From 69178945e5b64e73413169056673714a3db2d8cf Mon Sep 17 00:00:00 2001
From: Dawn <jiaxin.duan@psi.ch>
Date: Thu, 13 Aug 2026 10:31:57 +0200
Subject: [PATCH] fix: prevent libzmq abort and per-frame decode errors in
 prediction subscriber

Sniff for '{' before JSON-parsing zmq parts: JPEG frames start with 0xff
and raised UnicodeDecodeError on every frame. Stop closing the socket
from the GUI thread in stop(): zmq sockets are not thread-safe and the
concurrent close/recv tripped the signaler.cpp POLLIN assertion (core
dump); run() closes it in its own thread via the 500ms RCVTIMEO.

uv.lock: sync version to 0.8.3.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 src/aare/gui/threads/prediction_subscriber.py | 14 ++++++-----
 tests/unit/gui/test_prediction_subscriber.py  | 24 +++++++++++++++++++
 uv.lock                                       |  2 +-
 3 files changed, 33 insertions(+), 7 deletions(-)

diff --git a/src/aare/gui/threads/prediction_subscriber.py b/src/aare/gui/threads/prediction_subscriber.py
index 044fe99f..29fe7876 100644
--- a/src/aare/gui/threads/prediction_subscriber.py
+++ b/src/aare/gui/threads/prediction_subscriber.py
@@ -87,6 +87,10 @@ class PredictionSubscriber(QThread):
         self._measure_focus = enabled
 
     def _try_parse_json(self, part: bytes) -> dict | None:
+        # Binary frames (JPEG starts with 0xff) are not JSON; decoding them as
+        # utf-8 raises UnicodeDecodeError on every frame. Cheap sniff instead.
+        if not part.lstrip()[:1] == b"{":
+            return None
         try:
             decoded = json.loads(part.decode("utf-8"))
             return decoded if isinstance(decoded, dict) else None
@@ -254,11 +258,9 @@ class PredictionSubscriber(QThread):
         self.running = False
         self.requestInterruption()
 
-        try:
-            if self._sock is not None:
-                self._sock.close(0)
-        except Exception:
-            logger.debug("Error while stopping the prediction subscriber", exc_info=True)
-
+        # Do NOT close the socket here: zmq sockets are not thread-safe, and
+        # closing from the GUI thread while run() uses it aborts libzmq
+        # (Assertion failed: pfd.revents & POLLIN, signaler.cpp). RCVTIMEO=500ms
+        # guarantees run() notices running=False and closes it in its own thread.
         if not self.wait(1500):
             logger.warning("PredictionSubscriber did not stop within timeout")
diff --git a/tests/unit/gui/test_prediction_subscriber.py b/tests/unit/gui/test_prediction_subscriber.py
index e69de29b..6617ba55 100644
--- a/tests/unit/gui/test_prediction_subscriber.py
+++ b/tests/unit/gui/test_prediction_subscriber.py
@@ -0,0 +1,24 @@
+"""Checks the binary-frame sniff in _try_parse_json: JPEG frames (0xff lead
+byte) must be rejected without attempting utf-8 decode (previously raised
+UnicodeDecodeError on every frame)."""
+
+from aare.gui.threads.prediction_subscriber import PredictionSubscriber
+
+
+def _parse(part: bytes):
+    # Called unbound: _try_parse_json touches no instance state, so no
+    # QThread/zmq construction is needed.
+    return PredictionSubscriber._try_parse_json(object(), part)
+
+
+def test_binary_jpeg_frame_is_not_json():
+    assert _parse(b"\xff\xd8\xff\xe0somejpegbytes") is None
+
+
+def test_json_dict_is_parsed():
+    assert _parse(b'{"encoding": "jpeg"}') == {"encoding": "jpeg"}
+
+
+def test_json_non_dict_is_rejected():
+    assert _parse(b"[1, 2]") is None
+    assert _parse(b"") is None
diff --git a/uv.lock b/uv.lock
index 9a39da31..72a73a21 100644
--- a/uv.lock
+++ b/uv.lock
@@ -31,7 +31,7 @@ wheels = [
 
 [[package]]
 name = "aaredaq"
-version = "0.8.1"
+version = "0.8.3"
 source = { editable = "." }
 dependencies = [
     { name = "aarecommon" },