A C++ toolchain from 357 bytes, in Bazel

Stage0 is defined as a chain of compilers and assemblers that can be built from source, beginning with a 357-byte seed program. Once that seed yields a recent GCC, the resulting compiler can be used to build any C or C++ program. This mirrors the strategy adopted by NixOS and other distributions to shrink their binary seed and simplify porting to new architectures.

The author previously attempted to create a stage0-based hermetic C++ toolchain for Bazel in late 2024, making substantial progress but stopping short of usability. By leveraging an LLM to finish the remaining work, the toolchain now supports Bazel’s cc_binary rule and can compile external dependencies directly from the Bazel Central Registry. The setup involves declaring the stage0-bazel dependency and registering the clang and cc toolchains provided by the repository.

To confirm hermeticity, the repository includes an audit report generated via Bazel aspects. The report checks every action in the build graph and fails if any action runs a program not produced by the repository itself. The output shows that all actions are internal except for a few audited seed binaries, namely the hex0 seed and a bash binary used by genrule, which remains an external input because the shell path is not declared as a toolchain input.

Although companies like Google have not made bootstrap seeds a priority, given their control over the entire build environment, the increasing adoption of Bazel and similar build systems in the open-source world creates a need for reproducible, auditable toolchains. This work shows that a minimal seed can scale to a functional C++ toolchain, offering a concrete step toward more transparent and portable builds for the broader community.