Small, Hackable, Safety-Oriented Compiler
A small experimental C compiler for memory, lifetime, ownership, and thread safety.
Safe C explores a single question:
How much of Rust's memory, lifetime, ownership, and thread safety can be brought into the C ecosystem while preserving C's familiarity, interoperability, and incremental migration path?
This is not a Rust clone, and not just another tiny C compiler. It aims to be a small, understandable, hackable reference implementation for experimenting with safety mechanisms in the C language family.
Smallness is a means, not the goal: the compiler stays small enough that one developer can hold its major components in their head, which is what makes experimentation practical.
- Safety IR is the central abstraction. The AST is never lowered directly to LLVM IR. Safety analysis operates on a dedicated semantic IR that can express ownership, lifetimes, borrowing, regions, and thread relationships.
- Inference is separated from enforcement. Results are
Safe,Unsafe, orUnknown. The analyzer does not pretend every C program can be proven safe. - Gradual adoption. Existing C code can enter the system without a rewrite.
- Hackability over abstraction purity. The architecture optimizes for making experiments easy.
C source
|
v
Lexer -> Parser -> AST -> Semantic analysis -> Typed AST
|
v
Safety IR <-- central boundary
|
+--> Safety analysis (memory / lifetime / ownership / thread)
|
+--> Backend (LLVM IR, interpreter)
Longer term, a Clang adapter feeds the same Safety IR, so the safety model can be validated against real-world C and C++ code.
docs/ holds what the project wants to build: the concept,
the safety model, the architecture,
and the roadmap. docs/adr/ holds why each
cross-cutting decision was taken, and what would reverse it.
cargo build
cargo run -p safecRequires a Rust toolchain with edition 2024 support (1.85 or later).
Licensed under either of
- Apache License, Version 2.0 (LICENSE-APACHE)
- MIT license (LICENSE-MIT)
at your option.