Time-skip: a big file that did not change is not listed as changed

apps/host/server/timeskip.mjs

@@ -118,6 +118,16 @@ function readText(path) {
   } catch { return { missing: true }; }
 }
 
+/** Whether two files hold the same bytes (a big file that did not change must not be listed as changed). Files over 64 MB are treated as changed. */
+function sameBytes(a, b) {
+  try {
+    const sa = statSync(a), sb = statSync(b);
+    if (sa.size !== sb.size) return false;
+    if (sa.size > 64 * 1024 * 1024) return false;
+    return readFileSync(a).equals(readFileSync(b));
+  } catch { return false; }
+}
+
 /** What the assistant is told for one revision attempt. It works on a copy, so it is told to stay in it and to say what it did. */
 export function timeskipRevisionMessage(request) {
   return [
@@ -235,6 +245,8 @@ export function diffRun(record) {
     const status = a.missing ? 'added' : b.missing ? 'removed' : null;
     const same = !status && a.text !== undefined && a.text === b.text;
     if (same) continue;
+    // A file too big to read as text is the same file when its bytes are the same: it must not show up as a change.
+    if (!status && (a.big || b.big) && sameBytes(join(baseRoot, path), join(updatedRoot, path))) continue;
     if (imageTypeOf(path)) {
       // A picture is compared as bytes and shown from the file itself, never read as text.
       let differs = true;

apps/host/tests/timeskip-compare-server.test.mjs

@@ -65,3 +65,17 @@ test('the page that was made comes first; a browser profile and other housekeepi
     assert.deepEqual(onlyNoise.files, [], 'when only housekeeping changed nothing is shown as changed'); assert.equal(onlyNoise.housekeeping, 12);
   } finally { rmSync(dir, { recursive: true, force: true }); }
 });
+
+test('a big file that did not change is not listed as changed; one that did is', () => {
+  const dir = mkdtempSync(join(tmpdir(), 'strata-ts-big-'));
+  try {
+    const base = join(dir, 'base'), cand = join(dir, 'cand'); mkdirSync(base); mkdirSync(cand);
+    const big = Buffer.alloc(400 * 1024, 7);
+    writeFileSync(join(base, 'archive.bin'), big); writeFileSync(join(cand, 'archive.bin'), big);
+    writeFileSync(join(base, 'data.json'), Buffer.alloc(300 * 1024, 65)); writeFileSync(join(cand, 'data.json'), Buffer.alloc(300 * 1024, 66));
+    writeFileSync(join(cand, 'page.html'), '<p>x</p>');
+    const found = diffRun({ checkpoints: [{ id: 'cp1-baseline', label: 'baseline', path: base }], selected_checkpoint: 'cp1-baseline', attempts: [{ index: 1, path: cand }] });
+    assert.deepEqual(found.files.map(f => f.path), ['page.html', 'data.json'], 'the untouched big file is gone from the list; the changed one is last (too big to show)');
+    assert.equal(found.files[1].unreadable, true);
+  } finally { rmSync(dir, { recursive: true, force: true }); }
+});