@monsieuricon i was *just* about to ask you why shallow clones are garbage from your last blog, i guess that bundles missing might not help with that for sure. and god i wish that client would flip, it's a huge deal for large monorepos (e.g. our firefox clone...)
@monsieuricon id love this. We do weekly ish builds of kernel and to keep size down I put bundles into a docker image in our local cache (only docker images get cached, thusβ¦) but itβs still pretty rough to get up to a current tag.
So the bundle clones would help me at least to not hammer the bandwidth each attempt.
@monsieuricon This is what we've been doing for Armbian build system: https://github.com/armbian/shallow
This takes the kernel.org bundles (torvalds & stable), updates them from live trees, and produces .git tarballs in OCI (ORAS) artifacts backed by GH's CDN, ready to go, in a few variants (full tree, shallow-per-version, which includes all -rc tags and stable point releases).
eg to get a shallow 6.18.y tree: get oras `apt install oras`, then `oras pull ghcr.io/armbian/shallow/kernel-git-shallow-6.18:latest`, untar, and you get a `.git`.
Compared to shallow cloning from git, it reduces server load drastically, but also by delivering a .git (instead of cloning) it saves client cycles and time.
build/armbian uses a few heuristics (disk free space/speed) to auto-determine for the (casual) user if a full or shallow tree is better suited for their specific build...