IRC channel logs
2024-02-10.log
back to list of logs
<matrix_bridge><Lance R. Vick> Possibly more reliable, but using a proprietary Microsoft product as the source of truth for a Linux distro is suboptimal IMIO. Codeberg has support for hosting arbitrary artifacts with no limits for open source projects though. I am considering mirroring all the external source files for stagex, including for live-bootstrap, there. <sam_>oriansj, fossy, et. al: thank you for the takes! <GoogulatorMobile>lrvick2: unless Microsoft is able to preimage SHA256, the worst they can do even with full control over the source tarballs is DoSing live-bootstrap, which is essentially what already happens with unstable sources, albeit not intentionally. <stikonas>fossy: is bzip2 download broken for you? <snuik>stikonas, you have 1 message <snuik>stikonas, Googulator says: amd64 and riscv64 checksums might need updating too <stikonas>hmm, perhaps just my one system is boken <oriansj>Googulator: well we should probably make many possible root sources of the tarballs. <oriansj>say git-lfs, gnunet, ipfs, torrents, hosting servers, etc <oriansj>I was also thinking of a gemini file share; as a quite minimal client could be used. <oriansj>say bootstrap gmni and just download the rest of the bootstrap tarballs. <stikonas>some of the sources are not easy to maintain, since tarballs change a bit as we develop live-bootstrap <lrvick2>Really strugging to get bison 3.8.2 reproducible. Headers all out of order. Having the same problem with gettext, but every other package is bit for bit identical. https://sprunge.us/37CODL <lrvick2>starting to conside the consequences of using the older version live-bootstrap uses <oriansj>Googulator: or if we trust unecrypted protocols, we could leverage bash to do the downloads: { echo -e "GET /gnu/wget/ HTTP/1.0\r\nHost: ftp.gnu.org\r\n\r" >&3; cat <&3 ; } 3<> /dev/tcp/ftp.gnu.org/80 <GoogulatorMobile>The issue isn't with trusting unencrypted protocols themselves, but rather for how long they will remain usable <GoogulatorMobile>Especially since that's a HTTP 1.0 request, which causes problems with virtual servers <GoogulatorMobile>IIRC Host: isn't even a valid header for HTTP 1.0, but 1.1 will probably need more headers to be valid <oriansj>That isn't a big problem either, as we could have a minimal client that just reads and writes to /dev/tcp/$HOST <oriansj>and protocols like torrents can surivive for decades without a single party being responsible for seeding <deesix>The usage of github for the whole development is, in my opinion, really-really disturbing, indeed. <oriansj>deesix: well stage0 and all related too tools could easily be moved to sourcehut (they are already mirrored there) <oriansj>but ultimately it is dependent on what other people like to use <deesix>The web of hardcoded submodules is not easy to undo; not to mention all the data about issues, etc... <oriansj>deesix: it is easier than you think. Just move the M2libc repo, then all the rest and finally stage0-posix/stage0-uefi/stage0 <oriansj>which when combined with: { echo -e "GET /gnu/wget/wget2-latest.tar.gz HTTP/1.0\r\nHost: ftp.gnu.org\r\n\r" >&3; cat <&3 ; } 3<> /dev/tcp/ftp.gnu.org/80 >| wget2-latest.tar.gz will get you the wget.tar.gz <oriansj>you know what, I might as well write a little C program that uses /dev/tcp to download files for mescc-tools-extra <oriansj>I wonder if Fiwix supports that too? <deesix>oriansj, good luck with those "just" and "easier". Moving everything out is practically imposible. <stikonas>and submodules are independent, they don't have to be in the same location <stikonas>so one can move main repo and submodules independently <stikonas>e.g. I self-host stage0-uefi myself but various submodules (M2-Planet, etc...) point to oriansj's repos <stikonas>what is slightly trickier is having two maintained upstream repos (e.g. like mescc-tools have) <oriansj>correction /dev/tcp is a bash abstraction that doesn't actually exist on linux <deesix>stikonas, consider all the existing history with the hardcoded URLs in the submodules, for example. I'd love to have a mirror of all repos, all with relative submodules; it's a mess (if even possible). And, again... the lock-in is not just about the repo (the "forge" contains lots of data). <oriansj>so I will probably need to read a bit more to make a tcp client <stikonas>there isn't that much data in the forge... mostly merge requests, most without even description <oriansj>deesix: if there is a sufficiently good reason to replace submodules with something better, we will do so <stikonas>one can always switch to relative paths for submodules <oriansj>deesix: as long as the developers here agree on what to do, the rest can be replaced. <stikonas>probably a bit more in individual submodule such as M2-Planet <oriansj>heck we did a vote to move here from freenode <oriansj>moving where submodules point is not an issue <oriansj>especially if the current git repos can just remain as minor mirrors <oriansj>with a line in the README pointing to where the master repo is located. <deesix>oriansj, how can one replace the already commited URLs in the history?