archilyzer

commit 3ceadf7606a69ca9b95e9a7a00622c2344e3bce9
parent 02318dfc0e426f2bc020b0c5ffb4ec5fed9e92f6
Author: I Mean I'm Just Saying <imeanimjustsaying@kiwifarms.st>
Date:   Fri, 25 Sep 2026 18:47:40 -0400

common: a new video dir rebases the channel's title memo instead of dropping it

The source-3 title memo was keyed by data/'s mtime and dropped whole when
it changed, so every new dir a download lane added re-read every
metadata.info.json on the channel (8,049 on the-quartering-rumble). It is
now rebased: one readdir keeps the titles of the dirs still present and
only new dirs are read; a removed dir's title goes with it (release 8
review, V). videoTitles.test.ts: the memo case asserts 1 read for a new
dir and a fall-through for a removed one.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Diffstat:
M
common/controller/videoTitles.test.ts
 | 
28
++++++++++++++++++++++------
M
common/controller/videoTitles.ts
 | 
25
++++++++++++++++++++++---

2 files changed, 44 insertions(+), 9 deletions(-)
diff --git a/common/controller/videoTitles.test.ts b/common/controller/videoTitles.test.ts
@@ -289,22 +289,38 @@ test("metadata titles are memoized per channel until data/ changes", async () =>
     assert.equal(second.get("b")?.title, "Bravo");
     assert.equal(videoTitleMetadataReadCount() - r1, 0);
 
-    // A new video dir changes data/'s mtime: the channel's memo is dropped and
-    // every id is read again ("a" now has no file, so no title).
+    // A new video dir changes data/'s mtime: the memo is REBASED — the dirs
+    // still there keep their titles and only the new one is read (release 8
+    // review, V). "a"'s dir is still there, so its memoized title stands.
     await writeInfo(paths, "c", { id: "c", title: "Charlie" });
     const r2 = videoTitleMetadataReadCount();
     const third = await readChannelVideoTitles(paths, SLUG, ["a", "b", "c"]);
-    assert.equal(third.has("a"), false);
+    assert.equal(third.get("a")?.title, "Alpha");
+    assert.equal(third.get("b")?.title, "Bravo");
     assert.equal(third.get("c")?.title, "Charlie");
-    assert.equal(videoTitleMetadataReadCount() - r2, 3);
+    assert.equal(videoTitleMetadataReadCount() - r2, 1);
+
+    // A REMOVED dir takes its title with it: the rebase keeps only the dirs
+    // still present, so "a" falls through, is read, and has no title.
+    await rm(path.join(dataDir, "a"), { recursive: true, force: true });
+    // Two changes inside one filesystem timestamp tick would share an mtime;
+    // pin a distinct one so the rebase certainly fires.
+    const later = new Date("2026-01-02T00:00:00Z");
+    await utimes(dataDir, later, later);
+    const r3 = videoTitleMetadataReadCount();
+    const fourth = await readChannelVideoTitles(paths, SLUG, ["a", "b", "c"]);
+    assert.equal(fourth.has("a"), false);
+    assert.equal(fourth.get("c")?.title, "Charlie");
+    assert.equal(videoTitleMetadataReadCount() - r3, 1);
 
     // A miss is not memoized: a title that lands later is found.
+    await mkdir(path.join(dataDir, "a"), { recursive: true });
     await writeFile(
       path.join(dataDir, "a", "metadata.info.json"),
       JSON.stringify({ id: "a", title: "Alpha again" }),
     );
-    const fourth = await readChannelVideoTitles(paths, SLUG, ["a"]);
-    assert.equal(fourth.get("a")?.title, "Alpha again");
+    const fifth = await readChannelVideoTitles(paths, SLUG, ["a"]);
+    assert.equal(fifth.get("a")?.title, "Alpha again");
   } finally {
     resetVideoTitleMemo();
     await rm(dir, { recursive: true, force: true });
diff --git a/common/controller/videoTitles.ts b/common/controller/videoTitles.ts
@@ -29,7 +29,7 @@
 // invariant).
 
 import { existsSync } from "node:fs";
-import { open as openFile, stat } from "node:fs/promises";
+import { open as openFile, readdir, stat } from "node:fs/promises";
 import path from "node:path";
 import { open } from "lmdb";
 import type { Paths } from "../lib/paths";
@@ -156,7 +156,10 @@ async function readMetadataTitle(videoDir: string): Promise<string | null> {
 // 8,049 such dirs: 5,024 ms cold / 224 ms warm per render, measured 2026-09-25.
 //
 // Per channel, keyed by the `data/` directory's mtime: adding or removing a
-// video dir changes it and drops the channel's entry. Rewriting a
+// video dir changes it, and the entry is then REBASED rather than dropped — the
+// titles of the dirs still there carry over (one readdir, no file reads), and
+// only the new dirs are read (release 8 review, V: a busy download lane adds a
+// dir every few minutes, and each one used to re-read all 8,049). Rewriting a
 // metadata.info.json INSIDE an existing dir does not — accepted, because a
 // video's title does not change after download. Misses (no file yet, e.g. a dir
 // mid-download) are NOT memoized, so a title that lands later is picked up.
@@ -196,8 +199,24 @@ async function memoForChannel(
   }
   const memo = titleMemo();
   let entry = memo.get(key);
-  if (!entry || entry.dataDirMtimeMs !== mtimeMs) {
+  if (!entry) {
     entry = { dataDirMtimeMs: mtimeMs, titles: new Map() };
+  } else if (entry.dataDirMtimeMs !== mtimeMs) {
+    // Keep the titles of the dirs that are still there; a removed dir's title
+    // goes with it, so a re-listed id falls through to the other sources.
+    let present: Set<string> | null = null;
+    try {
+      present = new Set(await readdir(dataDir));
+    } catch {
+      present = null;
+    }
+    const kept = new Map<string, string>();
+    if (present) {
+      for (const [id, title] of entry.titles) {
+        if (present.has(id)) kept.set(id, title);
+      }
+    }
+    entry = { dataDirMtimeMs: mtimeMs, titles: kept };
   }
   // Re-insert so iteration order is least-recently-used first.
   memo.delete(key);