IQ.Pilot Release Commit @ 717ce45

iqpilot/cereal/messaging/messaging.h

@@ -12,6 +12,9 @@
 #include "common/timing.h"
 #include "msgq/ipc.h"
 
+const char *messaging_registry_tag();
+bool messaging_has_service(const char *name);
+
 class SubMaster {
 public:
   SubMaster(const std::vector<const char *> &service_list, const std::vector<const char *> &poll = {},

iqpilot/cereal/messaging/socketmaster.cc

@@ -49,7 +49,7 @@ SubMaster::SubMaster(const std::vector<const char *> &service_list, const std::v
   poller_ = Poller::create();
   for (auto name : service_list) {
     if (services.count(std::string(name)) == 0) {
-      fprintf(stderr, "SubMaster: unknown service '%s', skipping subscription\n", name);
+      fprintf(stderr, "SubMaster: unknown service '%s' in %s, skipping subscription\n", name, SERVICES_REGISTRY_TAG);
       continue;
     }
 
@@ -71,6 +71,14 @@ SubMaster::SubMaster(const std::vector<const char *> &service_list, const std::v
   }
 }
 
+const char *messaging_registry_tag() {
+  return SERVICES_REGISTRY_TAG;
+}
+
+bool messaging_has_service(const char *name) {
+  return services.count(std::string(name)) != 0;
+}
+
 void SubMaster::update(int timeout) {
   for (auto &kv : messages_) kv.second->updated = false;
 

iqpilot/cereal/services.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+import hashlib
 from enum import IntEnum
 from typing import Optional
 
@@ -134,6 +135,22 @@ SERVICE_LIST = {name: Service(*vals) for
                 idx, (name, vals) in enumerate(_services.items())}
 
 
+REGISTRY_TAG_PREFIX = "IQ_SERVICES_REGISTRY:"
+
+
+def registry_hash() -> str:
+  digest = hashlib.sha256()
+  for name in sorted(SERVICE_LIST):
+    v = SERVICE_LIST[name]
+    decimation = -1 if v.decimation is None else v.decimation
+    digest.update(f"{name}|{int(v.should_log)}|{v.frequency:f}|{decimation}|{int(v.queue_size)}\n".encode())
+  return digest.hexdigest()[:16]
+
+
+def registry_tag() -> str:
+  return REGISTRY_TAG_PREFIX + registry_hash()
+
+
 def build_header():
   h = ""
   h += "/* THIS IS AN AUTOGENERATED FILE, PLEASE EDIT services.py */\n"
@@ -151,6 +168,7 @@ def build_header():
     h += '  { "%s", {"%s", %s, %f, %d, %d}},\n' % \
          (k, k, should_log, v.frequency, decimation, v.queue_size)
   h += "};\n"
+  h += f'static const char SERVICES_REGISTRY_TAG[] = "{registry_tag()}";\n'
 
   h += "#endif\n"
   return h

iqpilot/selfdrive/iqlocd/atlas_loc_core.cc

@@ -734,6 +734,13 @@ int AtlasLocator::run() {
   const std::initializer_list<const char *> service_list = {gps_location_socket, "cameraOdometry", "extrinsicsCalibration",
                                                           "carState", "accelerometer", "gyroscope"};
 
+  for (const char *service : service_list) {
+    if (!messaging_has_service(service)) {
+      LOGE("service '%s' is missing from the compiled service registry %s: stale build, rebuild required", service, messaging_registry_tag());
+      return 2;
+    }
+  }
+
   SubMaster sm(service_list, {}, nullptr, {gps_location_socket});
   PubMaster pm({"iqLiveLocation"});
 

iqpilot/selfdrive/iqlocd/tests/test_iqlocd.py

@@ -128,12 +128,13 @@ class TestIQLocdProc:
     if case in ("odometry_short", "odometry_empty", "odometry_nan", "odometry_nan_std"):
       msg = messaging.new_message("cameraOdometry")
       odo = msg.cameraOdometry
-      if case == "odometry_short":
-        odo.rot = [0.0] * 6; odo.trans = [0.0] * 6; odo.rotStd = [0.01] * 6; odo.transStd = [0.01] * 6
-      elif case == "odometry_nan":
-        odo.rot = [float("nan"), 0.0, 0.0]; odo.trans = [0.0] * 3; odo.rotStd = [0.01] * 3; odo.transStd = [0.01] * 3
-      elif case == "odometry_nan_std":
-        odo.rot = [0.0] * 3; odo.trans = [0.0] * 3; odo.rotStd = [float("nan"), 0.01, 0.01]; odo.transStd = [0.01] * 3
+      shapes = {
+        "odometry_short": ([0.0] * 6, [0.0] * 6, [0.01] * 6, [0.01] * 6),
+        "odometry_nan": ([float("nan"), 0.0, 0.0], [0.0] * 3, [0.01] * 3, [0.01] * 3),
+        "odometry_nan_std": ([0.0] * 3, [0.0] * 3, [float("nan"), 0.01, 0.01], [0.01] * 3),
+      }
+      if case in shapes:
+        odo.rot, odo.trans, odo.rotStd, odo.transStd = shapes[case]
     elif case in ("calibration_short", "calibration_nan"):
       msg = messaging.new_message("extrinsicsCalibration")
       msg.extrinsicsCalibration.calStatus = "calibrated"

iqpilot/system/manager/build.py

@@ -51,7 +51,44 @@ class _SilentProgress:
     pass
 
 
-def build(spinner, dirty: bool = False, minimal: bool = False, show_error_window: bool = True) -> None:
+REGISTRY_ARTIFACTS = [
+  "iqpilot/cereal/services.h",
+  "iqpilot/cereal/messaging/socketmaster.o",
+  "iqpilot/cereal/libsocketmaster.a",
+  "iqpilot/cereal/messaging/bridge",
+  "iqpilot/selfdrive/iqlocd/iqlocd",
+  "iqpilot/selfdrive/pandad/pandad",
+  "iqpilot/system/camerad/camerad",
+  "iqpilot/system/loggerd/loggerd",
+  "iqpilot/system/loggerd/encoderd",
+  "iqpilot/system/loggerd/bootlog",
+]
+
+
+def stale_registry_artifacts(basedir: str = BASEDIR) -> list[str]:
+  from iqpilot.cereal.services import REGISTRY_TAG_PREFIX, registry_tag
+  expected = registry_tag().encode()
+  prefix = REGISTRY_TAG_PREFIX.encode()
+  stale = []
+  for rel in REGISTRY_ARTIFACTS:
+    path = os.path.join(basedir, rel)
+    if not os.path.isfile(path):
+      continue
+    with open(path, "rb") as f:
+      data = f.read()
+    if prefix in data and expected not in data:
+      stale.append(rel)
+  return stale
+
+
+def purge_registry_artifacts(stale: list[str], basedir: str = BASEDIR) -> None:
+  for rel in set(stale) | set(REGISTRY_ARTIFACTS[:3]):
+    path = os.path.join(basedir, rel)
+    if os.path.isfile(path):
+      os.remove(path)
+
+
+def build(spinner, dirty: bool = False, minimal: bool = False, show_error_window: bool = True, registry_retry: bool = False) -> None:
   env = os.environ.copy()
   env.pop('PWD', None)
   env['SCONS_PROGRESS'] = "1"
@@ -110,6 +147,22 @@ def build(spinner, dirty: bool = False, minimal: bool = False, show_error_window
         t.wait_for_exit()
     exit(1)
 
+  stale = stale_registry_artifacts()
+  if stale and not registry_retry:
+    cloudlog.error(f"compiled service registry is stale in {', '.join(stale)}, rebuilding messaging")
+    purge_registry_artifacts(stale)
+    build(spinner, dirty, minimal, show_error_window, registry_retry=True)
+    return
+  if stale:
+    error_s = "compiled service registry is still stale after rebuild: " + ", ".join(stale)
+    add_file_handler(cloudlog)
+    cloudlog.error(error_s)
+    spinner.close()
+    if not os.getenv("CI") and show_error_window:
+      with TextWindow("IQ.Pilot failed to build\n \n" + error_s) as t:
+        t.wait_for_exit()
+    exit(1)
+
   # enforce max cache size
   cache_files = [f for f in CACHE_DIR.rglob('*') if f.is_file()]
   cache_files.sort(key=lambda f: f.stat().st_mtime)

iqpilot/system/manager/test/test_services_registry.py

@@ -0,0 +1,45 @@
+import os
+import re
+
+from iqpilot.cereal.services import REGISTRY_TAG_PREFIX, SERVICE_LIST, build_header, registry_hash, registry_tag
+from iqpilot.system.manager.build import REGISTRY_ARTIFACTS, purge_registry_artifacts, stale_registry_artifacts
+
+
+def test_registry_tag_is_derived_from_the_service_list():
+  assert re.fullmatch(r"[0-9a-f]{16}", registry_hash())
+  assert registry_tag() == REGISTRY_TAG_PREFIX + registry_hash()
+  assert "extrinsicsCalibration" in SERVICE_LIST
+
+
+def test_generated_header_embeds_the_tag():
+  header = build_header()
+  assert f'static const char SERVICES_REGISTRY_TAG[] = "{registry_tag()}";' in header
+  assert '{ "extrinsicsCalibration", {"extrinsicsCalibration"' in header
+
+
+def _write(tmp_path, rel, payload):
+  path = tmp_path / rel
+  path.parent.mkdir(parents=True, exist_ok=True)
+  path.write_bytes(payload)
+  return path
+
+
+def test_stale_detection_flags_only_binaries_carrying_an_old_tag(tmp_path):
+  current = registry_tag().encode()
+  old = (REGISTRY_TAG_PREFIX + "0" * 16).encode()
+  _write(tmp_path, "iqpilot/selfdrive/iqlocd/iqlocd", b"\x7fELF" + old)
+  _write(tmp_path, "iqpilot/system/loggerd/loggerd", b"\x7fELF" + current)
+  _write(tmp_path, "iqpilot/system/camerad/camerad", b"\x7fELF no registry linked")
+  assert stale_registry_artifacts(str(tmp_path)) == ["iqpilot/selfdrive/iqlocd/iqlocd"]
+
+
+def test_purge_removes_the_stale_binary_and_the_messaging_table(tmp_path):
+  for rel in REGISTRY_ARTIFACTS:
+    _write(tmp_path, rel, b"x")
+  purge_registry_artifacts(["iqpilot/selfdrive/iqlocd/iqlocd"], str(tmp_path))
+  remaining = [rel for rel in REGISTRY_ARTIFACTS if os.path.isfile(tmp_path / rel)]
+  assert "iqpilot/selfdrive/iqlocd/iqlocd" not in remaining
+  assert "iqpilot/cereal/services.h" not in remaining
+  assert "iqpilot/cereal/messaging/socketmaster.o" not in remaining
+  assert "iqpilot/cereal/libsocketmaster.a" not in remaining
+  assert "iqpilot/system/loggerd/loggerd" in remaining