@cueloop/schema 0.1.0-alpha.65 → 0.1.0-alpha.67

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cueloop/schema",
-  "version": "0.1.0-alpha.65",
+  "version": "0.1.0-alpha.67",
   "description": "The ReviewSession primitive: types, markdown block model, quote-primary anchors, feedback serialization",
   "homepage": "https://github.com/mmurakaru/cueloop#readme",
   "bugs": {

package/src/fuzzy.test.ts

@@ -89,4 +89,22 @@ describe("fuzzyFindBestMatch", () => {
     // Assert
     expect(match.start).toBe(0);
   });
+
+  test("a long quote fuzzed over a long document bails within its work budget", () => {
+    // Arrange - a stale multi-line anchor whose text is gone, over a large document; the naive
+    // scan is O(haystack * needle^3) and would pin the CPU for a minute
+    const needle = Array.from({ length: 200 }, (_unused, index) => `token-${index}`).join(" ");
+    const haystack = Array.from(
+      { length: 400 },
+      (_unused, index) => `line ${index} of the diff`,
+    ).join("\n");
+
+    // Act
+    const started = Date.now();
+    const match = fuzzyFindBestMatch(needle, haystack, 0.75);
+
+    // Assert - it returns (here: no match clears the bar) fast, never spinning
+    expect(match).toBeNull();
+    expect(Date.now() - started).toBeLessThan(2000);
+  });
 });

package/src/fuzzy.ts

@@ -62,11 +62,21 @@ export function similarityRatio(left: string, right: string): number {
   return 1 - levenshteinDistance(left, right) / longestLength;
 }
 
+/**
+ * Character comparisons this scan may spend before it gives up and returns the
+ * best window found so far (or null). The scan is O(haystack * needle^3) - a
+ * long quote fuzzed over a long document (a stale multi-block anchor whose text
+ * is gone) would otherwise pin the CPU for a minute; a passage that is truly gone
+ * is better left orphaned than fuzzed. Small everyday quotes never approach it.
+ */
+const FUZZY_WORK_BUDGET = 150_000_000;
+
 /**
  * Find the haystack window most similar to the needle, or null when nothing
  * clears `minimumSimilarity`. Windows range around the needle length so the
  * match tolerates a few inserted or deleted characters, not only substitutions.
- * Ties keep the earliest window. Offsets index the haystack directly.
+ * Ties keep the earliest window. Offsets index the haystack directly. The scan
+ * stops once it has spent `FUZZY_WORK_BUDGET`, returning the best window so far.
  */
 export function fuzzyFindBestMatch(
   needle: string,
@@ -80,6 +90,7 @@ export function fuzzyFindBestMatch(
   const maximumWindowLength = needle.length + lengthTolerance;
 
   let bestMatch: FuzzyMatch | null = null;
+  let remainingWork = FUZZY_WORK_BUDGET;
 
   for (let start = 0; start < haystack.length; start++) {
     for (
@@ -87,6 +98,9 @@ export function fuzzyFindBestMatch(
       windowLength <= maximumWindowLength && start + windowLength <= haystack.length;
       windowLength++
     ) {
+      // each window costs a needle x window edit-distance pass; spend that from the budget
+      remainingWork -= needle.length * windowLength;
+      if (remainingWork <= 0) return bestMatch;
       const window = haystack.slice(start, start + windowLength);
       const similarity = similarityRatio(needle, window);
 

package/src/types.ts

@@ -12,6 +12,10 @@ export const SCHEMA_VERSION = "1";
 export interface WorkspaceKey {
   repoRoot: string;
   branch: string;
+  /** Earliest root commit SHA; the project key that survives moving or re-cloning the repo. Absent for a standalone thread. */
+  rootCommit?: string;
+  /** Origin remote URL when present; for display and repair only, never the project key. */
+  remote?: string;
 }
 
 /**