gnome parity gap

Measured 2026-08-11 21:51:17 UTC by scripts/measure-gnome-parity-gap.py (tunaos-packages#132). Two measurements, both from live sources: GHCR amd64 compressed layer sizes (the issue's own method, refreshed), and the per-base requested package lists from the tunaOS desktop manifests resolved against the GNOME desktop contract. Machine-readable result: docs/gnome-parity-gap.json.

Reproduce with:

scripts/measure-gnome-parity-gap.py

1. Image size deltas (GHCR, amd64, compressed)

variant	base	gnome	delta	2026-07-30 delta (issue)
sailfin	1.52	1.96	+0.44	+0.16
flounder	1.14	1.75	+0.61	+0.33
grouper	1.87	2.32	+0.45	+0.32
skipjack	2.60	2.95	+0.35	+0.99
albacore	2.60	2.95	+0.35	+1.00
yellowfin	2.53	2.88	+0.35	+0.99
bonito	3.27	3.96	+0.69	+0.70
marlin	1.36	2.10	+0.74	+0.68
guppy	3.34	4.90	+1.56	+1.76

All nine editions were rebuilt 2026-08-10 (config created timestamps). The EL-family deltas dropped from ~+1.0 GB to ~+0.35 GB because those editions moved from COPR group installs to the leaner native GNOME 50 RPM chain; the non-RPM variants rose from +0.16/+0.33/+0.32 to +0.44/+0.61/+0.45 as the manifest fixes from 2026-07-30 were published. Size alone is no longer a discriminator between healthy and thin editions — which is why the contract below, not the size table, is the gate.

2. GNOME contract coverage per variant

Core (contract + session surface)

variant	gdm	gnome-shell	mutter	gnome-session	gnome-keyring	gvfs	nautilus	portal-gnome	gnome-settings-daemon	gnome-control-center	portal-gtk
sailfin	✔	◆	◆	✔	✔	✔	✔	✔	✔	✔	✔
flounder	◆	◆	◆	◆	◆	◆	✔	✔	◆	◆	✔
grouper	◆	◆	◆	✔	✔	◆	✔	✔	◆	◆	✔
skipjack	✔	✔	◌	◌	◌	✔	✔	✔	✔	✔	◌
albacore	✔	✔	◌	◌	◌	✔	✔	✔	✔	✔	◌
yellowfin	✔	✔	◌	◌	◌	✔	✔	✔	✔	✔	◌
bonito	✔	✔	◌	◌	◌	✔	✔	✔	✔	✔	✔

✔ listed explicitly · ◆ supplied by a listed metapackage (measured fact) · ◌ supplied by an installed dnf group (runtime-verified on the reference editions) · ✘ nothing in the list supplies it.

Apps measured absent from the openSUSE pattern (2026-07-30)

variant	gnome-bluetooth	gnome-online-accounts	gnome-initial-setup	gnome-disk-utility	fwupd	yelp	orca	search-index	gnome-color-manager	gnome-remote-desktop	gnome-user-docs
sailfin	✔	✔	✔	✔	✔	✔	✔	✔	✔	✔	✔
flounder	◆	◆	◆	✔	◆	◆	◆	◆	◆	◆	◆
grouper	◆	◆	◆	✔	◆	✔	◆	◆	◆	◆	✔

search-index is tracker/tracker-miners on dnf/apt and tinysparql/localsearch on openSUSE.

3. Findings

• The requested lists now cover the full GNOME contract — As of the tunaOS manifests fetched for this run, every audited variant (sailfin, flounder, grouper, skipjack, albacore, yellowfin, bonito) requests every core contract component and every app the 2026-07-30 audit measured missing. sailfin resolves the apps explicitly (62 requested names); flounder through Debian's gnome-core metapackage plus explicit nautilus/portals; grouper through ubuntu-desktop-minimal plus the gnome-keyring and libpam-gnome-keyring the metapackage was measured to omit.

• The 2026-07-30 images are stale; current images are in the reference bracket — The issue measured sailfin +0.16 / flounder +0.33 / grouper +0.32 GB against a ~+1 GB healthy delta. The manifest fixes landed in tunaOS main on 2026-07-30/31 (ce11d21, d8ebdf6) and all nine editions were rebuilt 2026-08-10: sailfin is now +0.44 GB, flounder +0.61 GB, grouper +0.45 GB — inside the same bracket as the EL-family references (skipjack/albacore/yellowfin +0.34–0.36 GB, which dropped from ~+1.0 GB when they moved to the native GNOME 50 RPM chain).

• Remaining gap 1 — published apt images cannot be inventoried — bootc does not commit /var, so dpkg-query is empty on flounder/grouper images and no installed-package inventory exists to diff. The build-time contract (verify-desktop-experience.sh in tunaOS, verify-gnome-desktop-experience.py here) is the only guard for the apt bases; a build-time inventory written to /usr (issue ask 2) would make them diffable directly.

• Remaining gap 2 — sailfin's published image has no published inventory — rpm -qa works on sailfin:gnome, but no inventory has been published alongside the image, so ask 1 of the issue (diff effective installed sets against yellowfin:gnome) still has no committed answer. Publishing the rpm -qa output with the image metadata closes it.

• Next candidates — kde/xfce/niri zypper lists are still thin — The same shape that produced this issue is unfixed for the other desktops: kde.yaml lists 3 zypper names, xfce.yaml 3, niri.yaml 2. The pattern-family resolution these rely on should get the same explicit-component treatment sailfin's gnome.yaml got (tunaos-packages#133 tracks the broader audit).

4. Provenance

• gnome.yaml: b354498ae888 (2026-08-11) — docs: add references to package sourcing policy (#1319)

• gnome-debian.yaml: 28ac5e8ea9fe (2026-08-07) — feat(flatpak): Flathub everywhere, curated preinstall set, and a contract that keeps both (#1062)

• Registry: GHCR reachable at 2026-08-11 21:51:17 UTC.

5. Caveats

• Runtime package counts are unreliable for the apt bases: bootc does not commit /var, so dpkg-query returns empty inside flounder:base/grouper:base. A build-time installed-package inventory written to /usr is required to diff published apt editions directly (issue thread, 2026-08-10). Until that lands, the apt rows above verify the requested lists, and the tunaOS build gate (verify-desktop-experience.sh) verifies the built image.
• This audit reads the tunaOS manifests at their fetched commit, not the state an image was built from; a list change between that commit and the next publish will not show up until the next run.
• sailfin:gnome is the only GNOME edition that currently builds an ISO end-to-end (tuna-os/iso-builder#32); it is no longer the thinnest edition, but it is also the only one whose published image can be inventoried with rpm -qa — publishing that inventory would make ask 1 of the issue directly checkable.