S64: fix 5 bugs — dedup v11.0, RAM guard, watchdog, engine_log, status

app.py

@@ -96,13 +96,11 @@ ENGINE_SCRIPTS = {
     "v11.1": BASE / "true_engine_itiqan_v2_fixed.py",
     "v11.2": BASE / "engine_isteidad_v12.py",
     "v10.0": BASE / "engine_v100.py",
-    "v11.0": BASE / "engine_tajalli_v1.py",
     "v9.0":  BASE / "engine_v90.py",
     "v8.5":  BASE / "engine_v85.py",
-    "v8.4":  BASE / "engine_v84.py",
     "v8.0":  BASE / "engine_v80.py",
     "v7.0":  BASE / "engine_v70.py",
-}
+}  # S64: removed duplicate v11.0 + dead v8.4
 REF_DIR    = BASE / "reference_audio"
 CHUNK_SIZE = 4 * 1024 * 1024       # 4 MB upload chunks
 MAX_DOWNLOAD_CHUNK = 32 * 1024 * 1024  # 32 MB max per download_chunk call
@@ -519,10 +517,10 @@ def _run_engine(job_id):
 
     # ── RAM guard: require ≥ 3.5 GB available before launching subprocess ──
     ram_gb = _available_ram_gb()
-    if ram_gb < 0.5:
+    if ram_gb < 1.5:  # S64: raised from 0.5 — engines need ~3.5GB
         with JOBS_LOCK:
             job["status"] = "error"
-            job["error"]  = f"Insufficient RAM ({ram_gb:.1f} GB free, need 0.5 GB)"
+            job["error"]  = f"Insufficient RAM ({ram_gb:.1f} GB free, need 1.5 GB)"
             job["label"]  = "خطأ: ذاكرة غير كافية — أعد المحاولة لاحقاً"
         return
 
@@ -555,16 +553,18 @@ def _run_engine(job_id):
             for rf in ref_files[:3]:
                 cmd += ["--ref", str(rf)]
 
+            # S64: watchdog — kill engine if it runs > 35 min
+            _ENGINE_TIMEOUT = 35 * 60  # 35 minutes
             proc = subprocess.Popen(
                 cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True
             )
 
-            engine_lines = []   # DEBUG: capture all output
+            engine_lines = []   # last 60 lines for debug
             for line in proc.stdout:
                 line = line.strip()
-                engine_lines.append(line)        # DEBUG
-                if len(engine_lines) > 60:       # DEBUG: keep last 60 lines
-                    engine_lines.pop(0)          # DEBUG
+                engine_lines.append(line)
+                if len(engine_lines) > 60:
+                    engine_lines.pop(0)
                 # Progress markers — plain scalar writes, no lock needed
                 if "Pass 1" in line or "[٧]" in line or "[١]" in line:
                     job["progress"] = 45; job["label"] = "Pass 1 — تحليل الطيف..."
@@ -595,17 +595,31 @@ def _run_engine(job_id):
                         except Exception:
                             pass
 
-            proc.wait()
+            # S64: timed wait — kill if engine hangs past timeout
+            try:
+                proc.wait(timeout=_ENGINE_TIMEOUT)
+            except subprocess.TimeoutExpired:
+                proc.kill()
+                proc.wait()
+                job["engine_error"] = "Engine timed out after 35 min — killed"
+                job["engine_log"]   = engine_lines
+                with JOBS_LOCK:
+                    job["status"] = "error"
+                    job["error"]  = "Engine timed out (35 min limit)"
+                    job["label"]  = "خطأ: انتهت مهلة المحرك — أعد المحاولة"
+                return
             _out = Path(job["out_path"])
             # engine_v100 exits 1 for score<85 but still writes the file.
             # Use file existence + size as the real success signal.
             if _out.exists() and _out.stat().st_size > 0:
                 success = True
             else:
-                job["engine_rc"] = proc.returncode
+                job["engine_rc"]  = proc.returncode
+                job["engine_log"] = engine_lines  # S64: persist last 60 lines
 
         except Exception as exc:
             job["engine_error"] = str(exc)
+            job["engine_log"]   = engine_lines  # S64: persist for /status debug
 
     if not success:
         # Status transition: running → error (held under lock)
@@ -664,10 +678,13 @@ def status(job_id):
     if "error" in job:
         resp["error"] = job["error"]
 
-    # DEBUG: expose last engine output lines so we can see the crash
-    if job.get("status") == "error" and "engine_log" in job:
-        resp["engine_log"] = job["engine_log"]
-        resp["engine_rc"]  = job.get("engine_rc")
+    # S64: expose engine debug info on error
+    if job.get("status") == "error":
+        if "engine_log" in job:
+            resp["engine_log"] = job["engine_log"]   # last 60 stdout lines
+        if "engine_error" in job:
+            resp["engine_error"] = job["engine_error"]  # exception message
+        resp["engine_rc"] = job.get("engine_rc")
 
     return jsonify(resp)