feat!: raise error to eager len calls (#416)

src/dask_awkward/lib/core.py

@@ -704,7 +704,12 @@ def repartition(
 
     def __len__(self) -> int:
         if not self.known_divisions:
-            self.eager_compute_divisions()
+            raise NotImplementedError(
+                "Cannot determine length of collection with unknown partition sizes without executing the graph.\n"
+                "Use `dask_awkward.num(..., axis=0)` if you want a lazy Scalar of the length.\n"
+                "If you want to eagerly compute the partition sizes to have the ability to call `len` on the collection"
+                ", use `.eager_compute_divisions()` on the collection."
+            )
         return self.divisions[-1]  # type: ignore
 
     def _shorttypestr(self, max: int = 10) -> str:

tests/test_core.py

@@ -100,7 +100,20 @@ def test_compute_typetracer(daa: Array) -> None:
 
 def test_len(ndjson_points_file: str) -> None:
     daa = dak.from_json([ndjson_points_file] * 2)
-    assert len(daa) == 10
+    assert not daa.known_divisions
+    with pytest.raises(
+        NotImplementedError,
+        match=(
+            "Cannot determine length of collection with unknown partition sizes without executing the graph.\\n"
+            "Use `dask_awkward.num\\(\\.\\.\\., axis=0\\)` if you want a lazy Scalar of the length.\\n"
+            "If you want to eagerly compute the partition sizes to have the ability to call `len` on the collection"
+            ", use `\\.eager_compute_divisions\\(\\)` on the collection."
+        ),
+    ):
+        assert len(daa) == 10
+    daa.eager_compute_divisions()
+    assert daa.known_divisions
+    assert len(daa) == 10  # type: ignore
 
 
 def test_meta_exists(daa: Array) -> None:

tests/test_io.py

@@ -281,12 +281,18 @@ def test_from_awkward_empty_array(daa: dak.Array) -> None:
     assert len(c1) == 0
     a1 = dak.from_awkward(c1, npartitions=1)
     assert_eq(a1, c1)
-    assert len(a1) == 0
+    assert not a1.known_divisions
+    a1.eager_compute_divisions()
+    assert a1.known_divisions
+    assert len(a1) == 0  # type: ignore
 
     # with a form
     c2 = ak.typetracer.length_zero_if_typetracer(daa.layout)
     assert len(c2) == 0
     a2 = dak.from_awkward(c2, npartitions=1)
+    assert not a2.known_divisions
+    a2.eager_compute_divisions()
+    assert a2.known_divisions
     assert len(a2) == 0
     daa.layout.form == a2.layout.form
 

tests/test_io_json.py

@@ -58,6 +58,18 @@ def concrete_data(json_data_dir: Path) -> ak.Array:
 def test_json_sanity(json_data_dir: Path, concrete_data: ak.Array) -> None:
     source = os.path.join(str(json_data_dir))
     ds = dak.from_json(source)
+    assert not ds.known_divisions
+    with pytest.raises(
+        NotImplementedError,
+        match=(
+            "Cannot determine length of collection with unknown partition sizes without executing the graph.\\n"
+            "Use `dask_awkward.num\\(\\.\\.\\., axis=0\\)` if you want a lazy Scalar of the length.\\n"
+            "If you want to eagerly compute the partition sizes to have the ability to call `len` on the collection"
+            ", use `\\.eager_compute_divisions\\(\\)` on the collection."
+        ),
+    ):
+        assert ds
+    ds.eager_compute_divisions()
     assert ds
 
     assert_eq(ds, concrete_data)