Stage0: hand-typed hex programs bootstrap a C compiler from a seed binary under 1KB

Stage0 attacks one of computing's hardest trust problems: how do you get a compiler when you can trust no existing compiler, firmware or operating system? Its answer is a constraint masquerading as madness — manually created hex programs whose only goal is a bootstrapping path to a C compiler capable of compiling GCC, starting from a single binary of 1 KByte or less. A Hacker News reader called it 'a wonderfully mad, cool goal.'

The design builds upward in auditable layers. A Hex0 monitor assembles hex programs typed by hand and provides minimal text input; stage one adds assemblers of growing power (Hex0, Hex1, Hex2) with comments, labels and addressing; a minimal macro assembler follows; and the first C compiler on the chain was written as a cross-compiler for x86 precisely so that everyone could verify what it did and watch it self-host. Two rules hold every step: the code must be understandable by 70% of programmers, and whether built manually or by automation, the resulting binaries MUST be identical.

The project grew out of the bootstrapping movement's worries about unverifiable toolchains — the thread that greeted it connected the work to escaping trusting-trust attacks and to preserving knowledge in forms that do not depend on today's compilers. The README credits its inspirations, from Edmund Grimley Evans's bcompiler and cc500 to Jonesforth, the PDP-1 'Expensive Typewriter', and a thought experiment about programming while stuck in a room; it also documents false starts the author kept in the repo for people who want to build the root binary independently.

The repository snapshot shows the chain reached stage three: a M2-Planet C compiler at version 1.0 with an x86 port, a dwarf-stub generator used in mescc-tools bootstrapping, and x86 tools that let builders verify programs for hardware they do not even own. Master hosting moved to GNU Savannah, POSIX ports cover x86, AMD64 and AArch64, and the stated future is more hardware ports plus implementing the Knight processor in FPGA and then TTL.