fix: correct stream implementation when dataset is empty (#103)

For saving and loading

Author: nfrasser

cryosparc/dataset.py

@@ -685,6 +685,9 @@ def _load_stream(
         # Calling addrows separately to minimizes column-based
         # allocations, improves performance by ~20%
         dset = cls.allocate(0, descr)
+        if header["length"] == 0:
+            return dset  # no more data to load
+
         data = dset._data
         data.addrows(header["length"])
         loader = Stream(data)
@@ -799,6 +802,9 @@ def stream(self, compression: Literal["lz4", None] = None) -> Generator[bytes, N
         yield u32bytesle(len(header))
         yield header
 
+        if len(self) == 0:
+            return  # empty dataset, don't yield anything
+
         for f in self:
             fielddata: "MemoryView"
             if f in compressed_fields:

tests/test_dataset.py

@@ -56,6 +56,27 @@ def small_dset_stream(small_dset):
     return stream
 
 
+@pytest.fixture
+def empty_dset():
+    field3 = "long/fieldwithsuperduperlongcolumnnamethatislongandtestable"
+    return Dataset.allocate(
+        0,
+        fields=[
+            ("field/1", "u8", (2,)),
+            ("field/2", "f4"),
+            (field3, "O"),
+        ],
+    )
+
+
+@pytest.fixture
+def empty_dset_stream(empty_dset):
+    stream = BytesIO()
+    empty_dset.save(stream, format=CSDAT_FORMAT)
+    stream.seek(0)
+    return stream
+
+
 def test_allocate():
     storage = Dataset.allocate(size=2000000, fields=[("field1", "u8"), ("field2", "f4"), ("field3", "O")])
     assert storage is not None
@@ -270,6 +291,11 @@ def test_load_stream_fields(small_dset, small_dset_stream):
     assert result == small_dset.filter_fields(["field/2"], copy=True)
 
 
+def test_load_empty_stream(empty_dset, empty_dset_stream):
+    result = Dataset.load(empty_dset_stream)
+    assert result == empty_dset
+
+
 def test_pickle_unpickle():
     import pickle