Copilot for coding assistance.

Discussion on AI and C++ Standards

• Michael Wong introduces a paper on C++ directions for AI, focusing on governance, ecosystem improvement, and strategic priorities.
• The paper addresses the use of AI in generating normative wordings and the importance of human authors taking responsibility for the accuracy and logic of their work.
• The need for a curated, high-quality dataset of modern idiomatic C++ code for AI tooling is discussed.
• The conversation includes the potential for a call to action for the C++ community to create this dataset.

Strategic Priorities for C++ and AI

• Michael Wong outlines the strategic priorities for C++, including the need for a data frame and the ability to read CSV files.
• The discussion includes the importance of MD array for homogeneous multi-dimensional numeric data and its role in AI workflows.
• The potential for using compilers and modules to provide modern usage information to AI tools is explored.
• The conversation highlights the need for a comprehensive approach to support AI in C++, including data ingestion, generation, and infrastructure.

Feedback and Next Steps

• Speaker 1 and Michael Wong discuss the feedback received on the paper, including suggestions for branchless special value detection and handling insufficient values.
• The importance of consistency in error handling and the potential for future enhancements is emphasized.
• The conversation includes the need for a separate proposal for branchless optimizations and the potential for integrating these enhancements in future versions.