Replace hand-rolled spatial canvas drag with Konva

The custom pointer-capture + frozen-camera drag math was fragile and
felt worse than before. Konva handles the drag gesture itself; the
"me" node's screen position stays pinned to the stage center via
dragBoundFunc + a worldLayer offset that recenters on every dragmove,
so the world pans under the avatar instead of the avatar drifting off
center. Grid, peers, and the audible-range circle all live in the same
world-space layer so they pan together.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

package-lock.json

@@ -16,6 +16,7 @@
         "@opentelemetry/sdk-logs": "^0.222.0",
         "@opentelemetry/sdk-node": "^0.222.0",
         "@opentelemetry/semantic-conventions": "^1.43.0",
+        "konva": "^10.6.0",
         "matrix-js-sdk": "^42.4.0",
         "ws": "^8.18.0"
       },
@@ -3940,6 +3941,38 @@
         "bignumber.js": "^9.0.0"
       }
     },
+    "node_modules/konva": {
+      "version": "10.6.0",
+      "resolved": "https://registry.npmjs.org/konva/-/konva-10.6.0.tgz",
+      "integrity": "sha512-6dyI91lZTx2mDu9DRo5zrT2Pg+Q3TFeyLfGQSAPJMhfvKjelbjPnTJimYQISam/EW+g0+pjLk/yztJKpSuxFPQ==",
+      "funding": [
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/lavrton"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/konva"
+        },
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/lavrton"
+        }
+      ],
+      "license": "MIT",
+      "peerDependencies": {
+        "canvas": "^3.0.0",
+        "skia-canvas": "^3.0.0"
+      },
+      "peerDependenciesMeta": {
+        "canvas": {
+          "optional": true
+        },
+        "skia-canvas": {
+          "optional": true
+        }
+      }
+    },
     "node_modules/lightningcss": {
       "version": "1.33.0",
       "resolved": "https://registry.npmjs.org/lightningcss/-/lightningcss-1.33.0.tgz",

package.json

@@ -17,6 +17,7 @@
     "@opentelemetry/sdk-logs": "^0.222.0",
     "@opentelemetry/sdk-node": "^0.222.0",
     "@opentelemetry/semantic-conventions": "^1.43.0",
+    "konva": "^10.6.0",
     "matrix-js-sdk": "^42.4.0",
     "ws": "^8.18.0"
   },

public/app.js

@@ -1,4 +1,5 @@
 import * as matrixcs from 'matrix-js-sdk';
+import Konva from 'konva';
 
 // ---------------------------------------------------------------------------
 // state
@@ -904,124 +905,132 @@ function closePopout(key) {
 }
 
 // ---------------------------------------------------------------------------
-// spatial canvas
+// spatial canvas (Konva) — the "me" node is draggable and stays visually
+// pinned to the stage center: every world-unit it moves, the layer is
+// re-centered by the same amount in the opposite direction, so the world
+// pans underneath it instead of the dot sliding away from center.
 // ---------------------------------------------------------------------------
-const ctx2d = els.spatial.getContext('2d');
-let dragging = null;
+const pad = 30;
+let sx = 1, sy = 1; // px per world unit, recomputed on resize
 
-// Camera is always centered on the listener (me), matching how the audio
-// panning already works (peer positions are relative to S.me). While
-// dragging, the camera freezes at its pre-drag position so the world
-// doesn't slide under the pointer as S.me moves.
-function spatialMetrics(camX = S.me.x, camY = S.me.y) {
-  const r = els.spatial.getBoundingClientRect();
-  const dpr = window.devicePixelRatio || 1;
-  if (els.spatial.width !== r.width * dpr) {
-    els.spatial.width = r.width * dpr;
-    els.spatial.height = r.height * dpr;
+const spatialStage = new Konva.Stage({ container: 'spatial', width: 1, height: 1 });
+const worldLayer = new Konva.Layer();
+spatialStage.add(worldLayer);
+
+// grid lives inside worldLayer so it pans with the camera along with everyone else
+const gridGroup = new Konva.Group({ listening: false });
+worldLayer.add(gridGroup);
+
+const rangeCircle = new Konva.Circle({
+  radius: 0, stroke: 'rgba(88,101,242,.35)', strokeWidth: 1, dash: [5, 5], listening: false,
+});
+worldLayer.add(rangeCircle);
+
+const avatarNodes = new Map(); // id -> {group, circle, initial, nameText}
+function avatarNode(id, isMe) {
+  let node = avatarNodes.get(id);
+  if (node) return node;
+  const r = isMe ? 16 : 13;
+  const group = new Konva.Group({ draggable: isMe });
+  const circle = new Konva.Circle({ radius: r, fill: '#5865f2' });
+  const initial = new Konva.Text({
+    text: '', fontStyle: 'bold', fontSize: Math.round(r), fill: '#fff',
+    width: r * 2, height: r * 2, offsetX: r, offsetY: r, align: 'center', verticalAlign: 'middle',
+  });
+  const nameText = new Konva.Text({
+    text: '', fontSize: 11, fill: isMe ? '#b5bac1' : '#80848e', y: r + 4, align: 'center',
+  });
+  group.add(circle, initial, nameText);
+  worldLayer.add(group);
+  if (isMe) {
+    group.dragBoundFunc(function (pos) {
+      // pos is in worldLayer's parent (stage) coordinates; clamp the
+      // resulting world position to the room bounds, not the screen.
+      const worldX = Math.min(ROOM_UNITS, Math.max(0, (pos.x - worldLayer.x()) / sx));
+      const worldY = Math.min(ROOM_UNITS, Math.max(0, (pos.y - worldLayer.y()) / sy));
+      return { x: worldLayer.x() + worldX * sx, y: worldLayer.y() + worldY * sy };
+    });
+    group.on('dragmove', () => {
+      S.me.x = (group.x() - worldLayer.x()) / sx;
+      S.me.y = (group.y() - worldLayer.y()) / sy;
+      recenterCamera();
+      sendMyPos();
+    });
+    group.on('dragstart', () => els.spatial.classList.add('dragging'));
+    group.on('dragend', () => {
+      els.spatial.classList.remove('dragging');
+      sendMyPos(true);
+    });
   }
-  ctx2d.setTransform(dpr, 0, 0, dpr, 0, 0);
-  const pad = 30;
-  const sx = (r.width - 2 * pad) / ROOM_UNITS;
-  const sy = (r.height - 2 * pad) / ROOM_UNITS;
-  const cx = r.width / 2, cy = r.height / 2;
-  return { w: r.width, h: r.height, pad, sx, sy,
-    toPx: (x, y) => [cx + (x - camX) * sx, cy + (y - camY) * sy],
-    fromPx: (px, py) => [camX + (px - cx) / sx, camY + (py - cy) / sy] };
+  node = { group, circle, initial, nameText };
+  avatarNodes.set(id, node);
+  return node;
+}
+
+function recenterCamera() {
+  const cx = spatialStage.width() / 2, cy = spatialStage.height() / 2;
+  worldLayer.position({ x: cx - S.me.x * sx, y: cy - S.me.y * sy });
+}
+
+function layoutSpatialGrid() {
+  gridGroup.destroyChildren();
+  for (let i = 0; i <= ROOM_UNITS; i += 10) {
+    gridGroup.add(new Konva.Line({
+      points: [i * sx, 0, i * sx, ROOM_UNITS * sy], stroke: 'rgba(255,255,255,.05)', strokeWidth: 1,
+    }));
+    gridGroup.add(new Konva.Line({
+      points: [0, i * sy, ROOM_UNITS * sx, i * sy], stroke: 'rgba(255,255,255,.05)', strokeWidth: 1,
+    }));
+  }
+}
+
+function resizeSpatial() {
+  const r = els.spatial.getBoundingClientRect();
+  spatialStage.width(r.width);
+  spatialStage.height(r.height);
+  sx = (r.width - 2 * pad) / ROOM_UNITS;
+  sy = (r.height - 2 * pad) / ROOM_UNITS;
+  layoutSpatialGrid();
+  recenterCamera();
+}
+new ResizeObserver(resizeSpatial).observe(els.spatial);
+resizeSpatial();
+
+function updateAvatarNode(node, id, name, muted, speaking, isMe, worldX, worldY) {
+  // Safe to set even for "me" while dragging: dragmove already derived
+  // S.me from the node's own current position, so this is a no-op then.
+  node.group.position({ x: worldX * sx, y: worldY * sy });
+  node.circle.stroke(muted ? '#f23f43' : speaking ? '#23a55a' : '');
+  node.circle.strokeWidth(3);
+  node.circle.fill(colorFor(id));
+  node.initial.text((name || '?').slice(0, 1).toUpperCase());
+  node.nameText.text(name + (isMe ? ' (you)' : ''));
+  node.nameText.offsetX(node.nameText.width() / 2);
 }
 
 function drawSpatial() {
-  const camX = dragging ? dragging.camX : S.me.x;
-  const camY = dragging ? dragging.camY : S.me.y;
-  const m = spatialMetrics(camX, camY);
-  ctx2d.clearRect(0, 0, m.w, m.h);
+  // range circle stays centered on "me", which the camera keeps at center
+  rangeCircle.position({ x: S.me.x * sx, y: S.me.y * sy });
+  rangeCircle.radius(Math.max(0, Math.min(56 * sx, spatialStage.height() / 2 - pad)));
 
-  // grid (world bounds only — clamped so it doesn't imply an infinite plane)
-  ctx2d.strokeStyle = 'rgba(255,255,255,.05)';
-  ctx2d.lineWidth = 1;
-  for (let i = 0; i <= ROOM_UNITS; i += 10) {
-    const [gx] = m.toPx(i, 0);
-    const [, gy] = m.toPx(0, i);
-    ctx2d.beginPath(); ctx2d.moveTo(gx, m.toPx(0, 0)[1]); ctx2d.lineTo(gx, m.toPx(0, ROOM_UNITS)[1]); ctx2d.stroke();
-    ctx2d.beginPath(); ctx2d.moveTo(m.toPx(0, 0)[0], gy); ctx2d.lineTo(m.toPx(ROOM_UNITS, 0)[0], gy); ctx2d.stroke();
+  const seen = new Set(['me']);
+  const meId = S.myId || 'me';
+  const meNode = avatarNode('me', true);
+  updateAvatarNode(meNode, meId, S.name || 'me', S.muted, els.meAvatar.classList.contains('speaking'), true, S.me.x, S.me.y);
+  meNode.group.moveToTop();
+
+  for (const peer of S.peers.values()) {
+    seen.add(peer.id);
+    const node = avatarNode(peer.id, false);
+    updateAvatarNode(node, peer.id, peer.name, peer.state.muted, peer.speaking, false, peer.x, peer.y);
+  }
+  for (const [id, node] of avatarNodes) {
+    if (!seen.has(id)) { node.group.destroy(); avatarNodes.delete(id); }
   }
 
-  // audible range around me (~gain 0.05 at 56 units with ref=4, rolloff=1.4)
-  const [mx, my] = m.toPx(S.me.x, S.me.y);
-  const rangeR = 56 * m.sx;
-  ctx2d.strokeStyle = 'rgba(88,101,242,.35)';
-  ctx2d.setLineDash([5, 5]);
-  ctx2d.beginPath(); ctx2d.arc(mx, my, Math.max(0, Math.min(rangeR, m.h / 2 - m.pad)), 0, Math.PI * 2); ctx2d.stroke();
-  ctx2d.setLineDash([]);
-
-  // peers
-  for (const peer of S.peers.values()) drawAvatarDot(peer.name, peer.id, peer.x, peer.y, peer.speaking, peer.state.muted, false, m);
-  // me (draggable listener, always drawn at the camera's center)
-  drawAvatarDot(S.name || 'me', S.myId || 'me', S.me.x, S.me.y, els.meAvatar.classList.contains('speaking'), S.muted, true, m);
-
-  if (!dragging) requestAnimationFrame(drawSpatial);
+  worldLayer.batchDraw();
 }
-requestAnimationFrame(drawSpatial);
-
-function drawAvatarDot(name, id, x, y, speaking, muted, isMe, m) {
-  const [px, py] = m.toPx(x, y);
-  const r = isMe ? 16 : 13;
-  ctx2d.beginPath();
-  ctx2d.arc(px, py, r, 0, Math.PI * 2);
-  ctx2d.fillStyle = colorFor(id);
-  ctx2d.fill();
-  if (speaking) {
-    ctx2d.strokeStyle = '#23a55a';
-    ctx2d.lineWidth = 3;
-    ctx2d.stroke();
-  }
-  if (muted) {
-    ctx2d.strokeStyle = '#f23f43';
-    ctx2d.lineWidth = 3;
-    ctx2d.beginPath();
-    ctx2d.moveTo(px - r, py + r);
-    ctx2d.lineTo(px + r, py - r);
-    ctx2d.stroke();
-  }
-  ctx2d.fillStyle = '#fff';
-  ctx2d.font = `bold ${Math.round(r)}px system-ui`;
-  ctx2d.textAlign = 'center';
-  ctx2d.textBaseline = 'middle';
-  ctx2d.fillText(name.slice(0, 1).toUpperCase(), px, py + 1);
-  ctx2d.font = '11px system-ui';
-  ctx2d.fillStyle = isMe ? '#b5bac1' : '#80848e';
-  ctx2d.fillText(name + (isMe ? ' (you)' : ''), px, py + r + 12);
-}
-
-els.spatial.addEventListener('pointerdown', (ev) => {
-  const m = spatialMetrics();
-  const rect = els.spatial.getBoundingClientRect();
-  const px = ev.clientX - rect.left, py = ev.clientY - rect.top;
-  const [ux, uy] = m.fromPx(px, py);
-  const d = Math.hypot(ux - S.me.x, uy - S.me.y);
-  if (d < 10) {
-    dragging = { dx: S.me.x - ux, dy: S.me.y - uy, camX: S.me.x, camY: S.me.y };
-    els.spatial.classList.add('dragging');
-    els.spatial.setPointerCapture(ev.pointerId);
-  }
-});
-els.spatial.addEventListener('pointermove', (ev) => {
-  if (!dragging) return;
-  const m = spatialMetrics(dragging.camX, dragging.camY);
-  const rect = els.spatial.getBoundingClientRect();
-  const [ux, uy] = m.fromPx(ev.clientX - rect.left, ev.clientY - rect.top);
-  S.me.x = Math.min(ROOM_UNITS, Math.max(0, ux + dragging.dx));
-  S.me.y = Math.min(ROOM_UNITS, Math.max(0, uy + dragging.dy));
-  drawSpatial();
-  sendMyPos();
-});
-els.spatial.addEventListener('pointerup', () => {
-  if (!dragging) return;
-  dragging = null;
-  els.spatial.classList.remove('dragging');
-  sendMyPos(true);
-  requestAnimationFrame(drawSpatial); // rAF loop stopped while dragging
-});
+requestAnimationFrame(function loop() { drawSpatial(); requestAnimationFrame(loop); });
 
 // ---------------------------------------------------------------------------
 // Matrix text chat (experimental) — native OAuth2/OIDC auth against the

public/index.html

@@ -72,7 +72,7 @@
         <span>SPATIAL AUDIO</span>
         <span class="hint">drag <b>your</b> avatar &mdash; others hear you shift; you hear them relative to position</span>
       </div>
-      <canvas id="spatial"></canvas>
+      <div id="spatial"></div>
     </div>
   </main>
 </div>