Relay host/join/refresh/sign-in could hang forever: none of the UGS
Sessions 2.2.4 calls accept a CancellationToken, so a stuck op left _busy
true and locked the whole browser UI with no recovery.
Wrap every hangable call in a Timed() helper (Task.WhenAny + Task.Delay,
the only option the API allows) driven by a new connectTimeoutSeconds
inspector field (default 25s, 0 = wait forever). On timeout the status is
reported, _busy is cleared and the event-driven UI recovers immediately.
Guard the two ways a late/partial op can corrupt shared state:
- LeaveIfDisplaced: an aborted CreateSessionAsync keeps running and may
materialise a ghost host seconds later; leave it, but only when the user
has no tracked session (_session == null), else it would tear down the
shared NetworkManager and kill a fresh host.
- Partial-start shutdown: only for connection-owning ops (host/join), so a
refresh timeout during hosting can't shut down the user's own host.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Introduce PersistentPlayerId: a stable id that survives an app restart and is
unique per install (built on the UGS install-path-namespaced anonymous
PlayerId, with a local PlayerPrefs GUID fallback for loopback). Clones on the
same machine stay distinct.
Each client reports its id to the server on connect via a new Match_Identity
named message (mirroring Match_ReportHit, so no ConnectionApproval timing
risk); the host records its own directly. The server keeps an ephemeral
clientId -> persistent id map, exposed via PersistentIdOf(clientId) as the
foundation for later reconnect recognition. Map entries are cleared on
disconnect and net stop.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The freshly converted hunter's body was hidden for the settle window via a
WaitForSeconds coroutine. If the object was disabled or the converting owner
disconnected mid-settle, the coroutine was stranded and _renderGate stayed
true forever, leaving the avatar invisible on every remote.
Replace it with a frame-driven Time.time deadline (_renderGateUntil) lifted
in UpdateVisibility, which runs every frame the component is active and can
never get stuck. Remove the coroutine and the now-unused System.Collections
using. Visible behaviour in the normal case is unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The FreeCamera's CharacterController stayed enabled permanently, so its
invisible capsule sat in the play area at (0,1,0) and blocked players even
when the freecam was not in use: the controller script disables only itself
in Awake, not the independent CharacterController collider.
Keep the capsule out of the physics world unless the freecam is actually
flying: disable it in Awake and toggle it with SetActive (enabled before the
teleport in SyncToCurrentView, which preserves the current enabled state).
Also ignore collision against all live players instead of just the first on
each activation, since Physics.IgnoreCollision is reset when the capsule is
toggled off and back on.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The orbit camera suppressed wheel zoom whenever Shift/Ctrl was held, to
reserve Shift/Ctrl+wheel for the paint brush resize. But Shift is also
the sprint key, so sprinting blocked zoom outside paint mode.
Gate the reservation on PanEnabled (set only by PaintMode.Enter/Exit), so
the wheel is only claimed for brush resize while actually in paint mode;
otherwise it zooms normally.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
MapPropIds (in MeshPainter.Backend.cs) referenced the *Id fields from the
MeshPainter.cs partial in its field initializer. Static field-initializer
order across partial-class files is unspecified in C#; after the script
reorg the composite array initialized before the *Id fields and captured 0
for every property id. HasProperty(0) is always false, so AssignMapTexture
bound no maps at all -> the shader fell back to its white default textures,
making every fresh avatar fully glossy (smoothness 1) and emissive.
Build MapPropIds lazily on first access (runtime, after all static
initializers) so the file order no longer matters. The per-stroke allocation
savings are preserved; only a single null-check per call is added.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Move scripts under Assets/Scripts/ into Components, GameMode, Helper,
Network, Enum, Struct and ScriptableObject folders. Pure file moves
(plus folder .meta files); no code changes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Trim restated narration, dead commented-out code, and oversized class
headers; keep all why-comments (authority, timing, edge cases), tooltips,
and serialization caveats.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Extract byte-identical duplicated code into shared helpers (behaviour-neutral):
- New InputUtil.cs: single source for mouse-delta/scroll/cursor-lock, modifier
and mouse-button reads, previously copied verbatim across camera controllers,
MeshPainter, FirstPersonMode and PlayerCharacterController.
- GetMatFloat/SetMatFloat replace the 12x material-instance + HasProperty guard
pattern for the camo/height float properties.
- Route two inline button hooks through the existing Hook() helper.
Bespoke variants deliberately kept local (FreeCam SprintHeld = LeftShift only,
MeshPainter.TryGetMouse with out-params).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Force the local avatar out of Paint mode and block the switch key from
re-entering it while the match is in its resolution (reveal) window.
Painting is off for the local player regardless of role. No restore
needed: the match ends by despawning/respawning the avatar, rebuilding
the mode cycle at index 0.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Select the whole pre-filled painting name one frame after activating the
input field so the first keystroke replaces it. Done explicitly so it works
regardless of the field's authored "On Focus Select All" toggle.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Ramp the along-surface climb velocity via a gentle acceleration instead of
snapping to full speed, so a brief tap nudges only a little. Sprint scales
the accel so it stays snappy; releasing decelerates at the ground decel rate.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the fixed additive brush-size step with a fraction of the current
size, so small brushes resize a texel at a time (fine control) while large
ones jump in bigger increments. Always moves at least one texel per notch.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
While a match is in the resolution sub-phase, hiders that have not
been found hold their current pose and position: the pose graph is
frozen (mixer speed 0, current frame re-asserted) and owner input is
disabled while gravity/grounding keep them planted.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The pipette now reads the height (occlusion) value from the own
paint mesh and enables the height toggle; picking anything else
disables the toggle while leaving value and alpha untouched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Brush radius now derives from the mesh-wide average UV density instead of
the hit triangle's own density. Per-triangle density varied up to ~7.6x
across the mesh (spiking at UV seams and stretched islands), so the same
brushSize painted a visibly larger patch on seam triangles than on
evenly-mapped ones. A single global scale keeps the brush the same physical
size everywhere; the stamp remains a 3D box, so seams are still bridged.
Also decouple the hover preview opacity from the brush albedo alpha: the
preview now stays fully opaque and is scaled only by the Preview Opacity
property, since the ring is a size/shape indicator, not a paint-strength
preview.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A slab that spans a concave silhouette (e.g. a table side = two legs
plus a top rail, all coplanar and thin) produced one convex hull that
filled the open gap between the legs, so a player standing between them
counted as embedded. Add a refinement pass: each slab whose in-plane
coverage grid shows its convex-hull footprint is less than a
configurable fraction (hullFillThreshold, default 0.6) covered by real
surface is recursively halved along its larger PCA axis until every
piece hugs actual geometry. Solid boards (the shelf) are ~fully covered
and pass through untouched. Verified over a 9600-point grid: zero
bridging (no collider interior lies >6cm from the source surface),
shelf output unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Generate multiple convex collider proxies from a mesh by growing
triangles into thin slabs (PCA minor-axis thickness gate) and hulling
each slab, so shelves and open compartments stay hollow while boards
count as embedding surfaces. Baked meshes get readable, deterministic
asset names (scene/prefab path + hierarchy + collision index) instead
of GUIDs. ConcealmentMeter now also excludes a furniture's own concave
mesh collider when a ConcealmentColliderBaker sits in its parent,
symmetric to the existing ConcealmentVolume rule.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
SkinnedMeshRenderer.BakeMesh(mesh) bakes the renderer's world scale
into the vertices. SkinnedColliderBaker already accounts for this and
divides it back out, but ConcealmentMeter did not: it lifted the
already-scaled verts to world via localToWorldMatrix (scale again),
yielding scale-squared voxel positions. This threw off the embedded
broadphase box, the EmbeddedPercent computation (which now drives the
X-ray reveal) and the debug gizmos whenever PlayerScaler != 1.
Rescale the baked mesh to unit scale right after baking (no-op at
scale 1), matching the SkinnedColliderBaker convention documented in
PlayerScaler, so world = localToWorldMatrix holds on unit geometry.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a second, additive HDR emission color (_XRayColor) that reveals a
player through walls as a pulsing energy glow, driven during matches:
- Shader: XRayOverlay pass (ZTest Greater, additive) for occluded parts
plus a matching additive term in the forward fragment for visible
parts, so the glow shows everywhere and never double-blends. Exposes
_XRayColor, _XRayPulseSpeed and _XRayPulseStrength for a breathing pulse.
- Render it after opaques via a RenderObjects feature (custom XRayReveal
light-mode tag) on both PC and Mobile renderers, so the through-wall
test is depth-sort independent and no longer drops out at some camera
angles/distances.
- MeshPainter drives _XRayColor alpha through a framerate-independent
fade (xrayFadeDuration) rather than a hard toggle.
- Reveal is driven per avatar on every client: all remaining un-found
hiders glow during the Resolution phase, and a hider whose
ConcealmentMeter EmbeddedPercent exceeds MatchManager.embeddedRevealPercent
glows via an owner-authoritative NetworkVariable.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The pipette now samples only what is painted (albedo, metallic,
smoothness, emission), not how strongly - the albedo alpha,
metallic/smoothness opacity and emission alpha are preserved instead
of being forced to 1 in every pick path and in FinalizePick.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Wire three material-backed look sliders (occlusion AO shift, relief normal
intensity, parallax intensity) through the paint HUD, its layout/state
persistence, the .paint slot format (bumped to v4), owner-authoritative
network sync and the runtime reset, mirroring the existing camo-knob path so
they save, restore, replicate and reset alongside the brush values.
Change the occlusion AO mapping from a division threshold to a linear shift:
AO fades from 1 down to (1 - shift) at the deepest painted hole, so a 0 shift
no longer leaves black spots. Enable _NORMALMAP alongside _PARALLAXMAP so URP
transforms the painted relief normal into world space (otherwise it discards
normalTS and the relief is invisible), and pin the added keyword in the build
keep-alive material.
Stop the pipette from picking the occlusion/height value and turn its toggle
off after a pick. Default the knobs to 0.2 shift / 1.4 normal / 1.0 parallax
across the shader, painter, network and HUD fallbacks.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Reuse the existing occlusion paint channel (no new map, MAP_COUNT stays 3):
0.5 is neutral, above 0.5 raises a hill, below 0.5 sinks a hole and also
fades ambient occlusion (saturate(2*occ)). Bind the MSO map to _ParallaxMap,
enable _PARALLAXMAP, derive a relief normal from the height gradient, and
reseed unpainted occlusion to 0.5 so untouched surfaces stay visually neutral.
Pin the _PARALLAXMAP runtime variant in the build keep-alive material.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Load now applies the selected slot immediately instead of going through a
confirmation dialog. Save keeps its OK button enabled even when the name
already exists (the warning text stays, purely informational) and writes
through with overwrite, so re-saving under the same name replaces the slot.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The cutout cone's radius and feather are now multiplied by the hero target's
uniform world scale (lossyScale.x, the same body-scale source the orbit input
uses). The authored values are the scale-1 sizes, so the see-through hole and
its soft edge keep the same on-screen proportion for a small hider or a large
hunter, and a mid-match conversion tracks live since lossyScale is read every
frame.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Destroy every GameObject tagged "PreMatch" locally when the match enters
the Seek sub-phase (hooked in ApplyPhaseLocal, runs on every machine; the
objects are plain scene geometry with no NetworkObject, so each client
cleans up its own copies). Block hunters from starting a wall- or
ceiling-climb during the Hide sub-phase via a WallClimbBlockedNow() gate
in UpdateWallClimbState; an in-progress climb is exited cleanly.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two paint-mode pan refinements:
- The max-distance clamp is back to body-relative only (CurrentScale, no zoom
factor), so the world-space pan limit is identical at every zoom level;
only the sensitivity stays zoom-scaled (finer when zoomed in).
- Don't zero the pan offset in OnDisable. The only thing that disables this
component is FreeCamMode taking over, and the freecam snaps to the live
(still panned) view. Zeroing here jumped the orbit camera the brain blends
FROM to center while the freecam sat at the panned pose - the visible
"snap to 0, then translate to the offset" on the paint->free-cam switch.
Leaving the offset frozen keeps blend-from == blend-to (both panned) for a
seamless handoff; the pan glides home on its own once the orbit input next
runs in a non-paint mode.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The save-dialog name field is a focused TMP_InputField, but global discrete
hotkeys kept reading the keyboard every frame, so typing a name like
"Checkers" cycled the pose (C) and shut the server down (K). Add a shared
TypingGuard.IsTyping check (focused TMP/legacy InputField) and short-circuit
the global handlers on it: pose cycle, network dev keys, mode switch, and
paint undo/redo. Movement was already gated out via ControlEnabled.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The middle-mouse pan now persists when the button is released and only glides back to center on leaving paint mode. Pan sensitivity and max distance scale with the orbit zoom distance, so panning stays consistent on screen at any zoom and unchanged at the default zoom.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Save and restore the emission intensity and the three camo knobs
(half-Lambert, half-Lambert brightness, specular dampen) with a named
painting. The .paint format moves to v3 with a small scalar block
between the header and the pixel maps; v2 paintings still load with
their painter values left untouched. Loading re-syncs the emission and
camo sliders, and peers pick the values up for free through the
existing owner-authoritative paint NetworkVariables.
Add a HUD reset (Lib_ResetHud button plus context-menu fallback) that
behaves as if the saved hud-layout file were deleted: it removes the
file, restores panels to their authored prefab positions, and resets
the brush values to the painter/material defaults captured before any
carry/restore runs.
Remove the alpha bar from HsvColorPicker so alpha is driven by
standalone Slider_AlbedoAlpha / Slider_EmissionAlpha bindings; the
picker now writes RGB only and the sliders write alpha only, with no
feedback between them. Add a generic SliderCurve component that reshapes
a slider's handle position into a non-linear value (with an inverse for
setting values back), and make the slider wiring curve-aware.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fix the in-place Conversion respawn snapping to the origin (0,0,0),
which surfaced when a hider was tagged mid wall-climb. The owner holds
authority over the NetworkTransform, so the server's Instantiate
position is ignored on the owner. Deliver the tagged world position
through a replicated _spawnOverride the server writes in OnNetworkSpawn,
and have the owner re-assert it before PlayerCharacterController's first
Update snapshots the kill-plane rescue target. This replaces the fragile
SpawnedMidSeek phase check.
Also hide a freshly converted hunter on other clients for a short
settle window (conversionRevealDelay) so nobody sees it pop in or
interpolate from the spawn frame; the shot hider is already despawned,
then the hunter cleanly materializes in its place.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Introduce a second match mode alongside Classic. In Conversion mode
hiders are invisible to each other (and to hunters during Hide), and a
hunter hit converts the hider into a full hunter in place instead of
sending them to the spectator freecam. The mode is replicated to all
clients through the existing phase named-message byte; visibility is
computed locally per client per mode.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Drive the match through three server-authoritative phases (hide, seek,
resolution) with per-phase durations, broadcasting the active phase and
countdown to clients over the existing named-message infrastructure so
MatchManager stays a plain MonoBehaviour and Main keeps zero in-scene
network objects.
During hide, hider meshes are culled on hunter clients only (each client
computes visibility locally from role, phase and found state). During
seek, hunters shoot a single raycast from the first-person camera centre
(left click, cooldown-gated); a validated hit marks the hider found,
which hides its mesh for everyone and binds the owner to spectator
freecam. Seek ends early once every hider is found, then resolution runs
before returning to the lobby.
Force AlwaysAnimate on remote replicas so the animator keeps writing bone
transforms while the body is hidden or off-screen. The Paintman animator
is authored CullUpdateTransforms, so a culled renderer froze the pose on
the rest pose and the skinned collider baked that frozen pose, landing
replicated paint stamps on the wrong body location and leaving the
painting desynced after hiders were shown again.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Choose the spawn area by player role: Hunter avatars look for a
HunterSpawnArea-tagged collider, Hiders for HiderSpawnArea. When the
role-specific area is missing, fall back to the shared SpawnArea tag, and
when that is absent too, keep the existing client-id ring placement.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
PickFromForeignMaterial read rend.sharedMaterial, which is always slot 0, so any renderer with several materials (one per submesh) always sampled the first material's values. Resolve the submesh from the hit triangle (collider triangleIndex, or the recovered render-mesh triangle) and read sharedMaterials[submesh] (clamped) instead. Single-material renderers stay on slot 0, byte-identical to before.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Replace the multi-sample silhouette probe with one camera->target raycast: any world geometry in the line of sight arms the cut, a clear line keeps occluders solid. The shader carves a world-space cone (apex at the camera, radius growing toward the target) so the screen-space hole stays a constant size regardless of occluder depth, with a soft radial fade plus a symmetric axial fade around the target so geometry just behind the player feathers out instead of hard-cutting along a plane. Anything well past the target stays solid. Also suppresses the deoccluder pull-in so the camera holds its orbit distance and the cut takes over.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a geometry-based gate in PaintAtScreen that estimates how much of the
brush footprint lands on faces whose normals deviate sharply from the hit
surface. When that fraction exceeds the threshold the whole dab is
suppressed (no draw, emit, or undo mark) instead of smearing around the
edge. Backend-independent (no GPU readback) and winding-robust. Tunable via
cornerGuard / cornerGuardAngle / cornerGuardMaxFraction.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Move DefaultNetworkPrefabs.asset out of Assets/Settings back to the Assets root (where NGO expects it), and fix the PlaySceneWindow domain-reload toggle to read and write EnterPlayModeOptions correctly.
Surface the PlayerCamoLit material knobs (HalfLambert, HalfLambert brightness, specular dampen) as material-backed properties on MeshPainter, guarded by an IsReady gate since the HUD binds before the material instance exists. Wire the three HUD sliders to them and seed the sliders from the live material once the painter is ready. Persist the values into the same HUD JSON as the brush settings (with a hasCamoValues flag so pre-camo save files never clobber the authored material) and carry them across a scene-switch respawn. Replicate them owner-authoritatively via a NetworkVariable in PlayerPaintSync, exactly like emission intensity, so remotes and late joiners show the same camo.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The hybrid foreign-object pipette previously always overwrote albedo with the lit on-screen pixel. Gate that behind a new serialized pipetteAlbedoFromScreen flag (default off): off reads albedo texel-exact from the material like the other channels, on takes it from the rendered, fully-lit pixel. The data channels still always come from the material.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Move editor scripts to Assets/Scripts/Editor (PlaySceneWindow, SceneCollection),
move DefaultNetworkPrefabs and the Main Camera custom blends asset into
Assets/Settings, and shelve the unused HeroCutout shader/include under
Assets/Unused. Rename the Chameleon background texture, delete the obsolete
DefaultParentForEditor script and New Lighting Settings asset, add a The Room
SceneCollection asset, and drop the redundant Assets menu entry from the
StraightenKnees tool.
On a foreign object the pipette now reads the data channels
(metallic/smoothness/occlusion/emission) texel-exact from the material
while taking albedo from the rendered, fully-lit on-screen pixel - so the
picked colour honours baked lightmaps/ambient/SSAO while keeping the
material's PBR fidelity. Falls back to a pure screen pick when no readable
UV is available. Pull the render+readback out of PickFromScreen into a
shared TrySampleScreenColor.
The pipette now also arms on Ctrl (in addition to Shift), and Ctrl+wheel
resizes the brush; the orbit camera ignores the wheel while either modifier
is held so it does not zoom and resize at once.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
When leaving first person for an orbit mode (Paint / Player Control), snap
the orbit axes to Camera.main's current view direction instead of restoring
the orbit's last-used angle. With WorldSpace binding + Sphere orbit + a
centered composer the camera look rotation is exactly
Euler(VerticalAxis, HorizontalAxis, 0), so the inverse mapping is exact.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The orbit radius (min/max/default and per-notch step) was in absolute world
units and ignored the player's localScale, so a small Hider started too far
out and a large Hunter too close and could not zoom out far enough. Multiply
range, step and default radius by the follow target's lossyScale (PlayerScaler
keeps the body uniformly scaled), making the inspector fields size-1 design
values that frame every role body-relative. Re-clamp every frame so a runtime
grow/shrink stays in range. GameModeManager.BindLocalPlayer calls ResetZoom
after rebinding so each new avatar starts at its scaled default zoom.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
admin
