IRC channel logs

2024-10-25.log

back to list of logs

<aggi>i did check live-bootstrap tree for perl too, but i keep ~500ebuilds in maintenance here and ocassionally missed one or another spot
<aggi>seems the python-3.8.16 pass1.sh bothers with lib-dynload, which wouldn't be supported by tcc/static, besides tcc-assembler couldn't digest some mnemonics that python uses
<aggi>libffi iirc
<aggi>thing is, i'm less concerned about python/portage, that's not too challenging, but i scripted my whole workflow around gentoo crossdev
<aggi>and that's one i will be missing
<aggi>my focus is a full transition towards a tcc-toolchain only system (plus binutils required for 16bit real-mode asm)
<aggi>and this one is x86-exclusive anyway
<aggi>even when risc-v progress is made with tcc, i don't see a chance to compile any recent linux-5.x/6.x kernel version with tcc ever that risc-v would need
<aggi>hence the python/portage/crossdev issue isn't critical, since there isn't any supported target other than x86 to transition to anyway
<aggi>with tcc that is
<aggi>however, maybe the best approach would be, to spawn the full live-bootstrap seeding from i486-tcc-linux-musl.iso
<aggi>and reastablishing a gentoo-crossdev atop of that, with a verified/clean bootstrapping process for it
<aggi>because i don't want to loose the flexibility to transition to arm/risc whatever, and this is what crossdev is good at
<aggi>so, i'll get perl with tcc/statick back on board this week, since it is known-good among #bootstrappable i can trust
<aggi>and then, back to kernel+tcc, the linux-2.4 hackjob, don't know yet if musl-libc interacts with this kernel version
<aggi>another question: to my knowledge a gentoo/portage bootstrap was feasible with live-bootstrap
<daddy>stikonas: interesting, we had a dead mirror in stagex for openssl 1.1.1.
<stikonas>proably just moved to different url
<daddy>stikonas: might be related to the archive.org outage. does princeton's mirror have 3.whatever? https://codeberg.org/stagex/stagex/commit/02f339cbc9decb9d56db9b48a5734fc4b4744da6
<stikonas>like old.openssl.org
<aggi>however, i haven't got python itself with tcc/static-linking, and live-bootstrap relies upon python
<daddy>going to openssl.org points to openssl-libary.org which points to github archives
<aggi>python is problematic (cross-compilation, static linking, tcc), it's not a showstopper such as perl/autotools would have been
<aggi>gladly there seems not much that is written in python for live-bootstrap
<aggi>as far as libre/openssl are concerned, i did a test-run against wolfssl last week, but this can't avoid the ssl.h api related issues
<aggi>as soon as there's an #include <ssl.h> fun begins, with any libre/open/wolfssl, but this is what most pkgs need
<aggi>bearssl is much easier to handle, but it doesn't implement the entire ssl.h api, that's why it's much simpler
<aggi>but i don't see a realistic chance to avoid ssl.h, although curl supports bearssl for example, there's several dozen other pkgs that would need extensive re-write for a clean tls api/library such as bearssl
<aggi>so far this question is answered, i'll have to avoid python when focussing on the tcc/static system-profile
<aggi>python can be bootstrapped, but it's not self-hosting with tcc/static itself, ok.
<aggi>so, the most realistic approach to spawn i486-tcc-linux-musl.iso atop/with live-bootstrap is re-writing the python-scripting (rootfs.py)
<aggi>which is feasible
<aggi>the biggest show-stopper is linux-2.x/tcc still
<stikonas>as for building python with gcc, I don't have much insight here...
<stikonas>s/gcc/tcc/
<stikonas>haven't tried for more than a few minutes
<stikonas>and python really needs dynamic linking for some stuff
<aggi>README.rst "Without using Python", well, ok.
<aggi>my mistake
<aggi>to my understanding this part is sufficiently stable and completed by #bootstrappable
<aggi>and it's an almost trivial fork to stop at a working tcc-toolchain emitted; so i can skip the gcc related parts completely, and proceed with my i486-tcc_static-linux-musl.iso scripting
<aggi>this however, i have to rewrite, because i did the whole tracking with gentoo-crossdev and python is gone indefinitely
<aggi>no big issue
<aggi>then this yields a BUILDHOST with tcc, and this one can bootstrapp gcc/binutils/python/crossdev towards any desireable target
<aggi>except a bootable linux-2.x/tcc kernel (or any later although i have doubts this is feasible, others failed before)
<aggi>fyi, i just did git-commit and tag tcc-toolchain for the complete set of ~500 ebuilds i was tracking, i already got a root.squashfs for this
<aggi>but i can't boot this with a kernel that compiled/linked with tcc, not yet
<aggi>that's two big tasks remaining mainly for me: 1) rewriting the crossdev/portage scripting, that's not an all at once so i can do this gradually for those ~500 ebuilds beginning with crucial development utils, even gdb-7.12 compiled/linked with tcc seems to work now, since the cmp $0 patch was applied
<aggi>and 2) thats a total show-stopper currently, linux-2.x/tcc
<aggi>if and once this is completed, there's a candidate for a bootstrappable tcc distro
<aggi>something which was missing for decades
<stikonas>Googulator27: any thoughts about what to do with openssl?
<stikonas>archive.org link is not working...
<Googulator>hmm...
<Googulator>what version do we need now?
<matrix_bridge><Andrius Štikonas> 3.0.13
<stikonas>the link was http://web.archive.org/web/20240213065706im_/https://www.openssl.org/source/openssl-3.0.13.tar.gz
<stikonas>but I guess that's flexible...
<stikonas>See https://github.com/fosslinux/live-bootstrap/pull/481 for failed CI...
<lanodan>3.0.13 should be available on https://distfiles.gentoo.org/distfiles/c0/openssl-3.0.13.tar.gz for a bit at least, does openssl not keep archived versions?
<lanodan>Same hash as https://github.com/openssl/openssl/releases/download/openssl-3.0.13/openssl-3.0.13.tar.gz (found on https://openssl-library.org/source/old/3.0/index.html )
<stikonas>lanodan: that's https link
<stikonas>we can jus as well download from github...
<stikonas>oh, http works too there
<stikonas>Googulator: shall we switch to distfiles gentoo for now?
<lanodan>btw https://gitweb.gentoo.org/repo/gentoo.git/atom/dev-libs/openssl/Manifest?h=master for a feed to see when it'll eventually get removed
<stikonas>hmm, maybe that's not too ideal either
<stikonas>hm
<stikonas>wolfssl is potentially buildable very early
<lanodan>I picked bearssl on my side, at least curl (and so git as well) can use it.
<stikonas>it has far more dependencies thoguh
<stikonas>ior maybe not
<stikonas>still C
<stikonas>though again, does it have http only downloads...
<stikonas>maybe it doesn't matter that much, it's much smaller
<stikonas>so can be included in early image
<deesix>Maybe... http://deb.debian.org/debian/pool/main/o/openssl/openssl_3.0.13.orig.tar.gz
<lanodan>Oh wow yeah, 15M for OpenSSL 3.0.13 (and 18M for 3.3.2) vs 748K for bearssl
<deesix>Also around here, for example (they keep all) snapshot.debian.org/archive/debian/20240325T091329Z/pool/main/o/openssl/openssl_3.0.13.orig.tar.gz
<lanodan>Oh yeah snapshot.debian.org is probably a good idea
<deesix>Not downloading would be better ;)