@quolu/lattice 0.12.32 → 0.12.34

package/README.md

@@ -26,11 +26,17 @@ CLIの全体像は`lattice --help`、各公開namespaceの正規構文は
 npm test
 npm run check
 npm run ci
+node scripts/reap-orphan-test-daemons.mjs
 lattice sensor sync . --json
 spotter doctor
 codex-sidecar diagnostics --project . --preset auditor --json
 ```
 
+`reap-orphan-test-daemons.mjs`は、実daemonを起動するtestが取り残したprocessを一覧します。既定は表示
+だけで何も停めません。停めるのは`--reap`を付けた時に限り、対象はargvが指すfixtureのdirectoryが既に
+存在しないものだけです。実行中のtestを巻き込まないための条件なので、fixtureが残ったまま死んだ実行を
+含めたい場合だけ`--older-than-hours=<n>`で起動時刻による許可を明示します。
+
 未初期化projectで`sensor sync`した場合は`LATTICE_SENSOR_NOT_INITIALIZED`と正規`next_action`を返します。
 その他のsensor失敗もexit code、signal、bounded stderrをtyped detailへ残し、原因を隠しません。
 
@@ -137,6 +143,10 @@ LANや外部reverse proxyから閲覧するoptional bridgeは既定で無効で
 digest付きsidecarの欠落・改ざんはtyped failureになります。
 dashboard daemonは起動時に読み込んだ版数をhealthで名乗り、installされた版と食い違えば`lattice status`の
 たびに新版daemonへ置き換わります。publishしただけで配信面が古いまま残ることはありません。
+入れ替えは新daemonが登録済み全projectのstoreを読み終えるまで待つため、`lattice status`の応答が
+その間伸びます（実測: 8 project登録で約50秒台）。待ち時間は固定秒数ではなく、spawnした子が生きている
+間だけ待ち、子が死ねば即座に`DASHBOARD_DAEMON_UNAVAILABLE`を返します。既定120秒の上限は、応答を
+返さない子に対するbackstopであって正常な起動時間の見積りではありません。
 状態を書き込む`start / block / unblock / done / evidence promote / reopen / revise / revise-phase / revise-set`
 では、監査actorとして次の3環境変数をすべて設定してください。
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@quolu/lattice",
-  "version": "0.12.32",
+  "version": "0.12.34",
   "description": "Lattice — phase-aware TODO graph compiler and conflict-aware orchestration runtime",
   "type": "module",
   "license": "MIT",

package/src/todo-gantt-layout.mjs

@@ -416,9 +416,15 @@ export function layoutTodoGantt(readModel, chainProjection, options = {}) {
   const layers = orderLayers(nodes, wave, incoming, outgoing);
   // A ToDo with no registered dependency has nothing to line up with. Laying
   // every one of them out in a single row makes the canvas as wide as the plan
-  // is old — real stores have hundreds — so they wrap into a block underneath
-  // the wired ToDos of their stage. Nodes that carry edges keep the top row:
-  // edge routing assumes both endpoints of a stage share one baseline.
+  // is old — real stores have hundreds — so they wrap into a block of rows.
+  //
+  // That block goes ABOVE the wired ToDos of its stage. Below them it used to be
+  // pierced: a wired card's port descends from its own bottom edge to the routing
+  // band under the whole stage, straight through whatever independent card had
+  // wrapped into the same column, and the hidden half of that line read as a
+  // dependency the plan does not have. Independent ToDos carry no edge, so they
+  // have no incoming one either and always land in stage 0, which nothing ever
+  // arrives into — putting them on top leaves no line to cross them.
   const wired = new Set();
   for (const edge of edges) { wired.add(edge.from); wired.add(edge.to); }
   const transversePosition = new Map();
@@ -427,12 +433,15 @@ export function layoutTodoGantt(readModel, chainProjection, options = {}) {
   for (const layer of layers) {
     const connected = layer.filter((key) => wired.has(key));
     const loose = layer.filter((key) => !wired.has(key));
-    connected.forEach((key, index) => { transversePosition.set(key, index); stageRow.set(key, 0); });
     const columns = Math.max(LOOSE_COLUMNS_MINIMUM, connected.length);
-    const firstLooseRow = connected.length === 0 ? 0 : 1;
+    const looseRows = loose.length === 0 ? 0 : Math.ceil(loose.length / columns);
     loose.forEach((key, index) => {
       transversePosition.set(key, index % columns);
-      stageRow.set(key, firstLooseRow + Math.floor(index / columns));
+      stageRow.set(key, Math.floor(index / columns));
+    });
+    connected.forEach((key, index) => {
+      transversePosition.set(key, index);
+      stageRow.set(key, looseRows);
     });
     stageRowCount.push(Math.max(0, ...layer.map((key) => stageRow.get(key))) + 1);
   }