Rework property tests

Author: dcherian

tests/strategies.py

@@ -13,22 +13,22 @@
 
 Chunks = tuple[tuple[int, ...], ...]
 
-
-def supported_dtypes() -> st.SearchStrategy[np.dtype]:
-    return (
-        npst.integer_dtypes(endianness="=")
-        | npst.unsigned_integer_dtypes(endianness="=")
-        | npst.floating_dtypes(endianness="=", sizes=(32, 64))
-        | npst.complex_number_dtypes(endianness="=")
-        | npst.datetime64_dtypes(endianness="=")
-        | npst.timedelta64_dtypes(endianness="=")
-        | npst.unicode_string_dtypes(endianness="=")
-    )
-
-
+numeric_dtypes = (
+    npst.integer_dtypes(endianness="=")
+    | npst.unsigned_integer_dtypes(endianness="=")
+    | npst.floating_dtypes(endianness="=", sizes=(32, 64))
+    # TODO: add complex here not in supported_dtypes
+)
 # TODO: stop excluding everything but U
-array_dtypes = supported_dtypes().filter(lambda x: x.kind not in "cU")
-by_dtype_st = supported_dtypes()
+numeric_like_dtypes = (
+    numeric_dtypes | npst.datetime64_dtypes(endianness="=") | npst.timedelta64_dtypes(endianness="=")
+)
+supported_dtypes = (
+    numeric_like_dtypes
+    | npst.unicode_string_dtypes(endianness="=")
+    | npst.complex_number_dtypes(endianness="=")
+)
+by_dtype_st = supported_dtypes
 
 NON_NUMPY_FUNCS = [
     "first",
@@ -43,12 +43,15 @@ def supported_dtypes() -> st.SearchStrategy[np.dtype]:
 
 func_st = st.sampled_from([f for f in ALL_FUNCS if f not in NON_NUMPY_FUNCS and f not in SKIPPED_FUNCS])
 numeric_arrays = npst.arrays(
-    elements={"allow_subnormal": False}, shape=npst.array_shapes(), dtype=array_dtypes
+    elements={"allow_subnormal": False}, shape=npst.array_shapes(), dtype=numeric_dtypes
+)
+numeric_like_arrays = npst.arrays(
+    elements={"allow_subnormal": False}, shape=npst.array_shapes(), dtype=numeric_like_dtypes
 )
 all_arrays = npst.arrays(
     elements={"allow_subnormal": False},
     shape=npst.array_shapes(),
-    dtype=supported_dtypes(),
+    dtype=supported_dtypes,
 )
 
 calendars = st.sampled_from(

tests/test_properties.py

@@ -20,7 +20,7 @@
 from flox.xrutils import isnull, notnull
 
 from . import assert_equal
-from .strategies import array_dtypes, by_arrays, chunked_arrays, func_st, numeric_arrays
+from .strategies import by_arrays, chunked_arrays, func_st, numeric_dtypes, numeric_like_arrays
 from .strategies import chunks as chunks_strategy
 
 dask.config.set(scheduler="sync")
@@ -66,7 +66,7 @@ def not_overflowing_array(array: np.ndarray[Any, Any]) -> bool:
 
 @given(
     data=st.data(),
-    array=st.one_of(numeric_arrays, chunked_arrays(arrays=numeric_arrays)),
+    array=st.one_of(numeric_like_arrays, chunked_arrays(arrays=numeric_like_arrays)),
     func=func_st,
 )
 def test_groupby_reduce(data, array, func: str) -> None:
@@ -136,7 +136,7 @@ def test_groupby_reduce(data, array, func: str) -> None:
 
 @given(
     data=st.data(),
-    array=chunked_arrays(arrays=numeric_arrays),
+    array=chunked_arrays(arrays=numeric_like_arrays),
     func=func_st,
 )
 def test_groupby_reduce_numpy_vs_dask(data, array, func: str) -> None:
@@ -170,7 +170,7 @@ def test_groupby_reduce_numpy_vs_dask(data, array, func: str) -> None:
 @settings(report_multiple_bugs=False)
 @given(
     data=st.data(),
-    array=chunked_arrays(arrays=numeric_arrays),
+    array=chunked_arrays(arrays=numeric_like_arrays),
     func=st.sampled_from(tuple(NUMPY_SCAN_FUNCS)),
 )
 def test_scans(data, array: dask.array.Array, func: str) -> None:
@@ -297,8 +297,8 @@ def test_first_last_useless(data, func):
 @given(
     func=st.sampled_from(["sum", "prod", "nansum", "nanprod"]),
     engine=st.sampled_from(["numpy", "flox"]),
-    array_dtype=st.none() | array_dtypes,
-    dtype=st.none() | array_dtypes,
+    array_dtype=st.none() | numeric_dtypes,
+    dtype=st.none() | numeric_dtypes,
 )
 def test_agg_dtype_specified(func, array_dtype, dtype, engine):
     # regression test for GH388