Build the points before pickling

python/tests/test_transform_to_pickle.py

@@ -4,6 +4,7 @@
 import unittest
 from zoneinfo import ZoneInfo
 
+import pandas as pd
 import pytz
 
 from ..transform_to_pickle import TransformToPickle
@@ -78,5 +79,28 @@ def test_it_should_write_a_gzip_compressed_pickle(self):
                 actual = written.read(2)
             self.assertEqual(b'\x1f\x8b', actual)
 
+    def test_it_should_write_trajectories_with_point_locations(self):
+        # Readers access `trajectory.df.geometry` directly: movingpandas < 0.22 does it in
+        # `TrajectoryStopDetector`, Apps do it in their own code. A geometry movingpandas >= 0.22
+        # left to be built on demand arrives there as None.
+        # arrange
+        with tempfile.TemporaryDirectory() as tmp:
+            output = os.path.join(tmp, 'output_file')
+
+            # act
+            self.sut.convert(
+                input_data_file_name='./python/sample/input4/link.csv',
+                input_meta_file_name='./python/sample/input4/meta.csv',
+                output_file_name=output
+            )
+            actual = pd.read_pickle(output, compression='gzip')
+
+        # assert
+        for trajectory in actual.trajectories:
+            self.assertEqual(0, trajectory.df.geometry.isna().sum(), trajectory.id)
+        # csv: earliest fix of 'X742..deploy_id.56853924.' at 2013-08-08 06:47:31
+        first = actual.trajectories[0].df.geometry.iloc[0]
+        self.assertEqual((48.761833, 69.128117), (first.x, first.y))
+
 if __name__ == '__main__':
     unittest.main()

python/transform_to_pickle.py

@@ -1,5 +1,6 @@
 import os
 
+import geopandas as gpd
 import movingpandas as mpd
 import pandas as pd
 from dataclasses import dataclass
@@ -47,13 +48,19 @@ def adjust_timestamps(self, data, timezone, time_col_name):
         return data
 
     def create_moving_pandas(self, data, projection, track_id_col_name):
+        # Build the points here, not via movingpandas' `x`/`y`: since 0.22 that path leaves the
+        # geometry None and builds it on demand, which a reader on movingpandas < 0.22 - or one
+        # reading `trajectory.df.geometry` directly - never does. Dropping x/y keeps the columns
+        # movingpandas 0.17.2 handed over.
+        points = gpd.GeoDataFrame(
+            data.drop(columns=['coords_x', 'coords_y']),
+            geometry=gpd.points_from_xy(data['coords_x'], data['coords_y']),
+            crs=projection
+        )
         move = mpd.TrajectoryCollection(
-            data,
+            points,
             traj_id_col=track_id_col_name,
-            crs=projection,
-            t='timestamp_utc',  # use our converted timezone column (UTC)
-            x='coords_x',
-            y='coords_y'
+            t='timestamp_utc'  # use our converted timezone column (UTC)
         )
         print(move)
         return move