San Diego C++ Meetup Dec, end of the year edition - #81

Last meeting of 2025!
We will spend about 45-60 mins with our speaker and also catch up on latest news on C++ and some quiz if time permits.

• Title: Beyond the Paper: A Modular Workbench for Neural Code Analysis

1. Abstract: Implementing research papers often involves navigating a maze of hardcoded paths and fragile scripts. This presentation introduces BERT Builder, a comprehensive, "batteries-included" pipeline that transforms the theoretical work of Transformer architecture into a tangible, extensible workbench for developers and tinkerers alike. The speaker will demonstrate a complete, end-to-end tooling ecosystem that empowers users to own every step of the process. From the automated Corpus Engineering scripts that fetch and sanitize GitHub repositories, to a highly configurable CLI that allows for rapid experimentation with architectural hyperparameters, every component is designed for modification. The system's modular architecture is highlighted, which supports "plug-and-play" experimentation with distinct tokenizer strategies (Char, KeyChar, SentencePiece) and masking objectives (MLM, WWM). By bridging the gap between rigorous academic standards and DIY hackability, this toolkit provides a stable foundation for the community to explore, extend, and innovate on the intersection of NLP and Systems Programming.
2. Other talk Building Expresso: A Purely Functional, Immutable REPL in C and ANTLR This presentation explores the design and development of Expresso, a lightweight, interactive Read-Eval-Print-Loop (REPL) for evaluating expressions. Unlike traditional interpreters that rely on mutable state, Expresso is built on a strict philosophy of immutability and functional purity. Every user input is treated as a pure expression—supporting arithmetic, logical, bitwise, and string operations—that evaluates to a new, immutable value without side effects. We will examine the project's technical architecture, which leverages ANSI C17 for the core runtime and CLI, and C++17 with ANTLR for robust lexing and parsing. This hybrid approach ensures high performance while maintaining a flexible and type-safe grammar. The presentation will also detail the specification-driven development methodology used to build Expresso. We will discuss how detailed feature specifications (specs) guided a Test-First Development (TDD) process, ensuring that every operator and edge case—from complex operator precedence to numeric overflow handling—was rigorously defined and verified before implementation. Attendees will gain insights into:

• Implementing a "No-Assignment" language constraint to enforce functional purity.
• Integrating a C-based runtime with an ANTLR C++ parser backend.
• The role of "Speckit" and structured documentation in managing project complexity.
• How the modular architecture allows for future extensibility, such as "REPL-as-operand" meta-programming.

Join us to see how modern tooling and strict design constraints combined to create a clean, educational, and powerful expression evaluator.

• If time Permits
• Interesting news before quiz
• Quiz