bootstrappable.org project -*- org -*- * What? ** Full source bootstrapping for GuixSD * Why? ** Reproducibility is essential to Software Freedom Reproducible builds are a set of software development practices that create a verifiable path from human readable source code to the binary code used by computers. *** What about the compiler? We have the sources: they always lead to bitwise-same binary, but what about the compiler? *** The current way out: Ignore the problem ``recipe for yoghurt: add yoghurt to milk'' *** From the GuixSD manual The distribution is fully “bootstrapped” and “self-contained”: each package is built based solely on other packages in the distribution. The root of this dependency graph is a small set of “bootstrap binaries”, provided by the ‘(gnu packages bootstrap)’ module. For more information on bootstrapping, *note Bootstrapping::. *** New solution: Full source bootstrapping path, Stage0 and Mes * How? ** Software: Stage0 and Mes ** Stage0 *** hex.0: amazing ~300 byte self-hosting hex assembler that we consider to be source *** a M0 macro assembler written in .0 *** a M1 macro assembler written in M0 *** a hex2 linker written in M0 Look: https://git.savannah.nongnu.org/cgit/stage0.git/tree/Linux%20Bootstrap/hex0.hex Do: git clone https://git.savannah.nongnu.org/git/stage0.git cd stage0 make bin/hex < Linux\ Bootstrap/hex0.hex > hex-1 chmod +x hex-1 # later: bin/exec_enable hex-1 ./hex-1 < Linux\ Bootstrap/hex0.hex > hex-2 ** Mes https://gitlab.com/janneke/mes *** mes.c: a scheme interpreter prototyped in C ~1400 Lines *** mescc.scm: a C compiler written in Scheme (uses Nyacc C99 parser in Scheme) *** mes.M1: this scheme interpreter in annotated M1 assembly ** Naive recipe hex + hex.0 => hex-1 hex-1 + M0.0 -> M0 M0 + M1.M0 -> M1.0 M0 + hex2_linker.M0 -> hex2_linker M1 + mes.M1 -> mes.hex2 hex2_linker + stage0/elf32-header.hex2 + mes.hex2 + elf32-footer.hex2 -> mes mes + mescc.scm + tcc.c -> tcc.M1 -> tcc.hex2 -> mes-tcc mes-tcc + gcc.c -> gcc *done* ** Less naive recipe *** Use hex2_linker, M1 prototyped in C from https://github.com/oriansj/mescc-tools *** Remember that mes.M1 is compiled by mescc from prototyped src/mes.c in C. *** stage0/mescc-tools: TODO *** Mes M1 -f stage0/x86.M1 -f lib/crt1.M1 --LittleEndian --Architecture=1 > lib/crt1.hex2 M1 -f stage0/x86.M1 -f lib/mini-libc-mes.M1 --LittleEndian --Architecture=1 > lib/libc-mes.hex2 M1 -f stage0/x86.M1 -f src/mes.M1 --LittleEndian --Architecture=1 > src/mes.hex2 hex2 --LittleEndian --Architecture=1 --BaseAddress=0x1000000 -f stage0/elf32-header.hex2 -f lib/crt1.hex2 -f lib/libc-mes.hex2 -f src/mes.hex2 -f stage0/elf32-footer-single-main.hex2 > src/mes-mes exec_enable src/mes-mes *** Tinycc ./build.sh ./link.sh * DONE ** stage0: hex.0, M0 done; M1, hex2_linker prototyped in C ** tcc compiled with mescc correctly compiles: int main () {return 42;} ** mes+mescc.scm are mutual self hosting ** during development we run mescc.scm on Guile (mes is slooowww) ** tcc compiled with gcc is known to compile gcc * TODO ** fix mescc.scm so that tcc can correctly compile gcc ** fix bootstrap-loops: (Nyacc?, mes.M1?, psyntax.pp?) ** make GNU gcc (8.0?) bootstrappable again, remove [need for] tcc stage ** stage1/2 LISP, FORTH? ** rain1's LISP compiler? ** integrate with GuixSD ** x86_64, arm? * Contact ** #bootstrappable, #guix on freenode ** bootstrappable.org