Update casting behaviour

tests/test_properties.py

@@ -43,7 +43,9 @@ def not_overflowing_array(array) -> bool:
     else:
         return True
 
-    return bool(np.all((array < info.max / array.size) & (array > info.min / array.size)))
+    result = bool(np.all((array < info.max / array.size) & (array > info.min / array.size)))
+    # note(f"returning {result}, {array.min()} vs {info.min}, {array.max()} vs {info.max}")
+    return result
 
 
 @settings(suppress_health_check=[HealthCheck.filter_too_much])
@@ -57,14 +59,19 @@ def not_overflowing_array(array) -> bool:
 def test_groupby_reduce(array, dtype, func):
     # overflow behaviour differs between bincount and sum (for example)
     assume(not_overflowing_array(array))
-    # numpy-groupies always does the calculation in float64
-    assume(func != "var" and "f2" not in array.dtype.str)
     # arg* with nans in array are weird
     assume("arg" not in func and not np.any(np.isnan(array).ravel()))
 
     axis = -1
     by = np.ones((array.shape[-1],), dtype=dtype)
     kwargs = {"q": 0.8} if "quantile" in func else {}
+    # numpy-groupies always does the calculation in float64
+    if ("var" in func or "std" in func or "sum" in func) and array.dtype.kinde == "f":
+        # bincount accumulates in float64
+        kwargs.setdefault("dtype", np.float64)
+        cast_to = array.dtype
+    else:
+        cast_to = None
 
     with np.errstate(invalid="ignore", divide="ignore"):
         actual, _ = groupby_reduce(
@@ -73,4 +80,6 @@ def test_groupby_reduce(array, dtype, func):
         expected = getattr(np, func)(array, axis=axis, keepdims=True, **kwargs)
     note(("expected: ", expected, "actual: ", actual))
     tolerance = {"rtol": 1e-13, "atol": 1e-16} if "var" in func or "std" in func else {}
+    if cast_to:
+        actual = actual.astype(cast_to)
     assert_equal(expected, actual, tolerance)