From 79c0557d1f3187f4e8a1113888aa0e678684dd20 Mon Sep 17 00:00:00 2001
From: 4gray <fourgray@proton.me>
Date: Sun, 23 Aug 2026 09:05:08 +0200
Subject: [PATCH] fix(playback): detect CP1251 subtitles by letter bytes, not
 whole-file ratio
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Codex review on #1471: short-dialogue CP1251 SRT files are dominated by
ASCII cue counters and timing lines, so a whole-file high-byte ratio stayed
under the threshold and the Cyrillic text decoded as Windows-1252 mojibake.
The discriminator now compares high bytes against ASCII letters only —
timing scaffolding is identical in every candidate encoding and no longer
dilutes the signal.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .../external-subtitle-cues.util.spec.ts       | 19 +++++++++++++++++++
 .../external-subtitle-cues.util.ts            | 13 ++++++++++++-
 2 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.spec.ts b/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.spec.ts
index abc45ce55..dc0477b91 100644
--- a/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.spec.ts
+++ b/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.spec.ts
@@ -40,6 +40,25 @@ describe('external-subtitle-cues.util', () => {
             );
         });
 
+        it('detects CP1251 in a short-dialogue SRT where ASCII timing bytes dominate', () => {
+            // "Да" / "Нет" in CP1251 among full SRT timing scaffolding: only
+            // 5 of ~70 bytes are high, but ALL letter bytes are — the
+            // discriminator must ignore the timing lines.
+            const ascii = (text: string) =>
+                Array.from(text).map((c) => c.charCodeAt(0));
+            const srt = [
+                ...ascii('1\n00:00:01,000 --> 00:00:02,000\n'),
+                0xc4, 0xe0, // Да
+                ...ascii('\n\n2\n00:00:03,000 --> 00:00:04,000\n'),
+                0xcd, 0xe5, 0xf2, // Нет
+                ...ascii('\n'),
+            ];
+            expect(decodeExternalSubtitleBytes(toBuffer(srt))).toContain('Да');
+            expect(decodeExternalSubtitleBytes(toBuffer(srt))).toContain(
+                'Нет'
+            );
+        });
+
         it('decodes mostly-ASCII Windows-1252 bytes (sparse accents)', () => {
             // "resume: cafe" with two accented letters among ASCII.
             const cp1252 = [
diff --git a/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.ts b/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.ts
index 3837b2019..1e822056c 100644
--- a/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.ts
+++ b/libs/ui/playback/src/lib/web-video-support/external-subtitle-cues.util.ts
@@ -64,14 +64,25 @@ export function decodeExternalSubtitleBytes(buffer: ArrayBuffer): string {
         // Not valid UTF-8: a legacy single-byte encoding.
     }
 
+    // Compare only letter-carrying bytes: cue counters and timing lines are
+    // ASCII digits/punctuation and identical in every candidate encoding, so
+    // ratios over the whole file would drown short dialogue ("Да"/"Нет") in
+    // timing bytes. In a non-Latin-script file virtually every letter is a
+    // high byte; Latin text with accents stays dominated by ASCII letters.
     let highBytes = 0;
+    let asciiLetters = 0;
     for (const byte of bytes) {
         if (byte >= 0x80) {
             highBytes += 1;
+        } else if (
+            (byte >= 0x41 && byte <= 0x5a) ||
+            (byte >= 0x61 && byte <= 0x7a)
+        ) {
+            asciiLetters += 1;
         }
     }
     const encoding =
-        highBytes / Math.max(1, bytes.length) > 0.15
+        highBytes / Math.max(1, highBytes + asciiLetters) > 0.3
             ? 'windows-1251'
             : 'windows-1252';
     try {