IRC channel logs
2023-01-16.log
back to list of logs
<oriansj>well one could write a bigger compiler in the M2-Planet C subset but that just requires a good deal of work. <oriansj>and janneke has a good deal of experience with getting TCC bootstrapped. So there isn't a real advantage. <fossy>stikonas: i think i'm on the final revision-build cycle with python QEMU fixes now, python build system is a bit wack when it comes to machdep but figured it out (had to set some weird autoconf variable) <stikonas>fossy: nice. Feel free to push to PR, you can always push further fixes later <stikonas>fossy: I guess newest versions of Python don't need that fix? <fossy>stikonas, doras: then, new gcc is def next on my agenda <fossy>new pythons seem to have the same problem <stikonas>hmm, interesting, on my Gentoo system python reports as linux, not linux5 <fossy>hmm, perhaps i added something unnessecarily <fossy>i think we have discussed this in the past <stikonas>though I might have hit some problems with GCC 10, when I tried <stikonas>but in that case, just build some older version, like GCC9 or 8 <fossy>do you have any existing work on new gcc? <stikonas>and you can first try building with pregen files to see if it builds <stikonas>I tried running ./configure and make some time ago <fossy>yeah, no worries, i've done about the same :P <stikonas>but it was too long time ago, we have made significant changes now <fossy>new gcc has my focus after python, i'm hoping it should be relatively straightforward now that we have everything we could possibly need <stikonas>rebuilding all autotools stuff is probably the most annoying thing <fossy>especially autogen, will make it quite more straightforward <fossy>yeah, kernel is a bit of a different story haha <stikonas>but you'll still need to find exact versions of autotools that work <fossy>usually that isn't too bad, if you inspect the headers of the generated files in the tarball, no? <stikonas>yeah, usually they say what was used to generate it <stikonas>fossy: by the way, what should we do with package dependencies <fossy>yeah, i've been thinking a little bit about it <stikonas>now with shared libs, latest output tarballs start depending on others, at least musl <stikonas>so I've tried running python2 on my normal system and it works <fossy>i don't think there needs to be any significant revision to the build process to care about dependencies. live-bootstrap is a bit unique in that there's only one combination of packages - unlike distributions where there's a huge graph that has to be resolved, ours is all linear. so any notion of package dependencies doesn't really need to be presented outside of a development tool, imo <stikonas>no, it can be just some text file with recorded dependencies <fossy>yeah, that was my thoughts too <stikonas>I mean it's more for the purposes of using binary outside live-bootstrap <fossy>particularly in a human+machine readable format <stikonas>right now e.g. you can just run "bash" on any system <stikonas>can be musl distro, glibc distro, anything and it just runs <fossy>i think just a file with something like: "Python-3.11.1 rundep musl-1.2.1" or whatever is likely sufficient representation, it's really easy to parse and allows for different types of deps in place of "rundep" <stikonas>and for python, I have to unpack musl package; ln -s usr/bin bin; ln -s usr/lib lib; unpack python; PYTHONHOME=usr lib/ld-musl-i386.so.1 usr/bin/python <stikonas>yes, something like that would have sufficient info <fossy>yeah, doesn't need to be automatic <fossy>there's not any guesswork either if you use ldd <fossy>you just need to know where the library comes from <stikonas>we can emit list of so files that package provides <oriansj>well at live-bootstrap's current growth rate: fossy/stikonas will have their own Linux Distro with its own custom package manager <stikonas>well, one can always add extras, like openjdk :D <fossy>to some extent, live-bootstrap kinda is a jank as distro in its end form <fossy>but it's never really intended to be a distro <stikonas>but we had to add some distro like stuff (like packages) <fossy>yeah, useful abstractions that happen to come from distros, really <oriansj>well a package manager at its core only needs to download source files and orchestrate the building of packages <stikonas>by the way, at some point we probably want a new release of stage0-posix, I guess not urgent but we do have quite a few improvements now <stikonas>yeah DESTDIR stuff that comes from distros helped us a lot (we used to manually list individual files and their hashes before) <oriansj>well I guess rm, unbz2 and all of the major changes in M2libc/M2-Planet definitely would qualify for a fresh release <oriansj>did you want to do the honor stikonas? or shall I <stikonas>We should probably gather a changelog of all notable changes <stikonas>I can first do some more conversions of amd64/x86 defines... <oriansj>well if I was smarter, I probably should have added a CHANGELOG.org to stage0-posix like I have with mescc-tools <stikonas>well, we can start it for current release <stikonas>over the next days/weeks we can retroactively add stuff that was added since last release <stikonas>it's just that we are likely to get more coverage if we say what's new in the release. <stikonas>fossy: your last commit conflicts with my fix that I pushed <fossy>stikonas: i meant to rebase & remove mine, but forgot to before pushing <stikonas>that will help a bit with reviewing regenerated files <fossy>yes, it's not incredibly extensive, but the highlights are there