Fix timestamp offset bug in BLFWriter (#1921)

Fix timestamp offset bug in BLFWriter

In the BLF files, the frame timestamp are stored as a delta from the "start time". The "start time" stored in the file is the first frame timestamp rounded to 3 decimals. When we calculate the timestamp delta for each frame, we were previously calculating it using the non-rounded "start time".

This new code, use the "start time" as the first frame timestamp truncated to 3 decimals.

Author: pierreluctg

can/io/blf.py

@@ -530,7 +530,10 @@ def _add_object(self, obj_type, data, timestamp=None):
         if timestamp is None:
             timestamp = self.stop_timestamp or time.time()
         if self.start_timestamp is None:
-            self.start_timestamp = timestamp
+            # Save start timestamp using the same precision as the BLF format
+            # Truncating to milliseconds to avoid rounding errors when calculating
+            # the timestamp difference
+            self.start_timestamp = int(timestamp * 1000) / 1000
         self.stop_timestamp = timestamp
         timestamp = int((timestamp - self.start_timestamp) * 1e9)
         header_size = OBJ_HEADER_BASE_STRUCT.size + OBJ_HEADER_V1_STRUCT.size

test/data/example_data.py

@@ -160,7 +160,7 @@ def sort_messages(messages):
 
 TEST_MESSAGES_ERROR_FRAMES = sort_messages(
     [
-        Message(is_error_frame=True),
+        Message(is_error_frame=True, timestamp=TEST_TIME),
         Message(is_error_frame=True, timestamp=TEST_TIME + 0.170),
         Message(is_error_frame=True, timestamp=TEST_TIME + 17.157),
     ]