RapidProto documentation

RapidProto compiles a .proto schema into header-only C++ decoders: no runtime library to link and no third-party dependencies. It is decode-only - no serialization, no reflection - and generates two models from one schema: an arena decoder that materializes a message tree, and a streaming decoder that hands each field to a callback. Both validate untrusted wire input and never crash on malformed bytes. On a mixed real-world payload the arena decoder decodes about 7× as fast as protoc + Arena (benchmarks).

Start with the quick start; how it got fast walks the optimizations one at a time. The pages below are the topic-by-topic reference.

• arena.md - the arena decoder: accessors, the Arena, decode_owned, error handling
• streaming.md - the streaming decoder: field tags, the consumption patterns, aborting
• dumper.md - the --dump debug dumper: JSON-like inspection text, DumpOptions
• semantics.md - the shared rules: lifetimes, validation & trust, presence, open enums, duplicate fields
• using-both-models.md - both models in one TU; coexisting with protoc
• profiles.md - decode profiles (drop / raw) and unknown-field detection (arena)
• integration.md - the rapidprotoc CLI reference and the CMake helper
• osm-pbf.md - real-world walkthrough: OSM’s planet format with both models
• optimizations.md - how the decoders got fast: seven optimizations, re-enabled one at a time and measured
• benchmarks.md - the numbers and how to reproduce them

Contributor docs live at the repository root: architecture.md (internals, invariants, design rationale), CONTRIBUTING.md, SECURITY.md, and CHANGELOG.md.