IRC channel logs
2026-03-19.log
back to list of logs
<aggi>^ this provides an x86 assembler (gas syntax) very early in the live-bootstrap dependency chain <aggi>in particular for 16bit real-mode pieces to compile/link/assemble a kernel long before python/perl/autotools were available <aggi>i'll do the intel syntax AS86 one next, although that's not needed, just another option if anytime anyone sneaks into with intel syntax asm <aggi>wildwestrom: i've not reviewed all of this yet <aggi>tools/genperf/genperf.c: Generate Minimal Perfect Hash (genperf) <matrix_bridge><wildwestrom> This assembler is written in C. I guess you still need a better assembler even after reaching tcc in the chain? <aggi>only ensured most files yasm needs are re-generated (except those three which need modules/arch/x86/gen_x86_insn.py <aggi>wildwestrom: NO, so far i managed to maintain a complete no-c++/tcc-toolchain system profile including 500 ebuilds here <aggi>including linux-tcc kernel, syslinux bootloader etc. which need both 16bit asm and 32bit <aggi>and tcc assembler togethwer with yasm (for the 16bit pieces) did suffice for ALL of this <aggi>wildwestrom: i've created mentioned steps-tiny/yasm-1.3.0/pass1.sh with three pre-generated files added only <aggi>for that matter, if tcc did support x86 16bit asm (which it doesn't) then yasm would not be needed at all <matrix_bridge><wildwestrom> aggi It looks to me like yasm is indeed built with tcc from C source. Before I looked I thought yasm was going to be an asssembler written in assembly like fasm. Not experienced enough to know how early in the chain it is. <aggi>good point wildwestrom, i would prefer a capable assembler available even earlier in the dependency chain (currently yasm links against musl-libc) <aggi>i've not tried fasm before, and fear it may have it's own macro syntax incompatible with GNU gas or intel/as86 <matrix_bridge><wildwestrom> aggi: Me neither. Fasm does indeed has its own unique macro syntax though. I mostly watched tsoding work with fasm. <aggi>uhm, syslinux contains a few *.bin; isolinux.bin mainly <aggi>to my knowledge any bootable ISO may need one or another variant of this <luke-jr>Andrius Štikonas: it looks like a real piece of work :( the original Prolog compiler was never released open source afaik <luke-jr>and apparently erlc doesn't just run a straightforward process, it sets up a compile server and then uses that <aggi>seems gentoo syslinux.ebuild hasn't rebuild core/isolinux.asm with nasm assembler for 10years and shipped a prebuild .bin ever since <aggi>now that's interesting, to produce some iso bootcode that builds up a gigantic dependency graph (in particular for GRUB), or it's not been assembled from source <luke-jr>aggi: GRUB doesn't normally depend on syslinux ebuild? <aggi>no luke-jr, of cause not, it's different issues with either GRUB or syslinux/isolinux, both got problems <aggi>it seems not possible to integrate GRUB early into live-bootstrap where it belongs, and syslinux/isolinux.asm is difficult to rebuild from source <aggi>and of cause, some bootloader is needed, so there's no way around these <aggi>maybe lilo is an option, didn't try that one yet - i'll probably not touch GRUB <aggi>it's remarkable too, although GRUB bootloader is the first thing loaded during start, with live-bootstrap it's the last one build in the dependency chain <luke-jr>on x86, there's often UEFI which would not require a bootloader? <luke-jr>though I don't recall any issues with GRUB <aggi>with current live-bootstrap, you have to compile a hundred million lines of code before GRUB is compiled <matrix_bridge><Lance Vick> When someone bootstraps ada or rewrites the graphics stack of coreboot in something other than ada. _sigh_ <matrix_bridge><Lance Vick> Ada is the last known blocker for coreboot that I am aware of. <luke-jr>I guess Talos II would be a better target? ;) <matrix_bridge><Lance Vick> currently laying on the floor disconnected. a true crime against humantiy <luke-jr>mine was manhandled by the FBI, so idk that I'll ever trust it <luke-jr>I suppose on Talos II, you'd need to bootstrap on the BNC first? <matrix_bridge><Lance Vick> no one has claimed this and it has sat stalled for a year <luke-jr>or maybe there's a way to get a minimal BNC that just launches the host, so you can cross-compile from there <aggi>for the lols, i'm currently watching X-files... which is the kind of mood i'm in when reviewing bootstrapping <aggi>paranormal activity, CIA conspiracy <luke-jr>for that level of paranoia, we'll need a build-your-own-punchcard book <luke-jr>and design a punchcard circuit that writes a bootable USB stick <aggi>it's not paranoia, because there is evidence the whole thing got clusterf**d <aggi>paranoia is a psychological condition of imagining things which no evidence exists for <aggi>F-35 UFO and it's dozens of PPC chips with proprietary blobs isn't fiction - it's the reason why this flaying saucer is critized for remote-control kill-switch <aggi>Eurofighter had a track-record of vendor-insanity too <aggi>for that matter, i'm not in the mood for any SPARC nor Motorola bootstrap either <fossy>stikonas, Googulator: thoughts on an ai-generated contribution policy for live-bootstrap? <fossy>at least one of vxtls' documents is certainly ai-assisted <fossy>and i think it might be time for a clearer contribution policy <fossy>my rough opinion is that it's probably silly to try to totally ban it, because people will do it anyways <fossy>and it's increasingly difficult to decide whether they wrote it or not <fossy>i'll suggest that A) all LLM-assisted commits should have a Co-Authored-By line in the commit log, B) it is made totally clear that any code you PR against the repository is code you are taking responsibility for (no deferring to LLMs, e.g. "I did it this way because Claude told me so") <fossy>C) there will be a lower tolerance for low effort commits that are LLM-assisted <Googulator>fossy: I'd suggest the REUSE headers, rather than the commit log <lanodan>I think it should be banned out similarly to Gentoo, sure people will do it anyway, but that's also true for stuff like enforcing against using work that's proprietary or under a non-compatible license (plus like since when are rules only for stuff everyone will follow?) <matrix_bridge><Andrius Štikonas> the line between what is assisted and what is not might be a bit fuzzy though. For example, what if one uses LLM to "explain this codebase" but not to write code. Or perhaps explain some topic <matrix_bridge><Andrius Štikonas> though the line between what is pregenerated and human written file is also somewhat fuzzy, with some files partially generated, partially human written, some were originally generated, then mostly rewritten by human