fix: keep the word boundary for space-separated previews

The kept-ratio fallback was added for CJK summaries whose only Latin space
sits near the start, but it applied to every script. A space-separated
summary ending in a long token — a URL, a compound word — lost its word
boundary and was cut mid-token instead.

Gate the ratio on the discarded tail actually being non-spacing script,
reusing the character class the catalog search tokenizer already relies on
in convex/lib/searchText.ts.

Author: Yigtwxx

src/lib/truncateText.test.ts

@@ -31,6 +31,20 @@ describe("truncateText", () => {
     );
   });
 
+  it("keeps the word boundary when a long Latin word ends the slice", () => {
+    // The only space sits before 60% of the slice, but the discarded tail is space-separated
+    // text, so the preview must still end on a word rather than mid-word.
+    expect(truncateText("Read this hyperextendedword", 20)).toBe("Read this…");
+    expect(truncateText("Ship internationalization", 22)).toBe("Ship…");
+  });
+
+  it("keeps the Latin word boundary when only a short CJK tail is discarded", () => {
+    // Here the boundary already keeps most of the slice, so backtracking loses nothing but a
+    // stray CJK character that reads as noise on its own.
+    expect(truncateText("Deploy helper for 中文文档管理", 20)).toBe("Deploy helper for…");
+    expect(truncateText("Release notes generator 日本語対応", 28)).toBe("Release notes generator…");
+  });
+
   it("keeps the preview budget for Chinese summaries carrying an early Latin space", () => {
     // The single space after "5GC" is the only space in 104 characters, so backtracking
     // to it would leave a three-character preview.

src/lib/truncateText.ts

@@ -4,13 +4,19 @@ export const PUBLIC_CATALOG_NAME_PREVIEW_LENGTH = 70;
 // single Latin space near the start of a summary. Backtracking to that space would drop
 // nearly the whole preview, so only honour a word boundary that keeps most of the slice.
 const WORD_BOUNDARY_MIN_KEPT_RATIO = 0.6;
+// Same character class as the catalog search tokenizer in convex/lib/searchText.ts.
+const CJK_RE = /[\u4e00-\u9fff\u3400-\u4dbf\u3041-\u3096\u30a1-\u30fa\uac00-\ud7af]/;
 
 export function truncateText(value: string, maxLength: number) {
   const normalized = value.trim().replace(/\s+/g, " ");
   if (normalized.length <= maxLength) return normalized;
   const truncated = normalized.slice(0, Math.max(0, maxLength - 1)).trimEnd();
   const wordBoundary = truncated.lastIndexOf(" ");
+  // The ratio only guards non-spacing scripts. Space-separated text keeps its word boundary
+  // however long the trailing word is.
+  const discardsCJK = CJK_RE.test(truncated.slice(wordBoundary + 1));
   const keepsMostOfSlice = wordBoundary >= truncated.length * WORD_BOUNDARY_MIN_KEPT_RATIO;
-  const text = wordBoundary > 0 && keepsMostOfSlice ? truncated.slice(0, wordBoundary) : truncated;
+  const honoursBoundary = wordBoundary > 0 && (keepsMostOfSlice || !discardsCJK);
+  const text = honoursBoundary ? truncated.slice(0, wordBoundary) : truncated;
   return `${text}…`;
 }