Carbon and Modernizing C++ with Chandler Carruth

What It Covers

Chandler Carruth explains Carbon, Google's experimental successor to C++, designed to provide memory safety and modern features while maintaining seamless interoperability with existing C++ codebases through advanced compiler technology and bidirectional AST translation.

Key Questions Answered

• •C++ Interop Architecture: Carbon uses Clang to parse C++ into AST, synthesizes bidirectional representations between Carbon's semantic IR and C++'s AST, then merges both into LLVM IR. This enables zero-copy data passing and supports complex features like virtual inheritance across language boundaries without runtime overhead.
• •Parameter Passing Optimization: C++ forces developers to choose between pass-by-value and const-reference for every function parameter based on type size, creating decision overhead and performance penalties. Carbon provides one unified input parameter syntax that automatically optimizes for each type, eliminating pointer indirection for small types like integers.
• •Definition-Checked Generics: Carbon implements checked generic programming that validates code at definition time rather than instantiation, enabling memory safety features similar to Rust while maintaining template compatibility for C++ interop. This reduces code duplication and compile-time costs compared to C++ template instantiation for every type combination.
• •Incremental Migration Path: Developers can automatically translate C++ code to Carbon syntax, then incrementally add constraints to templates without breaking existing code. Once constraints propagate through dependencies, templates convert to checked generics, delivering better error messages and reduced binary size while maintaining full functionality throughout migration.
• •Governance Through Painter Role: Carbon's three-lead structure includes a painter who breaks aesthetic ties only when no technical rationale exists. This mechanism forces contributors to articulate technical reasons for preferences rather than arguing aesthetics, resulting in almost zero purely aesthetic decisions while maintaining rapid decision velocity.