24 Commits
Author SHA1 Message Date
MichaelScottsman 8b422d1cd6 AssPod in + > Launch Program: the installer adds it once VLC is there, and it opens the assPod with the theme on or off
lgs pod: with the theme on, the bar's asspod.open; with it off, only the assPod's runtime modules
in Steam (rt.podOnly, no liveness watch on the CSS core) and lgs-shell in pod mode (no theme, the
assPod's page script only), which exits once the assPod is away with nothing playing. lgs on
replaces a pod unit with a full one.
2026-10-10 14:37:53 -04:00
MichaelScottsman 80d4710f3f assPod: Nano Gen 2 body, chosen on the main menu and kept across launches
- assPod Model on the main menu switches between Video Gen 1 and Nano Gen 2; Color lists the
  current body's finishes (the nano: silver, green, blue, pink, black, (PRODUCT)RED). The choice is
  kept in ~/.config/glass-shell/asspod.json beside the glass tune.
- The nano: a procedural 2nd-generation model (tools/nano2-model) exported as render models, drawn
  by podd with anodized aluminum (tinted, bead-blasted, the laser-etched back paler), white end
  caps and a clickable wheel. Same menus on a 240 x 180 logical screen, so its text is larger, as on
  the real one; it plays no video. Shown 52.5 mm wide.
- podd: one box layout that holds either body, so a switch needs no restart; 64 px between atlas
  tiles and a 4 px clear band around each eye's image, which removes the hairline rectangle around
  the body (filtering at the stereo split took the other eye's edge); texture dump and debug modes.
- The left stick acts like the right: Steam's d-pad directions are no longer relayed (tilting the
  left stick down or left pressed the centre button or Menu).
2026-10-10 14:08:31 -04:00
MichaelScottsman 4fc0cfffd6 assPod: a music and video player in the hand, on VLC
A music note button on the dashboard bar, left of the paintbrush, opens the
assPod in the hand whose laser pressed it. VLC plays its media library's music
and videos in the background; both sticks turn the click wheel, A and stick
clicks press the centre, B is Menu and the hamburger button puts it away. The
rest of the UI fades out while it is open.

- device/asspod: the pod's OS, screens, VLC engine and scene graph (bundled by
  tools/build_asspod.py into device/vr/systemui.asspod.js), with the model
  exported from tools/asspod-model
- native/podd: renders the 3D body with real shaders into a box of stereo
  portal panels, and the playing video's picture under the screen
- device/shell_ext/asspod.py: VLC, the library scan, art, podd, and video
  frames from ffmpeg paced to VLC's clock; videos keep their place across
  restarts
- Reposition and Black/White finish on the main menu
- install.sh offers to install VLC when it is missing
2026-10-09 22:27:35 -04:00
MichaelScottsmanandClaude Opus 5.5 75a6e6c246 Fix Steam renderer crash when LiquidAss is launched from +
A top-level @starting-style on the light layer's ::after (registered
custom properties under the @container style() gate) SEGVs Steam's
Chromium 126 renderer when the theme is injected while the pointer rests
on a hovered element. Launching LiquidAss from + > Launch Program does
exactly that: the laser is still on the row when the stylesheet lands,
so every launch crashed the whole gamepad UI (reproduced 3/3 with real
clicks; CSS-only reproduces too; gone without the rule).

The ramp in from dark is now a keyframe animation (lgs-ill-in) scoped to
the illumination hosts' in-rule, so it no longer reaches Steam's own
::after either.

Home's gamepad focus and blur handlers return when the module is torn
down, so turning the theme off no longer throws in 41-home.js.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-09 17:21:16 -04:00
MichaelScottsman 056f69d418 Rename the app to LiquidAss: launcher entry, toasts and icon; README updated
Install paths and internal identifiers keep the glass-shell name so existing installs carry over.
Steam's program row is matched by the new name, with the old one still accepted.
2026-10-09 15:57:11 -04:00
MichaelScottsman c6427528e0 Pointer ring off by default: the glass sits behind Steam's panel, so SteamVR's own laser dot shows and ours doubled it 2026-10-09 15:15:30 -04:00
MichaelScottsman af4e73c1d3 Stereo glass: each eye sees the room behind the window glass from where it is; tuner fits its popup; the sidebar keeps native glass without the laser; Home leaves on the route exit
Stereo (glassd cap stereo, flag glassStereo, on)
- The window glass was one image both eyes saw on the glass plane, so the room behind it read as a picture
  pasted on the glass. glassd now renders each eye's view (SteamVR's eye-to-head offsets, refreshed every
  5 s): every region of a stereo surface's texture is a pair, left eye's copy then right eye's, and the
  scene graph's panel spans the pair with stereoscopy Parallel.
- Measured on the device with a lab probe (lgs_sg test.stereoProbe): SteamVR splits a Parallel panel at
  the middle of the panel's own uv rect, not at the texture's halves; hence the pair layout, and stereo only
  where the scene graph shows a surface whole (main, SteamVR's panels). The bar, the frame menu and popups
  show crops and stay mono. A slab clipped sideways is left out while it is.
- GPU about 2.9 -> 3.5 ms a frame.

Sidebar without the laser
- frame.menu is no longer laser-only: SteamVR shows it in gamepad mode (laser mouse off) and beside its own
  pages; keyed on the gamepad flag its native glass went off and the CSS glass showed. Verified with the
  flag forced to gamepad on Home and in VR Settings: the cover stays.

Home -> Library
- The shell stamps the leaving Home view with the moment it starts leaving (only on a way out of Home, kept
  across Steam's /library redirect); Home fades on the route exit and the reporter lets its plates go on
  that clock. Measured: all Home discs reported leaving in the first report, gone with the page.
- Plates whose element leaves the DOM are reported leaving from the mutation itself (no measuring), not
  after a frame-driven tick the new route's render starves.

Glass tuner
- The popup fits Steam's 600 px bar popup cap (rows 76 px: label and value on one line over a 40 px slider);
  open()/close() for tests.
2026-10-09 14:51:41 -04:00
MichaelScottsman 5af91fc1bd Glass leaves with its page; the sidebar keeps native glass in VR Settings; Friends clears the search row; a paintbrush on the bar tunes the glass
Route changes
- Plates fade with their content: the reporter reads a running opacity fade on a plate or its ancestors
  (Home -> Friends: the route's 150 ms exit) and reports exitAt/exitMs; glassd fades the plate's coverage
  linearly on that clock, optics unchanged (cap exitAt); the daemon keeps a leaving plate for exactly
  that fade. A plate whose element left the DOM (Home -> Library stays in one /library route: no fade,
  Home unmounts when Library has rendered) goes in 90 ms instead of the 350 ms dematerialize ramp that
  left Home's discs over the Library grid.
- The window cover arrives with the route: the shell sets the new route's glass mode in the history
  change itself, the reporter reports the mode flip at once (tickNow) with the time it happened, and
  glassd springs a kept surface's new cover from then (appearAt now also for main gaining its cover).
- A disc popped at the switch belongs to the page that leaves: it never flips the new page to the crops
  layout and goes with the page's fade (lgs_sg.js routeOut).

Sidebar in VR Settings
- The frame menu is a laser-only surface keyed on Steam's gamepad-nav flag, which goes stale on
  SteamVR's own pages (the laser works SteamVR's page): the native glass went on and off with every
  input Steam saw, its CSS glass in between. On a SteamVR page (the frame's VR Settings action active)
  the rule does not apply.

Friends
- The page starts at y 108 (C1a's pad + 40, as Settings' list): the sidebar platter no longer sits
  under the search capsule.

Glass tuner (flag glassTune, on)
- A paintbrush circle after the system capsule (its own glass piece and native cover) opens a bar popup
  through Steam's own bar popup button: colour (8 swatches) and tint strength, Glass intensity (the
  dial), Refraction, Frost, Highlights, Reset to default.
- Daemon action glass.tune (kind ui, its own 40 ms rate, outside the global window): glassd.json at once,
  ~/.config/glass-shell/tune.json 400 ms after the last change (kept across reboots by request; the
  defaults delete it; RT-6 allows exactly this file), the dial file mirrored for the SteamVR pages.
- glassd: tune {refract, frost, light, hue, hueK} applied in drawGlass, eased (40 ms) like the dial now
  is; the glass body's neutral moves toward the hue (uHue), per-piece tints stay on top (cap tune).
- The CSS glass follows live: a style per Steam window sets --lgs-dial / --lgs-tune-hue; the material
  tints mix the colour in.
2026-10-09 14:06:10 -04:00
MichaelScottsman 27edbab0f3 Home icons keep their tone on hover; the sidebar resizes with its glass when VR Settings opens
- Home discs are never occluders: a plate under a pop was its shadow (glassd x.55, no light, in one
  frame), so the hovered disc and the neighbours its name plate or card overlaps snapped dark on every
  hover. The pop keeps its own contact shadow.
- The frame menu's tab pitch (and place) are registered lengths that transition: VR Settings' shorter
  panel takes a smaller pitch, which changed the bar in one frame while glassd's shape followed a report
  (~120 ms) later. Now the bar resizes on the spring the reporter morphs the glass on (shrink fade, grow
  snappy); measured: Steam's crop moves over ~250 ms and glassd gets the morph from the first report.
2026-10-09 12:46:18 -04:00
MichaelScottsman ca78f26da9 Launcher: label fit cached by name and column, tagged while the popup is hidden: + opens ~200 -> ~135 ms
Steam renders new cells on every open, so the per-element cache missed and all 23 labels were measured again
(a forced layout each) inside the open. The verdict is now kept by label and column; tagging runs as the
pooled popup's content mounts, and the show itself only checks after its first frame. Stock: ~105 ms.
2026-10-09 02:12:51 -04:00
MichaelScottsman e1d7318559 Light layer only while lit: a third less style work per route change
The illumination ::after (04-states.css) painted nothing at rest but was computed for every host (46-93 per route,
~25 custom properties and six gradients each) at every style recalc. The generated host entries (fontkit.py --hooks)
now carry a :where() gate (hover, active, focus, focus-within, pressed, focus-in, .lgs-lit): the layer exists only
while lit. rt/06-states.js holds .lgs-lit for 600 ms after hover, focus or press ends, so the 441 ms fade-out still
plays; a top-level @starting-style (02-motion.nowrap.css) ramps the light in from dark. Verified: ramps unchanged
(press in 0/.66/.98/1, out .86/.45/.03), no layer at rest; route changes ~48 -> ~55 fps.
2026-10-09 01:52:07 -04:00
MichaelScottsman 8862d7174a Native plates are glassd's from their first frame; Settings tick reads no layout
- reporter: <html data-lgs-plates> on managed native windows; a plate glassd has not drawn for 1.5 s gets
  data-lgs-plate-css (CSS plate glass back). The theme drops CSS plate glass on every other plate at once,
  instead of after the ack: no CSS-then-native swap, no fill/shadow transition on each Home disc.
- settings: scrolled state from the scroll listener, the hero title measured in rAF (the 300 ms tick
  forced the page's layout mid-transition)
2026-10-09 01:29:34 -04:00
MichaelScottsman a757782a0e Route changes: compositor-only materialize keyframes, no nested Home grid fade, no forced layouts from the ornament and library scans
- lgs-mat-glass-in/out, lgs-toast-in: scale and opacity only (glassd materializes the glass; the fill,
  blur and shadow ramps re-blurred every animating element per frame on Steam's GPU)
- Home: the first grid arrives with the route's own fade (is-entering only for section/page changes)
- shell: the ornament is measured in the frame pass and only when its legends change
- library: section headers checked by style (checkVisibility), not a rect read
2026-10-09 01:02:03 -04:00
MichaelScottsman acab337eab Performance: Steam UI thread (route changes 42.5 -> ~52.6 fps; stock 58)
- 01-font: the every-element :where(:not(code, ..., code *, ...)) rule is two
  plain rules (body *, plus a monospace restore for code/pre/kbd/samp/tt):
  the :not(X *) ancestor walks ran for every element on every recalc.
- Reporter: measures after the frame (rAF -> task), not in its rAF callback;
  transitions/animations away from measured elements no longer trigger a full
  recompute nor flush the style-fact cache; plates a route inserts are marked
  glassd's at once without measuring (no forced layout of the half-built page),
  a glass-mode change marks pending and measures after the paint.
- 41-home: the top row's pill layout effect has dependencies (it read layout in
  every Home commit, ~450 ms over six route changes).
- 40-library: the sort pill's visibility via checkVisibility (no layout).
- 60-settings: the hero title width is measured when the title changes, not
  every 300 ms.
2026-10-08 23:53:33 -04:00
MichaelScottsman 9becb5a5c3 Native-only glass: no CSS-to-native swaps; popups armed ahead; menus cut the page out
Phase 1 (no handoff):
- lgs-native on every Steam window while native mode is live (daemon NATIVE_JS
  `all`, lgs_core sets it at injection), pooled popups included.
- Covers and plates are marked glassd's as soon as they are reported (optimistic),
  CSS glass back only if never acked within 1.5 s.
- Covers/plates that appear while animating are reported at their end state
  (seekEnd over all running animations) with appear timing; glassd ramps them
  from the page's own start time (appearAt).
- Pooled popups stay in the report while hidden and are measured as their content
  mounts ("armed"): glassd draws the glass ahead, lgs_sg builds the nodes in the
  same scene-graph update Steam inserts its panel in (panel watch), the hidden
  cover elements are pre-marked, glassd clears a surface's texture on hide.
- Plates inserted by a route are reported synchronously, ahead of the scroll
  deferral.

Phase 2 (menus):
- While a menu, alert or sheet is open in native mode, the page under it is cut
  out of Steam's texture (data-lgs-hole mask, faded on a registered property),
  held through the slab's out-ramp; the slab is transparent glass from the open.
- Modal layers are reported at once (no 607 ms settle wait).
2026-10-08 23:33:29 -04:00
MichaelScottsman dc3071aac0 Smooth transitions: sidebar width morph, window glass materialize, launcher plate exit
- Frame menu: measured expanded width so Steam's width transition animates;
  reporter sends cover.morph end shape + start time, glassd morphs the glass
  on the same spring (70 ms native lead).
- Main window: glass-mode change reported from the mutation callback and the
  cover marked data-lgs-cover="pending" (no opaque CSS window); glassd springs
  covers in/out; daemon keeps lgs-native 1.5 s between modes.
- Launcher: the name plate dissolves in place, then the label fades back.
2026-10-08 23:04:35 -04:00
MichaelScottsman 0267a9d13b Launcher: replace :has/nth-child row-mate rules with runtime tags; + menu opens in ~250 ms (was ~1 s) 2026-10-07 23:10:21 -04:00
MichaelScottsman 503017b154 Ship: native glass on by default, all phase 2 areas and real launches enabled (device/defaults.json) 2026-10-07 21:30:57 -04:00
MichaelScottsman 49286acda1 Checkpoint: R2 reviews of all 14 context packages, coordinator decisions, P10 and platform request handling 2026-10-07 17:43:20 -04:00
MichaelScottsman 769b4b2470 Checkpoint: R2 coordinator decisions (PLAN 1.19) and partial R2 reviews before usage-limit interruption 2026-10-07 15:15:00 -04:00
MichaelScottsman c01df80df4 Context round complete: all 14 context packages built to M3 (some M4) and report READY 2026-10-07 12:51:17 -04:00
MichaelScottsman ca69915d8f Platform packages P1-P10 built, reviewed and fixed (R1); P7 anchor fix (texture uv) and popup curvature; context round in progress 2026-10-07 12:42:46 -04:00
MichaelScottsman dd4f775645 WIP checkpoint: platform packages built (M3), core contexts C1a/C1c/C2a at M2, C4a at M3 2026-10-07 08:21:40 -04:00
MichaelScottsman 65ab8fc09b WIP checkpoint: phase 2 build round R1 interrupted by usage limit (contracts M1 for P1-P10, partial platform code, context M0 work) 2026-10-07 05:03:08 -04:00