IRC channel logs
2024-06-16.log
back to list of logs
<matrix_bridge><cosinusoidally> aggi: I think it's more likely that the GCC developers were making dubious "optimisations" on the basis of undefined behaviour. iirc Linux also makes heavy used of gcc extensions. I wonder if any of the bsd kernels are better in that regard. I would guess that they may be more stricter ansi C codebases. <aggi>fyi, linux-2.4 has some #ifdef for -D__STRICT_ANSI__ <aggi>but i haven't arrived there yet <stikonas>Googulator: hmm, tcc mob also has some problems with fiwix... <stikonas>tcc itself seems to insert calls to memset <stikonas>hmm, I wonder if we can disable this include <stikonas_>it's just that I treid to upgrade tcc and suddenly it fails to build <matrix_bridge><cosinusoidally> sorry I probably didn't phrase that right. I meant was there a ticket raised with fiwix about it not building with tcc mob? Out of interest, which version of mob are you testing? Are you planning to upgrade live-bootstrap to use some newer mob version? <matrix_bridge><Andrius Štikonas> I hit a problem on x86 with and older tcc <matrix_bridge><Andrius Štikonas> somehow updating bootstrappable tcc to newer version that fixes riscv bugs breaks fiwix <matrix_bridge><Andrius Štikonas> tcc-boot [old] -> tcc 0.9.27 -> fiwix worked <matrix_bridge><Andrius Štikonas> tcc-boot [updated] -> tcc 0.9.27 -> fiwix doesn't work <matrix_bridge><Andrius Štikonas> so I'm testing tcc-boot [updated] -> tcc mob -> fiwix <matrix_bridge><Andrius Štikonas> tcc-boot [updated] -> tcc 0.9.27 -> tcc 0.9.27 doesn't work due to some other bug (tcc 0.9.27 couldn't self-host with meslibc) <aggi>on the slackeware11 buildhost i could confirm tcc commit e7262accb6e6965452e7f82c0ff281ef8412af4d could build a linux-2.4.26 and it didn't crash (i didn't git-pull for a little while) <aggi>currently i am gradually re-covering all test-cases to arrive at linux-2.4.37.11 compiled with both gcc-4.7 and tcc do not crash when booting <aggi>together with the customized non-Kbuild scripting, and differnt kconfigs (because tcc cannot digest everything yet without patching, to arrive at [optional] ACPI support for example) <aggi>i am merely a little excited, the linux-tcc test-cases could be confirmed, after a year of stagnation, there's progress <aggi>this is the plan: bootstrappable -> your tcc-version -> my tcc-version -> c-only profile stage0-x86-linux-tcc.iso (no gcc, no binutils, no automake) -> confirming bootstrapping gcc/binutils gradually from there <aggi>i am struggling with the same issues, related to various compiler versions, cross-compilation leakage issues etc. <aggi>and it's a principle concern, buildhost was leaking into tcc-linux compilation, when it should not and must not have <aggi>because, i am rather cautious with gentoo-crossdev here, going back and forth from/to aarch32/64/x86/64 <aggi>all of the sudden an i386-tcc coming from such a host fails to compile linux without it crashing, and moving to slackware11 magically solved this issue <aggi>anyway, as another topic, which i haven't arrived at yet, i am looking forward to discussions among #gentoo-releng/toolchain, how i am supposed to bootstrap gentoo, without a _non-reproducible_ python.bin acquired from somewhere <aggi>yet, this is low priority, because i'll migrate my stuff away without relying on python anywhere