Wayland Protocol Surface — margo vs niri vs Hyprland vs mango#

Correctness update (2026-07-13, resolved in 1.1.9): margo briefly withdrew wp_fifo_v1 and wp_commit_timing_v1 because their managed commit barriers were never released (a hidden-then-shown Chromium surface could stall behind them). Both globals are advertised again as of 1.1.9, now driven by a real barrier-release scheduler (margo/src/state/pacing.rs): FIFO barriers signal at every per-output present, hidden surfaces release from the frame-callback fallback tick, and commit-timing deadlines get an exact one-shot wake. The advertised surface is back at 57.

margo refresh (2026-07-18): two corrections from a fresh source audit. (1) wp_color_representation_v1 is now advertised — a real hand-rolled implementation (margo/src/protocols/color_representation.rs) that only claims what the render path does: premultiplied_electrical alpha + identity/full coefficients; YCbCr matrices are deliberately not advertised while YUV conversion happens driver-side in EGL. Only margo and mango ship this protocol. (2) wp_color_management_v1 corrected ✅→⚠️: the delegate + full Phase-1 plumbing are compiled, but the global has been withheld since 28ad09bf ("do not register the global until Phase 2 lands" — the mpv/Chromium probe crashes). The delegate-macro count missed this; the advertised surface therefore stays 57 (−1 gated colour-management, +1 new colour-representation). When HDR Phase 2 flips the gate, both will be live and the count rises again. Same-day addendum: wp_drm_lease_device_v1 also shipped — the architecture blocker fell to the prescribed small backend-access change (a MargoState-held Rc into the udev BackendData), so the advertised surface is 58 (−1 gated colour-management, +colour-representation, +drm-lease). Only tearing_control remains unadvertised, and that one waits on upstream smithay.

mango column re-walk (2026-07-17, mango 0.15.2-10 / 242ffe8): mango moved to wlroots 0.20 + SceneFX and its column below was stale (carried from 0.13/0.14). Re-counted from source (hand-curated wlr_*_create protocol globals): headline ~53 → ~55. Cell flips: wp_color_management_v1 ❌→✅ (created whenever the FX renderer reports colour-transform support — the per-monitor hdr knob gates HDR output, not the global; mango also ships wp_color_representation_v1, which margo answered with its own on 2026-07-18) and xwayland_shell_v1 ❌→✅ (wlroots creates it in XWayland builds). Newly counted since the old walk: ext_data_control_v1, ext_foreign_toplevel_list_v1 + per-window/output image-capture-source managers (window-capture path), wl_fixes. The rest of the modern set is still missing (content-type, fifo, commit-timing, xdg-dialog, system-bell, pointer-warp, security-context). Only the mango column was touched this pass.

Last refreshed: 2026-06-17 — margo column re-counted directly from source at v1.0.7 (446f8597): margo advertises 57 protocol-bearing smithay delegate_* macros (excluding the two framework-internal delegate_dispatch / delegate_global_dispatch); all 8 hand-rolled GlobalDispatch globals overlap an existing delegate, so they add no extra. This corrects the prior headline ~60 (which over-counted — the standing-picture prose below already said ~57) to the source-verified 57. The protocol set is unchanged from the v1.0.3 audit: tearing_control and drm_lease are still the only two not advertised (both blocked, see gaps below). Sources (re-counted from each project's HEAD this pass): - margo v1.0.7 (446f8597) — this repo; 57 protocol delegate_* macros (+ overlapping hand-rolled GlobalDispatch). - niri v26.04-34 (fdb6d85) — same smithay method; 41 protocol delegates (unchanged from the 2026-05-28 walk). - Hyprland v0.55.0-189 (2d190ba) — src/protocols/*.cpp (63) + src/protocols/core/ (6) = 69. - mango 0.15.2-10 (242ffe8) — wlroots wlr_*_create() protocol globals, hand-curated to exclude non-protocol calls (wlr_scene_*, wlr_output_layout_*, device/backend objects); ~55, re-counted 2026-07-17 (wlroots 0.20 + SceneFX).

Companion: protocol-matrix.md is the internal view — for each protocol margo advertises, whether it's implemented and how it's tested (CI fixture vs manual vs none). This doc is the cross-project view (who advertises what).

This document is the source-of-truth audit of which Wayland protocols each daily-driver compositor advertises. mango — the C/wlroots (dwl-derived) compositor that margo is the Rust rewrite of — is included so the comparison shows what margo gained, kept, and dropped relative to its own origin. Maintained alongside road_map.md §15.10; this file is the read-side reference for "what does each compositor actually advertise."

Legend#

Mark	Meaning
✅	shipped / advertised
⚠️	partial (e.g. read-only, no write-side; or reached via a different mechanism)
❌	not advertised

Headline score#

Counted with each project's native method (delegate macros / protocol files / wlr_*_create globals), core protocols included so the numbers are comparable.

Compositor	Protocols (approx.)	Stack	Note
Hyprland `0.55.0`	69	C++ (hand-rolled)	Widest surface — 63 protocol modules + 6 core, plus Hyprland-only extensions (source-counted 2026-06-17)
margo `main`	58	Rust / smithay	Modern surface; ahead of niri and mango; pursuing Hyprland. FIFO/commit-timing rejoined in 1.1.9 (`state/pacing.rs` scheduler); colour-representation + drm-lease landed 2026-07-18
mango `0.15.2`	~55	C / wlroots	Broad and modernizing: wlroots 0.20 + SceneFX hand it the surface for free — now incl. HDR colour-management (re-counted 2026-07-17)
niri `26.04`	41	Rust / smithay	Tightest surface; deliberately minimal (source-counted 2026-06-17)

This refresh (2026-06-17) re-counts the surface from source; the standings are unchanged — only the margo headline number was corrected (~60 → source-verified 57, matching the standing-picture prose that already said ~57). What was re-counted directly from each project's checkout this pass: margo = 57 protocol delegate_* macros, niri = 41 (identical to the 2026-05-28 walk), Hyprland = 69 (63 src/protocols/*.cpp + 6 core/; was 68 — consistent with its steady module creep, e.g. the earlier hyprland_lock_notify addition). mango (~53) was not re-counted this pass — its wlroots globals need manual curation to separate true protocol globals from scene/output wlr_*_create calls, so the 2026-05-28 figure is carried forward.

Scope of this pass: only margo's column was re-verified against source (every margo ✅/❌ in the tables below was confirmed against the delegate_* list). The niri / Hyprland / mango per-protocol cells carry forward from the 2026-05-28 four-way walk and were not re-walked protocol-by-protocol here.

The current standing picture: margo (~57) narrowly leads its own C ancestor mango (~55) and niri (~41), having hand-rolled the three wlroots-freebies mango got for free — zwlr_foreign_toplevel_manager_v1 (write-side, P2), ext_workspace_v1 (P5), zwlr_virtual_pointer_manager_v1 (P7) — on top of the modern surface (content-type, fifo/commit-timing, security-context, pointer-warp, xdg-dialog, system-bell, toplevel-icon, toplevel-tag, xwayland-keyboard-grab) that mango's wlroots base doesn't expose. HDR colour-management left that list in mango 0.15: wlroots 0.20 ships it and mango wires it up, while margo's own manager global stays Phase-2-gated (see the 2026-07-18 note). mango's wp_color_representation_v1 lead lasted a day — margo advertises its own since 2026-07-18. output_power now ships (external DPMS control), and fifo/commit-timing are driven by a real presentation scheduler since 1.1.9. The two margo still doesn't advertise (tearing_control, drm_lease) are blocked, not just deferred — see "remaining gaps" below.

Core baseline — present in all four#

All four advertise these. No comparison table — assume ✅ everywhere.

wl_compositor / wl_subcompositor / wl_shm / wl_seat / wl_output
wp_viewporter, wp_presentation_time, wp_cursor_shape_v1, wp_fractional_scale_v1
zwlr_layer_shell_v1
xdg_shell, xdg_activation_v1, xdg_decoration_unstable_v1, xdg_foreign_v2
linux_dmabuf_v1
ext_session_lock_v1, ext_idle_notifier_v1, zwp_idle_inhibit_manager_v1
wl_data_device, zwlr_data_control_manager_v1, ext_data_control_manager_v1, zwp_primary_selection_device_manager_v1
zwp_pointer_constraints_v1, zwp_relative_pointer_manager_v1, zwp_pointer_gestures_v1
zwp_text_input_manager_v3, zwp_input_method_manager_v2, zwp_virtual_keyboard_manager_v1
zwp_tablet_manager_v2
zwlr_gamma_control_manager_v1, zwlr_screencopy_manager_v1
org_kde_kwin_server_decoration, wp_single_pixel_buffer_v1
zwp_keyboard_shortcuts_inhibit_v1
zwlr_output_manager_v1 (output-management — read topology + apply scale/transform/position; all four ship it, was previously missing from this audit)

Modern protocol surface — where margo leads the pack#

These are the protocols that separate a current compositor from a legacy wlroots one. margo and Hyprland carry almost the full set; niri and mango each miss large chunks.

Protocol	margo	niri	Hyprland	mango	Use case
`wp_color_management_v1` (HDR)	⚠️ opt-in	❌	✅	✅	HDR / wide-gamut output (margo: full plumbing incl. the complete info-event chain; advertise with `color_management = 1` — default off until the mpv/Chromium probe path is hardware-verified; mango since 0.15 via wlroots 0.20, renderer-gated `hdr` knob)
`wp_color_representation_v1`	✅	❌	❌	✅	Alpha-mode / YCbCr-matrix buffer metadata (margo since 2026-07-18: honestly advertises exactly what the render path does — premultiplied alpha, identity/full)
`linux_drm_syncobj_v1` (explicit sync)	✅	❌	✅	✅	Tear-free GPU sync, NVIDIA
`ext_image_copy_capture_v1` + capture-source	✅	❌	✅	✅	Modern screencast (replaces screencopy)
`xwayland_shell_v1`	✅	⚠️ diff path	✅	✅	HiDPI XWayland scaling (mango: created by wlroots in XWayland builds)
`wp_content_type_v1`	✅	❌	✅	❌	Game / video / photo hint
`wp_fifo_v1`	✅	❌	✅	❌	FIFO commit ordering; barriers released by `state/pacing.rs` since 1.1.9
`wp_commit_timing_v1`	✅	❌	✅	❌	Explicit commit-time targets; deadline wake + present-pass release since 1.1.9
`wp_alpha_modifier_v1`	✅	❌	✅	✅	Per-surface alpha hint
`xdg_wm_dialog_v1`	✅	❌	✅	❌	Modal dialog hint
`xdg_system_bell_v1`	✅	❌	✅	❌	System bell
`wp_pointer_warp_v1`	✅	❌	✅	❌	Programmatic cursor warp
`wp_security_context_v1`	✅	✅	✅	❌	Flatpak / sandboxed clients

margo's unique / near-unique protocols#

Protocol	margo	niri	Hyprland	mango	Note
`zwp_xwayland_keyboard_grab_v1`	✅	❌	❌	❌	margo unique — X11-side kb grab for VNC/VM/remote
`xdg_toplevel_icon_v1`	✅	❌	❌	❌	margo unique — inline app icons on toplevels
`xdg_toplevel_tag_v1`	✅	❌	✅	❌	Semantic toplevel tags (Hyprland ships `XDGTag` too)

Correction this refresh: xdg_toplevel_tag_v1 is no longer margo-unique — Hyprland 0.55 ships it as XDGTag.cpp. The previous "margo unique vs both niri and Hyprland" claim was stale. Only xwayland_keyboard_grab and xdg_toplevel_icon remain unique to margo across all four.

Recently shipped — the wlroots-freebie catch-up (P2 / P5 / P7)#

Three protocols mango/Hyprland get for free from wlroots, hand-rolled in margo (ported from niri) so it no longer trails on them:

Protocol	margo	niri	Hyprland	mango	What it unlocks
`zwlr_foreign_toplevel_manager_v1` (write-side)	✅	✅	✅	✅	Taskbar click-to-activate / close / (un)fullscreen (mshell active-window pill)
`ext_workspace_v1`	✅	✅	✅	✅	Workspace state for third-party bars (sfwbar, ironbar) — the standard bar-state path now that dwl-ipc is gone
`zwlr_virtual_pointer_manager_v1`	✅	✅	✅	✅	Synthetic pointer — `wtype --click`, remote desktop, accessibility

margo's foreign-toplevel write-side runs alongside the existing smithay ext-foreign-toplevel-list-v1 (read side stays untouched). ext_workspace_v1 maps each output to a workspace group with 9 fixed tag-workspaces (active = bitmask membership); it is the standard bar-state protocol now that dwl-ipc has been removed (mctl + mshell use the Unix control socket instead). virtual_pointer feeds margo's normal input path.

The remaining gaps — blocked, not just deferred#

These two are not a matter of effort — each hits a concrete upstream or architectural wall (re-confirmed by source audit 2026-05-28). (output_power was the third; it shipped in 1.0.2.)

Protocol	margo	niri	Hyprland	mango	Why margo can't ship it cleanly today
`zwlr_output_power_management_v1`	✅	❌	✅	✅	Shipped (1.0.2). `set_mode` maps onto the recoverable `request_dpms` deferred-queue path (`DrmCompositor::clear()`); input-wake + VT-switch both guarantee recovery
`wp_tearing_control_v1`	❌	❌	✅	✅	Upstream-blocked: smithay's `DrmCompositor` exposes no tearing / async page-flip (`FrameFlags` has no tearing variant) and the `wp_tearing_control` bindings aren't in the pinned wayland-protocols. Advertising it would be a no-op lie to clients
`wp_drm_lease_device_v1`	✅	✅	✅	✅	Shipped (2026-07-18). The blocker was resolved exactly as prescribed below: `MargoState` now holds an `Rc` handle into the udev `BackendData`, so `lease_request` reaches the live `DrmDevice` synchronously. Non-desktop connectors (VR headsets) are withheld from the output list and offered for lease; lease grants connector + free CRTC + primary plane; suspend/resume follows VT switches

margo now ships output_power (niri still lacks it) and, since 2026-07-18, drm_lease — niri no longer has any protocol margo lacks. The only gap left anywhere is tearing (upstream-blocked; Hyprland and mango have it). None of this blocks daily-driver use.

Compositor-specific protocols (informational, not gaps)#

Tied to each project's own ecosystem; not counted as margo gaps.

Protocol	Who	What it does
`focus_grab`	Hyprland	Internal grab semantics
`hyprland_lock_notify`	Hyprland	New 0.55.0-86 — notifies clients when the session locks/unlocks (Hyprland-flavoured; not the standard `ext_lock_notify_v1`)
`global_shortcuts`	Hyprland	xdg-desktop-portal global-shortcuts helper
`toplevel_export` / `toplevel_mapping`	Hyprland	Hyprland-flavored toplevel export
`hyprland_surface`	Hyprland	Rounding / opacity surface hints
`CTM_control`	Hyprland	Color-transform-matrix (pre-`wp_color_management`)
`mesa_drm`	Hyprland	Compat shim for older mesa
`background_effect`	niri and Hyprland	Surface blur / effect extension
`mutter_x11_interop`	niri	XWayland tweak inherited from mutter
`xdg_output_v1` (legacy)	Hyprland and mango	Pre-`wl_output v4` HiDPI workaround
`zwlr_export_dmabuf_manager_v1`	mango only	Legacy wlroots screencast (superseded by image-copy-capture)

Unblocking the remaining two#

zwlr_output_power_management_v1 shipped in 1.0.2 (DPMS via the recoverable request_dpms deferred-queue path). The two that remain each need a prerequisite, not just a coding session:

wp_tearing_control_v1 — wait for smithay's DrmCompositor to expose tearing / async page-flip (FrameFlags variant). Until then any advertisement is a no-op. Track upstream smithay.
wp_drm_lease_device_v1 — needs MargoState to reach the udev DrmDevice synchronously (a small backend-access change to the deliberate State/BackendData split). Lowest value (VR / leased connectors), so lowest priority.

See road_map.md §15.10 for the full work log and blocker analysis.