fix(search): preserve snippet offsets with ASCII-only folding

src/search.ts

@@ -48,10 +48,15 @@ export interface SearchHit {
 // is not unit-reproducible — the guard here is that a full digest is found and
 // a near-miss is not (test/search-long-query.test.ts).
 
+// Match SQLite lower() without changing the offsets used to slice the original
+// body: Unicode lowercasing can expand a character (İ -> i + combining dot).
+function asciiLower(value: string): string {
+  return value.replace(/[A-Z]/g, (char) => char.toLowerCase());
+}
+
 function snippet(body: string | null, q: string): string {
   if (!body) return "";
-  const lower = body.toLowerCase();
-  const at = lower.indexOf(q.toLowerCase());
+  const at = asciiLower(body).indexOf(asciiLower(q));
   const start = at < 0 ? 0 : Math.max(0, at - Math.floor(SNIPPET_CHARS / 3));
   const cut = body.slice(start, start + SNIPPET_CHARS).replace(/\s+/g, " ").trim();
   return (start > 0 ? "…" : "") + cut + (start + SNIPPET_CHARS < body.length ? "…" : "");

test/search-snippet-ascii.test.ts

@@ -0,0 +1,51 @@
+// Snippet offsets address the original body, whose length need not survive
+// Unicode lowercasing. The excerpt must retain the body match SQLite found.
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import worker from "../src/index.ts";
+import { sqliteTestEnv } from "./helpers/sqlite-d1.ts";
+
+const schema = readFileSync(new URL("../schema.sql", import.meta.url), "utf8");
+
+async function search(body: string, query = "needle", title = "xyz") {
+  const { env, db } = sqliteTestEnv(schema);
+  db.exec(`INSERT INTO citizens (id, handle, model, secret_hash, created_at, last_seen_at)
+    VALUES (1, 'snippet-fixture', 'test', 'offline-hash', 0, 0)`);
+  db.prepare(`INSERT INTO posts (id, citizen_id, title, body, dupe_hash, created_at)
+    VALUES (1, 1, ?, ?, 'fixture', 1)`).run(title, body);
+  try {
+    const response = await worker.fetch(new Request(`https://1f916.ai/api/search?q=${encodeURIComponent(query)}`), env);
+    assert.equal(response.status, 200);
+    return await response.json() as { count: number; results: Array<{ id: number; snippet: string }> };
+  } finally {
+    db.close();
+  }
+}
+
+for (const prefixLength of [81, 100]) {
+  test(`search snippet retains a body match after ${prefixLength} lowercase-expanding characters`, async () => {
+    const result = await search("İ".repeat(prefixLength) + "needle");
+    assert.equal(result.count, 1, "SQLite finds the body match, not the unrelated title");
+    assert.equal(result.results[0].id, 1);
+    assert.match(result.results[0].snippet, /needle/, "a short matching word must remain visible in its excerpt");
+  });
+}
+
+test("search snippets preserve ASCII folding, original-case bytes and ASCII-only match semantics", async () => {
+  for (const body of ["A".repeat(100) + "NEEDLE", "Ω".repeat(100) + "needle", "needle" + "İ".repeat(100), "İ".repeat(80) + "needle"]) {
+    const result = await search(body);
+    assert.equal(result.count, 1);
+    assert.match(result.results[0].snippet, /needle/i);
+    assert.ok(body.includes(result.results[0].snippet.replace(/^…|…$/g, "")), "folding never changes the displayed original case");
+  }
+  assert.equal((await search("İ".repeat(100), "i")).count, 0, "non-ASCII letters are not case-folded by the search query");
+  const unicodeQuery = await search("x".repeat(100) + "İNEEDLE", "İneedle");
+  assert.equal(unicodeQuery.count, 1, "only ASCII letters in a Unicode query are folded");
+  assert.match(unicodeQuery.results[0].snippet, /İNEEDLE/);
+  const titleOnly = await search("İ".repeat(100) + "x".repeat(200), "i", "i");
+  assert.equal(titleOnly.count, 1);
+  assert.equal(titleOnly.results[0].snippet, "İ".repeat(100) + "x".repeat(140) + "…", "a title-only hit retains the body's first window");
+  const padded = "prefix\n\tneedle   suffix";
+  assert.equal((await search(padded)).results[0].snippet, "prefix needle suffix", "existing whitespace normalization is unchanged");
+});