From fc8efc62e03ca3cb3caebb27fd65a983e8bf9a5a Mon Sep 17 00:00:00 2001
From: Brummel <mould@mycelium.de>
Date: Tue, 14 Apr 2026 11:02:37 +0200
Subject: [PATCH] Add user slug to transcription jobs

The `TranscribeJob` struct now includes a `user_slug` field. This is
necessary for the transcription worker to look up per-user settings such
as hotwords and initial prompts. The `collect_pending` function has been
updated to return user slugs along with audio paths, and the recovery
scan now enqueues jobs with the correct user slug.
---
 server/src/routes/upload.rs       |  5 ++++-
 server/src/transcribe/mod.rs      |  3 +++
 server/src/transcribe/recovery.rs | 22 ++++++++++++++++------
 3 files changed, 23 insertions(+), 7 deletions(-)

diff --git a/server/src/routes/upload.rs b/server/src/routes/upload.rs
index 760540a..389b457 100644
--- a/server/src/routes/upload.rs
+++ b/server/src/routes/upload.rs
@@ -86,7 +86,10 @@ pub async fn handle_upload(
 
     // Enqueue transcription. Failure here is non-fatal: on restart the recovery
     // scan will re-enqueue any .m4a without a .transcript.txt sibling.
-    match transcribe_tx.try_send(TranscribeJob { audio_path: file_path.clone() }) {
+    match transcribe_tx.try_send(TranscribeJob {
+        audio_path: file_path.clone(),
+        user_slug: user.slug.clone(),
+    }) {
         Ok(()) => {}
         Err(TrySendError::Full(_)) => {
             warn!(file = %file_path.display(), "Transcription queue full; will be picked up on restart");
diff --git a/server/src/transcribe/mod.rs b/server/src/transcribe/mod.rs
index cdd2bc4..96b25f2 100644
--- a/server/src/transcribe/mod.rs
+++ b/server/src/transcribe/mod.rs
@@ -13,6 +13,9 @@ pub const QUEUE_DEPTH: usize = 64;
 #[derive(Debug, Clone)]
 pub struct TranscribeJob {
     pub audio_path: PathBuf,
+    /// Slug of the user who owns this recording. Used by the worker to look up
+    /// per-user transcription settings (hotwords, language, initial_prompt).
+    pub user_slug: String,
 }
 
 pub type TranscribeSender = mpsc::Sender<TranscribeJob>;
diff --git a/server/src/transcribe/recovery.rs b/server/src/transcribe/recovery.rs
index c9a61e0..6d4b819 100644
--- a/server/src/transcribe/recovery.rs
+++ b/server/src/transcribe/recovery.rs
@@ -14,10 +14,11 @@ pub async fn scan_and_enqueue(data_path: &Path, tx: &TranscribeSender) {
     let total = pending.len();
     info!(count = total, "Recovery scan found pending recordings");
 
-    for audio_path in pending {
+    for (user_slug, audio_path) in pending {
         if tx
             .send(TranscribeJob {
                 audio_path: audio_path.clone(),
+                user_slug: user_slug.clone(),
             })
             .await
             .is_err()
@@ -28,8 +29,12 @@ pub async fn scan_and_enqueue(data_path: &Path, tx: &TranscribeSender) {
     }
 }
 
-async fn collect_pending(data_path: &Path) -> Vec<PathBuf> {
-    let mut out = Vec::new();
+/// Walks `$data_path/<slug>/open/<case>/*.m4a` and returns `(slug, audio_path)`
+/// tuples for every recording without a `.transcript.txt` sibling. The slug
+/// comes from the top-level user directory name — that's the only ownership
+/// signal after a restart (auth context is gone).
+async fn collect_pending(data_path: &Path) -> Vec<(String, PathBuf)> {
+    let mut out: Vec<(String, PathBuf)> = Vec::new();
 
     let mut users = match tokio::fs::read_dir(data_path).await {
         Ok(r) => r,
@@ -37,6 +42,9 @@ async fn collect_pending(data_path: &Path) -> Vec<PathBuf> {
     };
 
     while let Ok(Some(user_entry)) = users.next_entry().await {
+        let Some(slug) = user_entry.file_name().to_str().map(str::to_owned) else {
+            continue;
+        };
         let open_dir = user_entry.path().join("open");
         let mut cases = match tokio::fs::read_dir(&open_dir).await {
             Ok(r) => r,
@@ -61,13 +69,15 @@ async fn collect_pending(data_path: &Path) -> Vec<PathBuf> {
                 }
                 let transcript = path.with_extension("transcript.txt");
                 if !transcript.exists() {
-                    out.push(path);
+                    out.push((slug.clone(), path));
                 }
             }
         }
     }
 
-    // Oldest first (filenames are ISO timestamps).
-    out.sort();
+    // Oldest first (filenames are ISO timestamps). Path ordering puts same-user
+    // entries together, which is fine — we still process strictly by timestamp
+    // within a user directory.
+    out.sort_by(|a, b| a.1.cmp(&b.1));
     out
 }