IRC channel logs
2026-06-02.log
back to list of logs
<janneke>ACTION forgot what stabbles intended to work on? <janneke>version-0.27 is stable, there will be a 0.27.2 some time soon <janneke>`wip' will be 0.28, probably including wip-bootstrap-x86 <stabbles>OK, I'm looking at bootstrapping using 64-bit only (both x86_64 because some distros disabled 32-bit, and aarch64 because my macbook's CPU cannot run 32-bit). I added aarch64 support to MES-replacement, which still uses MES's libc (maybe that project should be renamed :p). And I managed to adapt the Guix bootstrap path for RISC-V to AArch64 using <stabbles>Linaro's old GCC 4.7 branch, with a few shortcuts (system make, coreutils, ...). Now I'm looking at building GNU make using tcc-mes, and I thought I'd try to build a recent GNU make with FIFO-jobserver support. That way I can exploit parallelism across packages at a later stage. But: gmake 4 a handful of small changes to GNU mes: (1) a fix for <stabbles>O_DIRECTORY on aarch64 and RISC-V too I think, (2) assert.h honoring NDEBUG, which I saw on the `x` branch, (3) define _POSIX_VERSION without BOOTSTRAP_WITH_POSIX -- this one I have to double check, (4) ernno.h: ENOTSUP, (5) dirent.h: DT_* defines, (6) fcntl.h: O_CLOEXEC, O_NOBLOCK, struct flock, F_*LCK, (7) stdlib.h: missing declaration of mktemp, <stabbles>which has an implementation, (8) linux/ftruncate.c wrapper, (9) mkfifo wrapper, (10) stdio.h: tmpfile. <stabbles>These are all tiny diffs, but I could see you wanna keep things small ;) <stabbles>Some of these are actual bug fixes, like O_DIRECTORY being incorrectly defined on aarch64, and the assert.h change too, the rest is exposing more linux/libc features. <janneke>yeah, possibly some stuff from wip-bootstrap-x86 can/should be moved down to wip, or even version-0.27 <stabbles>And what is the "policy" for adding new parts of libc/syscalls etc? I suppose the idea is to keep it as minimal as needed? Do you consider building gmake 4.4 to be within the scope? <janneke>stabbles: yes; there's an older wip-gcc4 branch that aims to build gcc-4.6 without musl, but it needs some work <janneke>wip-gcc-4 should probably be rebased onto wip-bootstrap-x86 <stabbles>janneke: good to know. I think I'm fine with musl, cause it builds fast and only so many patches are needed. For AArch64 the situation is somewhat similar to RISC-V, cause official target support in GCC was only added in 4.8. But the Linaro folks did the backporting work back in the day to 4.7, and that is what tcc can build, so no work is needed <stabbles>there. I'm not aware of backports of aarch64 to GCC 4.6. <aggi>it's a little sad, the gcc4 aarch64 backport hasn't been merged into mainline with gcc4 <aggi>some years ago i began the rollback to gcc4.x here, keeping x86/32/64 and aarch32/64 targets <aggi>and maintained additional cross-toolchains for the latter because i didn't know aarch64 got backported <aggi>and, it's just coincidence, but a moment ago there's been a little noise on tinycc-devel again <aggi>which i repsonded to with a reminder there tcc is blocked at the jannecke fork for tcc-0.9.26 still <aggi>i didn't succeed myself neither to rebase for a bootstrap with tcc-mob/HEAD directly <stabbles>oh, so you have an independent GCC < 4.8 aarch64 branch? I've only found linaro's release tarball so far :p <aggi>stabbles: asking me? nope, i did at all expect anyone had backported aarch64 at the time when i rolled-back to gcc4 <stabbles>Ah right. I was very pleased to find Linaro's backport... it's for exactly the last version of GCC that doesn't require a C++ compiler <aggi>janneke: it's not urgent but i think both tinycc-devel and bootstrappable should keep this in mind