79e2494fc62e9bc96c3ff3e01a126e8a0b8ac691
6
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  gromandClaude Opus 5
|
280c2e886f |
Planar water: the mirror works, and ladders in at waterQuality 3 (Ultra)
The starburst was three defects, and none of them was the oblique projection I blamed it on in the last commit. What found them was looking at the mirror's own render target instead of reasoning about the composited frame — the water reads it through a fresnel weight, a rippled UV, a plane-match band and an SSR march that paints over it, so from the outside a broken mirror and a broken water shader look identical. WOW_MIRROR_DUMP=1 now writes that target beside the capture. (Bevy's own save_to_disk cannot: try_into_dynamic knows four 8-bit formats and answers UnsupportedFormat for Rgba16Float, logging and writing nothing.) - The oblique row's scale. Replacing the near plane with the water plane costs the FAR plane: `z >= 0` stops meaning "nearer than CAM_FAR" and becomes `depth >= a*d`, a plane through the eye tilted by a. Lengyel's own scale puts that plane exactly through the frustum's far corner, and the sky dome is pinned just inside the far plane, so it straddled the boundary and was sliced along its own tessellation. Scale by the supremum of d/depth over the frustum instead, with 2% of margin. - The water surface lay ON its own clip plane. This was the starburst and it was fatal rather than untidy: depth now means "distance above the plane", so a fragment at d=0 reads z/w=1, the nearest value reverse-Z has, and beats every other surface in the frame. The mesh is wave-displaced, so its triangles straddled the plane instead of lying in it, and each was clipped to a sliver that kept the winning depth. SURFACE_CLEARANCE lifts the plane a yard: the swell sums to 0.34 and the probed height answers for the surface under the eye, not this fragment. - The mirror was stored upside down. A true reflection is what a camera on the far side sees with its up vector reflected too — an improper basis no Transform can hold. The honest right-handed camera sees the same surfaces through the same frustum and stores the image flipped, so the shader samples 1 - v; that costs no winding flip, so backface culling stays correct. It read as a lake reflecting the ridge the RIGHT way up. Measured: Lake Everstill 2.87 -> 3.36 ms mean over 600 frames (+17%), the Stormwind canal inside the noise, CPU +0.2 to +1.1 ms. It is the first lane in docs/GRAPHICS.md that spends main thread, so it is a rung of its own above High and the default stays 2. Seated as a setting, per the standing rule that no graphics lane hides behind an env var: waterQuality gains tier 3, the Ultra preset takes it, and the Advanced Graphics dropdown gains a fourth entry in both locales. With the lane off, both scenarios are byte-identical to the build before it — which took two findings of its own: - The system could not live in Update at all. `&mut Transform` on a camera is shared with everything that moves anything; declaring it put 59 undeclared orders into Update and tripped the schedule ratchet, which is what kept this behind an env var. It belongs in PostUpdate with the skybox and the billboards, after propagation and before visibility, where it also reads the eye's settled pose instead of a frame-old one. (The comment blaming ResMut<Assets<Image>> for those 59 was wrong — dropping the asset handle changed the count by zero.) - A system that runs and returns on its first line is still not free. Holding &mut Camera and &mut GlobalTransform for a frame is enough to change which undeclared orders the executor picks: the canal moved 0.75% of its pixels by up to 96/255 with the lane switched off. So the whole system sits behind a run condition rather than an early return — and the same effect killed a .before(UpdateDirectionalLightCascades) order that was correct in itself. PostUpdate's ambiguity ceiling goes 466 -> 495 with the reason (the pairs are structurally impossible: nothing else in the schedule names the mirror camera); Update's ratchets down 4973 -> 4952. Also restores scripts/gates.sh's `where` example, which had gone missing. Its absence was silent — cargo's error went to /dev/null and `|| true` made it read as "no install" — so the memo key and the player-tests gate saw none even when WOW_DATA was set. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SgstFWtn1RxEJ5kVrpgYwn |
||
|
Sam SungandClaude Opus 5.5
|
87a92eb295 |
Comments in benilla-world: one line of fact each, no history, no record numbers (#121)
Part of #41: the comments of `benilla-world`, Rust and WGSL, are rewritten to the rule in `docs/METHOD.md`, which is what the code does and the reference fact behind it, in a line. 24,226 comment lines become 8,863. - **The mechanical strip.** The citations that were only a private record number go first, and each edit is shown to be a pure removal. - **Wrong facts corrected.** The rewrite corrects the comments that stated wrong facts. Among them: - `0x6b62e0` is the WMO group's liquid renderer, not a doodad drawer. - The reference creates a WMO's props once, when a group enters its streaming window (`0x698720` → `0x695aa0`). It then admits them each frame the portal walk visits the group (`0x685d70` → `0x6838f0`). The old text had them created per visible group. - The liquid query's comments now state the reference's order: the map-object leg first, then the ADT (`0x69b6d0`). They no longer call the Outdoors/Inside split a match. - The interior probe runs from the character, not the camera. - The reference lights ground clutter. The old comment said it draws unlit. - The model-particle "pool order" deviation was wrong. Each instance is its own mesh, depth-sorted every frame. - The reference's vertical field of view follows the aspect ratio (44.1° at 16:9). The comment now gives it beside our fixed 45°. - The effect shader's far wall is the live `farclip`, 177 to 777 yd, not a fixed 777 yd. - Dozens more turned up on the way: renamed or deleted items, docs sitting on the wrong item, stale counts and links. - **Deviation labels.** "Deviation:" now marks only a choice with a reason. An unbuilt behaviour says so plainly. - **Verification.** Every file strips to identical code, and `cargo fmt --check` is clean. Five second readers compared every file old against new, restored the facts the rewrite had dropped, and checked each new citation against the reference. Defects the rewrite surfaced outside the comments are filed as their own issues (#106, #114, #115, #116). The comments here state what the code does. String literals are untouched; they are #92. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> |
||
|
SamandClaude Fable 5.1
|
9b6b84aea3 |
eighteen no-op writes gated, and the dead-gate finder
- eighteen per-frame systems now write only when their inputs moved: the equipment resolve, the skills, language and bind-location feeds, quest markers, shadows, the ambience seat and the weapon trail (2197) - a still remote mover writes no transform, so a city's idle crowd no longer re-propagates every model subtree each frame, and the WMO liquid latch no longer holds a city's visibility walk open (2197) - the bags are walked once a frame for every reagent and item slot instead of once per slot, and fourteen per-frame environment reads become one read each (2197) - the ffx blur chain is skipped where the combine multiplies it by zero, the audio pump compares before set_volume, and the QoS latch sits beside bevy's own exclusive system (2197) - WOW_RES_CENSUS counts on how many frames each resource read as changed, noisiest first — the finder that named these gates (2197) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
SamandClaude Fable 5.1
|
a5382d7b37 |
still frames skip settled work, and a guild can design its tabard
- on a still frame the world stops redoing settled work — each cut an exact gate, measured with the process pinned to the efficiency cores, where the main thread IS the frame: point lookups go through the water index, a terrain column tests its cell's four fan triangles, not the chunk's 256 (−1.3 ms), the flare march resolves its tile once, the static-world publish and light blob write only on change, billboards, emitter gates (off the new-portal edge), doodad-anim verdicts, the visibility walk, the unit publisher (its filter a named type), a rampless unit's alpha and the stable feed skip when nothing moved, the rig evaluator reuses its scratch, dormant emitters skip their step, child wiring visits pending emitters only, the shade map persists, and the terrain streamer skips its walk once settled; alone parked 25.4 → 21.6–22.8 ms of main thread, 32 → 36–38 fps there with nothing on screen (1979, 1982)
- the UI tick got the same treatment: a solo run's colour rides the material, not the mesh — the glow pulse's one `Assets<Mesh>` write a frame had armed bevy's asset-changed probes over ~44 k `Mesh3d` rows (−1.0 ms); script object ids index a dense map instead of a hash, where every scripted method call starts (−0.7 ms); the draw list is fingerprint-cached, walked every frame and sorted only when it changed (the list a named type); and the instrument that made the gates count at all: the first battery read flat until the probe printed how many frames each gate input had read as changed — `noisy=[cam:0,dbg:299,view:0,win:0,claim:0]` — one dev system took `&mut` on `DebugState` every frame to test a flag, so bevy marked it changed on 299 of 300 frames and silently disarmed every gate reading `debug.is_changed()`; it reads the flag first now, and the counters stay on the probe line (1982)
- two of the new gates were wrong the day they landed and are fixed in the same snapshot: the streamer's settled skip engaged the frame `complete` flipped, with the retained pass still baking the new cells, and froze the undrawn count `presentable()` reads above zero — the loading screen stayed up for good after a `.tele`; it now waits for a paced, presentable, complete world and keeps that count fresh; and the still-frame gates read the camera's `GlobalTransform`, which moves only after propagation — which runs after them — so a teleport frame carried every host's pre-snap verdict across the snap (a lamppost glow due to park kept its clock, over-bright after the `.tele`); the gates read the seat's own `Transform` write too (1982)
- a guild master can design the guild's tabard: the reference's own `TabardFrame.xml` joins the manifest (benilla never had the window), opened only from the wire — the click ladder's TabardDesigner arm now sends `MSG_TABARDVENDOR_ACTIVATE` and the Battlemaster arm its own hello (1861's two deviations retire); the `TabardModel` method table runs over five ints bounded by the `.rdata` table `{170, 17, 6, 17, 51}`, the design previews on the player's own body so the pane and the world show one look, and `Save` runs the fourteen pre-flight checks in the reference's order; the world frame is the reference's own `WorldFrame.xml` too — the framerate readout is live on CTRL-R — and the carve settled its kind: a `Frame` to Lua, a one-shot type, the sole tenant of stratum WORLD, and not "over the UI" when hovered, so world clicks and camera look stay live while its scripts fire (1977, 1983, 1984)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
|
||
|
SamandClaude Fable 5
|
838bc1fb0c |
the login theme rides the loading screen, and the mix gets a limiter
- the login theme no longer cuts off the instant the loading screen appears: it plays through the whole load and hands over on a 3 s fade once the world is ready, as the reference does — the stop sat on the same edge that raised the cover, and the reference never calls its music-stop on the enter-world path at all
- the mix had no headroom: every WoW sound is mastered to full scale, so five party-wide Fortitudes in one frame asked for 5× and clipped hard ("dirty, like a speaker breaking") — a stereo-linked look-ahead limiter now holds the output, a meter reports the peak the mix asked for, `SoundOutputLimiter 0` A/Bs it, and the 3D voice ceiling is named at 512 instead of inherited from the backend at 128
- a shaman's Rockbiter / Flametongue / Frostbrand / Windfury Weapon casts straight onto the equipped weapon on the press — no item cursor, no click — the reference's own `Attributes & 0x200` read; stones, oils and poisons still raise the cursor
- the nameplate category's player arm is pinned to the bytes (CanCooperate is faction-group-mask equality) and `/reaction` prints the one number it turns on: a GM-mode character plates as an enemy on a test realm because faction 35's mask is zero, exactly as 1.12 does
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
||
|
SamandClaude Fable 5
|
bdef9cd790 |
benilla-world: the engine becomes its own crate
- the world engine splits from the game into a benilla-world crate — 112 files, and the wall is ratcheted to zero names of the game inside the engine - the engine's public doorway is counted and gated at 32 items, and the render materials, shaders and asset cache move down to benilla-assets - the game's files leave the engine's asset root — game:// is its own space — and a spike binary boots the engine with no game attached - weather rides along: snowfall emission is denominated in the reference's frame cap, and a long teleport retires the stranded precipitation field instead of leaving it behind Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> |