No fill_value=None; use fill value out-of-bounds (#115)

Author: gjoseph92

stackstac/nodata_reader.py

@@ -1,12 +1,12 @@
-from typing import Optional, Tuple, Type, Union, cast
+from typing import Tuple, Type, Union, cast
 import re
 
 import numpy as np
 from rasterio.windows import Window
 
 from .reader_protocol import Reader
 
-State = Tuple[np.dtype, Optional[Union[int, float]]]
+State = Tuple[np.dtype, Union[int, float]]
 
 
 class NodataReader:
@@ -17,7 +17,7 @@ def __init__(
         self,
         *,
         dtype: np.dtype,
-        fill_value: Optional[Union[int, float]] = None,
+        fill_value: Union[int, float],
         **kwargs,
     ) -> None:
         self.dtype = dtype
@@ -36,13 +36,7 @@ def __setstate__(self, state: State) -> None:
         self.dtype, self.fill_value = state
 
 
-def nodata_for_window(
-    window: Window, fill_value: Optional[Union[int, float]], dtype: np.dtype
-):
-    assert (
-        fill_value is not None
-    ), "Trying to convert an exception to nodata, but `fill_value` is None"
-
+def nodata_for_window(window: Window, fill_value: Union[int, float], dtype: np.dtype):
     height = cast(int, window.height)
     width = cast(int, window.width)
     # Argument of type "tuple[_T@attrib, _T@attrib]" cannot be assigned to parameter "shape" of type "_ShapeLike"

stackstac/reader_protocol.py

@@ -26,14 +26,17 @@ class Reader(Pickleable, Protocol):
     Protocol for a thread-safe, lazily-loaded object for reading data from a single-band STAC asset.
     """
 
+    fill_value: Union[int, float]
+    dtype: np.dtype
+
     def __init__(
         self,
         *,
         url: str,
         spec: RasterSpec,
         resampling: Resampling,
         dtype: np.dtype,
-        fill_value: Optional[Union[int, float]],
+        fill_value: Union[int, float],
         rescale: bool,
         gdal_env: Optional[LayeredEnv],
         errors_as_nodata: Tuple[Exception, ...] = (),
@@ -113,13 +116,15 @@ class FakeReader:
     or inherent to the dask graph.
     """
 
-    def __init__(self, *, url: str, spec: RasterSpec, **kwargs) -> None:
+    def __init__(
+        self, *, dtype: np.dtype, fill_value: Union[int, float], **kwargs
+    ) -> None:
         pass
-        # self.url = url
-        # self.spec = spec
+        self.dtype = dtype
+        self.fill_value = fill_value
 
     def read(self, window: Window, **kwargs) -> np.ndarray:
-        return np.random.random((window.height, window.width))
+        return np.random.random((window.height, window.width)).astype(self.dtype)
 
     def close(self) -> None:
         pass

stackstac/rio_reader.py

@@ -279,7 +279,7 @@ class PickleState(TypedDict):
     spec: RasterSpec
     resampling: Resampling
     dtype: np.dtype
-    fill_value: Optional[Union[int, float]]
+    fill_value: Union[int, float]
     rescale: bool
     gdal_env: Optional[LayeredEnv]
     errors_as_nodata: Tuple[Exception, ...]
@@ -302,7 +302,7 @@ def __init__(
         spec: RasterSpec,
         resampling: Resampling,
         dtype: np.dtype,
-        fill_value: Optional[Union[int, float]],
+        fill_value: Union[int, float],
         rescale: bool,
         gdal_env: Optional[LayeredEnv] = None,
         errors_as_nodata: Tuple[Exception, ...] = (),
@@ -407,8 +407,6 @@ def read(self, window: Window, **kwargs) -> np.ndarray:
 
         result = result.astype(self.dtype, copy=False)
         result = np.ma.filled(result, fill_value=self.fill_value)
-        # ^ NOTE: if `self.fill_value` was None, rasterio set the masked array's fill value to the
-        # nodata value of the band, which `np.ma.filled` will then use.
         return result
 
     def close(self) -> None:

stackstac/stack.py

@@ -30,7 +30,7 @@ def stack(
     resampling: Resampling = Resampling.nearest,
     chunksize: int = 1024,
     dtype: np.dtype = np.dtype("float64"),
-    fill_value: Optional[Union[int, float]] = np.nan,
+    fill_value: Union[int, float] = np.nan,
     rescale: bool = True,
     sortby_date: Literal["asc", "desc", False] = "asc",
     xy_coords: Literal["center", "topleft", False] = "topleft",
@@ -192,9 +192,7 @@ def stack(
         don't set it here---instead, call ``.chunk`` on the DataArray to re-chunk it.
     dtype:
         The NumPy data type of the output array. Default: ``float64``. Must be a data type
-        that's compatible with ``fill_value``. Note that if ``fill_value`` is None, whatever nodata
-        value is set in each asset's file will be used, so that value needs to be compatible
-        with ``dtype`` as well.
+        that's compatible with ``fill_value``.
     fill_value:
         Value to fill nodata/masked pixels with. Default: ``np.nan``.
 
@@ -249,7 +247,7 @@ def stack(
     errors_as_nodata:
         Exception patterns to ignore when opening datasets or reading data.
         Exceptions matching the pattern will be logged as warnings, and just
-        produce nodata (``fill_value``). A non-None ``fill_value`` is required when using this.
+        produce nodata (``fill_value``).
 
         The exception patterns should be instances of an Exception type to catch,
         where ``str(exception_pattern)`` is a regex pattern to match against
@@ -282,9 +280,9 @@ def stack(
     if sortby_date is not False:
         plain_items = sorted(
             plain_items,
-            key=lambda item: item["properties"].get("datetime", ""),
+            key=lambda item: item["properties"].get("datetime", "") or "",
             reverse=sortby_date == "desc",
-        )  # type: ignore
+        )
 
     asset_table, spec, asset_ids, plain_items = prepare_items(
         plain_items,

stackstac/to_dask.py

@@ -24,18 +24,12 @@ def items_to_dask(
     chunksize: int,
     resampling: Resampling = Resampling.nearest,
     dtype: np.dtype = np.dtype("float64"),
-    fill_value: Optional[Union[int, float]] = np.nan,
+    fill_value: Union[int, float] = np.nan,
     rescale: bool = True,
     reader: Type[Reader] = AutoParallelRioReader,
     gdal_env: Optional[LayeredEnv] = None,
     errors_as_nodata: Tuple[Exception, ...] = (),
 ) -> da.Array:
-    if fill_value is None and errors_as_nodata:
-        raise ValueError(
-            "A non-None `fill_value` is required when using `errors_as_nodata`. "
-            "If an exception occurs, we need to know what to use as the nodata value, "
-            "since there may not be an open dataset to infer it from."
-        )
     errors_as_nodata = errors_as_nodata or ()  # be sure it's not None
 
     if fill_value is not None and not np.can_cast(fill_value, dtype):
@@ -114,17 +108,18 @@ def asset_entry_to_reader_and_window(
     spec: RasterSpec,
     resampling: Resampling,
     dtype: np.dtype,
-    fill_value: Optional[Union[int, float]],
+    fill_value: Union[int, float],
     rescale: bool,
     gdal_env: Optional[LayeredEnv],
     errors_as_nodata: Tuple[Exception, ...],
     reader: Type[ReaderT],
-) -> Optional[Tuple[ReaderT, windows.Window]]:
+) -> Tuple[ReaderT, windows.Window] | np.ndarray:
     asset_entry = asset_entry[0, 0]
     # ^ because dask adds extra outer dims in `from_array`
     url = asset_entry["url"]
     if url is None:
-        return None
+        # Signifies empty value
+        return np.array(fill_value, dtype)
 
     asset_bounds: Bbox = asset_entry["bounds"]
     asset_window = windows.from_bounds(*asset_bounds, transform=spec.transform)
@@ -159,11 +154,11 @@ def asset_entry_to_reader_and_window(
 
 
 def fetch_raster_window(
-    asset_entry: Optional[Tuple[Reader, windows.Window]],
+    asset_entry: Tuple[ReaderT, windows.Window] | np.ndarray,
     slices: Tuple[slice, ...],
 ) -> np.ndarray:
     current_window = windows.Window.from_slices(*slices)
-    if asset_entry is not None:
+    if isinstance(asset_entry, tuple):
         reader, asset_window = asset_entry
 
         # check that the window we're fetching overlaps with the asset
@@ -172,7 +167,11 @@ def fetch_raster_window(
             data = reader.read(current_window)
 
             return data[None, None]
+        fill_arr = np.array(reader.fill_value, reader.dtype)
+    else:
+        fill_arr: np.ndarray = asset_entry
 
     # no dataset, or we didn't overlap it: return empty data.
     # use the broadcast trick for even fewer memz
-    return np.broadcast_to(np.nan, (1, 1) + windows.shape(current_window))
+    return np.broadcast_to(fill_arr, (1, 1) + windows.shape(current_window))
+