From 0ac862c681fb43a992ec641c945bfa3cca77bc13 Mon Sep 17 00:00:00 2001
From: Worph <worph@users.noreply.github.com>
Date: Mon, 14 Sep 2026 22:29:50 +0200
Subject: [PATCH] improve search and fix languages

---
 src/plugin.ts         | 23 +++++++++++++++++++++--
 test/subtitle.test.ts | 11 +++++++++--
 2 files changed, 30 insertions(+), 4 deletions(-)

diff --git a/src/plugin.ts b/src/plugin.ts
index d0193d5..137303c 100644
--- a/src/plugin.ts
+++ b/src/plugin.ts
@@ -242,9 +242,22 @@ export async function midhashOverWebDAV(client: WebDAVClient, filePath: string,
 // Record keys
 // ----------------------------------------------------------------------------
 
+/**
+ * Keys on the video record for one linked subtitle (METADATA_KEYS.md §8, §9).
+ *
+ * The language lands in `subtitleLanguages/<lang3>` **and** in the
+ * `languages/<lang3>` union — §9 rule #7: "a union field is written, never
+ * computed". Nothing reconciles them afterwards, and `languages` is the only
+ * field a query filters on, so a subtitle language recorded in the split alone
+ * is invisible to every language filter on every peer. This mirrors
+ * subtitle-extractor's `videoSubtitleKeys`, which has always written both.
+ */
 export function videoSubtitleKeys(subCid: string, lang3: string): Record<string, string> {
     const keys: Record<string, string> = { [`subtitles/${lang3}/${subCid}`]: 'true' };
-    if (lang3 !== 'und') keys[`subtitleLanguages/${lang3}`] = 'true';
+    if (lang3 !== 'und') {
+        keys[`subtitleLanguages/${lang3}`] = 'true';
+        keys[`languages/${lang3}`] = 'true';
+    }
     return keys;
 }
 
@@ -299,7 +312,13 @@ async function writeVideoLinks(
         deletes.push('subtitleLanguages');
         for (const code of legacyLangs) {
             const l = languageToken(code);
-            if (l && l !== 'und') sets[`subtitleLanguages/${l}`] = 'true';
+            if (l && l !== 'und') {
+                sets[`subtitleLanguages/${l}`] = 'true';
+                // The legacy csv never had a union member to migrate, so the
+                // migration is also the only chance to satisfy §9 rule #7 for
+                // these records.
+                sets[`languages/${l}`] = 'true';
+            }
         }
     }
     for (const { subCid, lang3 } of links) deletes.push(...staleLanguageLeaves(flat, 'subtitles', subCid, lang3));
diff --git a/test/subtitle.test.ts b/test/subtitle.test.ts
index 41d7d7e..6aca9bd 100644
--- a/test/subtitle.test.ts
+++ b/test/subtitle.test.ts
@@ -220,8 +220,15 @@ describe('midhash256', () => {
 });
 
 describe('record keys (METADATA_KEYS.md §8)', () => {
-    it('video: language-nested leaf, plus the facet for a known language', () => {
-        expect(videoSubtitleKeys('bagsub', 'fre')).toEqual({ 'subtitles/fre/bagsub': 'true', 'subtitleLanguages/fre': 'true' });
+    it('video: language-nested leaf, plus the facet AND the union for a known language', () => {
+        // §9 rule #7 — the split never travels without the union, because
+        // `languages` is the only field a query filters on.
+        expect(videoSubtitleKeys('bagsub', 'fre')).toEqual({
+            'subtitles/fre/bagsub': 'true',
+            'subtitleLanguages/fre': 'true',
+            'languages/fre': 'true',
+        });
+        // `und` is never a member of either set (§9).
         expect(videoSubtitleKeys('bagsub', 'und')).toEqual({ 'subtitles/und/bagsub': 'true' });
     });