flatpak: hydrate a 1.45 GiB flatpak-ci shell instead of 10.4 GiB .#flatpak + 14.5 GiB .# #52
No reviewers
runner
android
linux
infra
windows
agent
devex
security-agent
assistant
ios
web
cmo
billing
forge
release-runner
cdo
manager
Labels
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
key-store/key.store!52
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "claude/flatpak-slim-shell"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
KSE-0186 D2 second slice. Dedicated minimal devShells.flatpak-ci (measured on the forge: 1.45 GiB / 353 paths vs 10.36 GiB for .#flatpak; the .# shell used for rclone was 14.5 GiB + 7 drvs built at job time). Arch-repo scripts, Merge, Build (Linux) upload and publish-flatpak download repointed. .#flatpak unchanged (Build (Linux) needs Bazel). Verified: nix eval + nix build of .#devShells.x86_64-linux.flatpak-ci on the forge; YAML + bash -n. Next slice: bind-mount the host /nix into the docker sandbox to remove the second hydration pass; cache volumes on the linux-large targets (common.nix, needs forge deploy).
The Flatpak arch-repo legs spent ~36 of their ~39 minutes (x86_64, task 76914) hydrating Nix closures: `.#flatpak` on the host for the GPG materialise (321 paths / 10.4 GiB -- devShellToolingInputs and the commonShellEnv string interpolations drag JDK/Bazel, the Swift toolchain, the Android SDK/NDK and jsign into a shell that only runs flatpak-builder), `.#flatpak` again inside the debian:12-slim sandbox with its own fresh /nix, and then the default `.#` shell (568 paths / 14.5 GiB, plus 7 derivations built at job time) for a single `rclone copy` (KSE-0186 D2). flatpak-builder + ostree themselves take ~1.5 min. - flake.nix: `devShells.flatpak-ci` is now a dedicated minimal shell (the tools Scripts/release/build-flatpak-arch-repo-ci.sh and Scripts/build-flatpak-testing-repo{,-docker}.sh actually exec: flatpak, flatpak-builder, bubblewrap, ostree, xdg-dbus-proxy, appstream, desktop-file-utils, gnupg, age, rclone, jq, git, curl, core utils) without devShellToolingInputs/commonShellEnv. Measured on the forge: 1.45 GiB (353 paths) vs 10.36 GiB for `.#flatpak`. `.#flatpak` itself is unchanged (Build (Linux) still needs Bazel from it). - The arch-repo scripts use `.#flatpak-ci` for the GPG materialise, the in-sandbox build/ostree calls, the direct-path build, and the rclone upload; Merge (Flatpak Repo), Build (Linux)'s upload and publish-flatpak's download use `.#flatpak-ci` for rclone instead of `.#`. Expected: x86_64 leg ~39 min -> single-digit minutes once the sandbox shares the host store (next slice); ~7x less NAR traffic from nix.key.store per leg immediately. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>