Looking at the rootfs size trend in: https://valkyrie.yocto.io/pub/non-release/20250506-54/testresults/buildperf-alma8/perf-alma8-vk_master_20250506090028_601b05a298.html for test1: Build core-image-sato: rootfs size, you can see that we have a significant increases recently: 1. Commit 78042, and 78059 which is mesa: https://git.yoctoproject.org/poky/log/?id=79fa12ae4b72b967051f7d19ffddac4fe3651171 Bill has questioned whether we just went from a barely unusable, very slow sato image to one with enough mesa to be performat and therefore usable. Regardless, we should investing these changes to see if the increase in rootfs size is sensible and worthwhile. Can we trim any content without giving up on useful features?
I have send couple of patches which should help with size and build time https://lists.openembedded.org/g/openembedded-core/message/216116 https://lists.openembedded.org/g/openembedded-core/message/216117
Bill is mostly correct: enabling llvmpipe in Mesa pulls in libllvm which is ~114MB. If a system wants llvmpipe then this is the cost. There's a good argument to be made for more modular packaging of mesa, so that systems that don't want it don't need to build or install it.
We may be able to share source between llvm and rust-llvm. Preliminary summary is that rust-llm has: 20 backported patches, and 2 rust specific ones. More data next week. Linked related "build time increase" as a depends since bugzilla doesn't have "Relates".
I think we've bottomed out on this. Disabling llvmpipe will reduce it back to where we were, which can be done if that functionality is not needed. Otherwise the size impact of having llvm is reduced from what it was.
Closing, I think we've done all we can do here for the immediate impact.