fix: batch 29 third-pass audit — 12 issues fixed across 8 files

chat_editor.py: multi-turn chat was completely broken — query_llm()
doesn't accept messages kwarg (TypeError at runtime). Rewrote to
embed conversation history in the prompt. Added _sessions_lock for
thread-safe session dict access.

Bare ffmpeg/ffprobe (3 final): styled_captions.py Popen used bare
"ffmpeg" (bypasses run_ffmpeg auto-resolve), audio_enhance.py used
bare "ffprobe" despite resolving path, color_match.py fallback path
used bare "ffmpeg" via subprocess.run.

system.py: removed double rate_limit_release on validation error in
install_whisper and whisper_reinstall (acquired flag handles it).

animated_captions.py: bare "ffmpeg" replaced with get_ffmpeg_path(),
_group_words_into_lines bare dict access changed to .get() defaults.

engine_registry.py: clear_cache() now acquires self._lock.

server.py: PID file open() calls now have encoding="utf-8".

Author: SysAdminDoc

opencut/core/animated_captions.py

@@ -17,7 +17,7 @@
 import tempfile
 from typing import Callable, Dict, List, Optional
 
-from opencut.helpers import ensure_package, get_video_info, run_ffmpeg
+from opencut.helpers import ensure_package, get_ffmpeg_path, get_video_info, run_ffmpeg
 
 logger = logging.getLogger("opencut")
 
@@ -264,7 +264,7 @@ def render_animated_captions(
 
     try:
         run_ffmpeg([
-            "ffmpeg", "-hide_banner", "-loglevel", "error", "-y",
+            get_ffmpeg_path(), "-hide_banner", "-loglevel", "error", "-y",
             "-i", tmp_video, "-i", video_path,
             "-map", "0:v", "-map", "1:a?",
             "-c:v", "libx264", "-crf", "18", "-preset", "medium",
@@ -295,16 +295,16 @@ def _group_words_into_lines(words: List[Dict], max_per_line: int) -> List[Dict]:
         if len(current_words) >= max_per_line:
             lines.append({
                 "words": current_words,
-                "start": current_words[0]["start"],
-                "end": current_words[-1]["end"],
+                "start": current_words[0].get("start", 0),
+                "end": current_words[-1].get("end", 0),
             })
             current_words = []
 
     if current_words:
         lines.append({
             "words": current_words,
-            "start": current_words[0]["start"],
-            "end": current_words[-1]["end"],
+            "start": current_words[0].get("start", 0),
+            "end": current_words[-1].get("end", 0),
         })
 
     return lines

opencut/core/audio_enhance.py

@@ -50,7 +50,7 @@ def _is_video(filepath):
     try:
         ffprobe_path = get_ffprobe_path()
         result = subprocess.run(
-            ["ffprobe", "-v", "quiet", "-select_streams", "v:0",
+            [ffprobe_path, "-v", "quiet", "-select_streams", "v:0",
              "-show_entries", "stream=codec_type", "-of", "csv=p=0", filepath],
             capture_output=True, text=True, timeout=10, env=_binary_env(ffprobe_path), check=False,
         )

opencut/core/chat_editor.py

@@ -25,6 +25,7 @@
 """
 
 import logging
+import threading
 import time
 from dataclasses import dataclass, field
 from typing import Dict, List, Optional
@@ -105,14 +106,15 @@ def get_llm_messages(self) -> List[Dict]:
 
 # Active sessions (in-memory, keyed by session_id)
 _sessions: Dict[str, ChatSession] = {}
+_sessions_lock = threading.Lock()
 
 # Session TTL: evict sessions inactive for more than 2 hours
 _SESSION_TTL = 2 * 60 * 60  # seconds
 _MAX_SESSIONS = 100
 
 
 def _evict_stale_sessions():
-    """Remove sessions that haven't been active within the TTL."""
+    """Remove sessions that haven't been active within the TTL. Must hold _sessions_lock."""
     now = time.time()
     stale = [
         sid for sid, s in _sessions.items()
@@ -136,28 +138,30 @@ def get_or_create_session(
     clip_info: Optional[Dict] = None,
 ) -> ChatSession:
     """Get an existing chat session or create a new one."""
-    _evict_stale_sessions()
-
-    if session_id in _sessions:
-        session = _sessions[session_id]
-        # Update filepath if changed
-        if filepath and filepath != session.filepath:
-            session.filepath = filepath
-            session.context = clip_info or {}
+    with _sessions_lock:
+        _evict_stale_sessions()
+
+        if session_id in _sessions:
+            session = _sessions[session_id]
+            # Update filepath if changed
+            if filepath and filepath != session.filepath:
+                session.filepath = filepath
+                session.context = clip_info or {}
+            return session
+
+        session = ChatSession(
+            session_id=session_id,
+            filepath=filepath,
+            context=clip_info or {},
+        )
+        _sessions[session_id] = session
         return session
 
-    session = ChatSession(
-        session_id=session_id,
-        filepath=filepath,
-        context=clip_info or {},
-    )
-    _sessions[session_id] = session
-    return session
-
 
 def clear_session(session_id: str):
     """Clear a chat session's history."""
-    _sessions.pop(session_id, None)
+    with _sessions_lock:
+        _sessions.pop(session_id, None)
 
 
 def chat(
@@ -209,12 +213,24 @@ def chat(
     # Query LLM with full conversation history
     messages = session.get_llm_messages()
 
+    # Build a combined prompt with conversation context
+    # query_llm() only supports single prompt + system_prompt,
+    # so we include recent history in the prompt itself
+    context_msgs = messages[1:-1]  # Skip system and current user msg
+    if context_msgs:
+        history_text = "\n".join(
+            f"{'User' if m['role'] == 'user' else 'Assistant'}: {m['content']}"
+            for m in context_msgs[-10:]  # Last 10 messages for context
+        )
+        full_prompt = f"Previous conversation:\n{history_text}\n\nUser: {user_message}"
+    else:
+        full_prompt = user_message
+
     try:
         response = query_llm(
-            prompt=user_message,
+            prompt=full_prompt,
             config=llm_config,
             system_prompt=system_prompt,
-            messages=messages[1:],  # Skip system (passed separately)
         )
         response_text = response.text if hasattr(response, "text") else str(response)
     except Exception as e:

opencut/core/color_match.py

@@ -11,12 +11,13 @@
 from typing import List
 
 try:
-    from ..helpers import run_ffmpeg
+    from ..helpers import get_ffmpeg_path, run_ffmpeg
 except ImportError:
     try:
-        from opencut.helpers import run_ffmpeg
+        from opencut.helpers import get_ffmpeg_path, run_ffmpeg
     except ImportError:
         run_ffmpeg = None  # type: ignore
+        get_ffmpeg_path = None  # type: ignore
 
 logger = logging.getLogger("opencut")
 
@@ -311,8 +312,9 @@ def color_match_video(
     logger.info("Merging audio from source into color-matched output")
 
     # Merge original audio from source using FFmpeg
+    _ffmpeg = get_ffmpeg_path() if get_ffmpeg_path else "ffmpeg"
     cmd = [
-        "ffmpeg", "-hide_banner", "-loglevel", "error", "-y",
+        _ffmpeg, "-hide_banner", "-loglevel", "error", "-y",
         "-i", temp_video,
         "-i", source_path,
         "-map", "0:v:0",

opencut/core/engine_registry.py

@@ -203,7 +203,8 @@ def export_preferences(self) -> Dict[str, str]:
 
     def clear_cache(self):
         """Clear the availability cache."""
-        self._availability_cache.clear()
+        with self._lock:
+            self._availability_cache.clear()
 
 
 # ---------------------------------------------------------------------------

opencut/core/styled_captions.py

@@ -19,6 +19,8 @@
 from dataclasses import dataclass
 from typing import Callable, Dict, List, Optional, Set, Tuple
 
+from opencut.helpers import get_ffmpeg_path
+
 from .captions import TranscriptionResult, Word
 
 logger = logging.getLogger("opencut.styled_captions")
@@ -992,7 +994,7 @@ def render_styled_caption_video(
     # Pipe raw RGBA frames to FFmpeg
     # -------------------------------------------------------------------
     cmd = [
-        "ffmpeg", "-hide_banner", "-loglevel", "warning", "-y",
+        get_ffmpeg_path(), "-hide_banner", "-loglevel", "warning", "-y",
         "-f", "rawvideo",
         "-pix_fmt", "rgba",
         "-s", f"{video_width}x{video_height}",

opencut/routes/system.py

@@ -749,7 +749,6 @@ def install_whisper(job_id, filepath, data):
     backend = data.get("backend", "faster-whisper")
     allowed = {"faster-whisper", "openai-whisper", "whisperx"}
     if backend not in allowed:
-        rate_limit_release("model_install")
         raise ValueError(f"Unknown backend: {backend}")
 
     try:
@@ -1050,7 +1049,6 @@ def whisper_reinstall(job_id, filepath, data):
     backend = data.get("backend", "faster-whisper")
     allowed_backends = {"faster-whisper", "openai-whisper", "whisperx"}
     if backend not in allowed_backends:
-        rate_limit_release("model_install")
         raise ValueError(f"Unknown backend: {backend}")
 
     cpu_mode = data.get("cpu_mode", False)

opencut/server.py

@@ -310,7 +310,7 @@ def _write_pid(port: int):
         # Atomic write: temp file + rename to prevent partial reads
         fd, tmp_path = tempfile.mkstemp(dir=pid_dir, suffix=".tmp", prefix="server.pid.")
         try:
-            with os.fdopen(fd, "w") as f:
+            with os.fdopen(fd, "w", encoding="utf-8") as f:
                 f.write(f"{os.getpid()}\n{port}\n")
             os.replace(tmp_path, PID_FILE)
         except BaseException:
@@ -326,7 +326,7 @@ def _read_pid():
     """Read PID and port from file. Returns (pid, port) or (None, None)."""
     try:
         if os.path.exists(PID_FILE):
-            with open(PID_FILE, "r") as f:
+            with open(PID_FILE, "r", encoding="utf-8") as f:
                 lines = f.read().strip().split("\n")
             pid = int(lines[0]) if lines else None
             port = int(lines[1]) if len(lines) > 1 else None