add test_to_rio_dataset_nodata_none() and test mask cacheing in test_…

…nodata_mask()

tests/test_fuse_api.py

@@ -124,7 +124,7 @@ def test_basic_fusion(
             )
         ),
         dict(
-            driver='GTiff', dtype='uint8', nodata=0,
+            driver='GTiff', dtype='uint8', nodata=None,
             creation_options=dict(
                 tiled=True, blockxsize=64, blockysize=64, compress='jpeg', interleave='pixel', photometric='ycbcr'
             )
@@ -139,8 +139,10 @@ def test_out_profile(file_rgb_100cm_float, tmp_path: Path, out_profile: Dict):
     with raster_fuse:
         raster_fuse.process(corr_filename, Model.gain_blk_offset, (3, 3), out_profile=out_profile)
     assert (corr_filename.exists())
+
     out_profile.update(**out_profile['creation_options'])
     out_profile.pop('creation_options')
+
     with rio.open(file_rgb_100cm_float, 'r') as src_ds, rio.open(corr_filename, 'r') as fuse_ds:
         # test output image has been set with out_profile properties
         for k, v in out_profile.items():

tests/test_raster_array.py

@@ -24,7 +24,7 @@
 import pytest
 import rasterio as rio
 from rasterio.crs import CRS
-from rasterio.enums import Resampling
+from rasterio.enums import Resampling, MaskFlags
 from rasterio.transform import Affine, from_bounds
 from rasterio.windows import Window
 from rasterio.warp import transform_bounds
@@ -72,12 +72,19 @@ def test_nodata_mask(ra_byte):
     assert (ra_byte.mask_ra.array == mask).all()
     assert ra_byte.mask_ra.transform == ra_byte.transform
 
+    # test mask changed after setting altered masked array
+    array = ra_byte.array.copy()
+    array[np.divide(mask.shape, 2).astype('int')] = ra_byte.nodata
+    mask[np.divide(mask.shape, 2).astype('int')] = False
+    ra_byte.array = array
+    assert (ra_byte.mask == mask).all()
+
     # test mask unchanged after setting stacked array
     ra_byte.array = np.stack((ra_byte.array, ra_byte.array), axis=0)
     assert (ra_byte.mask == mask).all()
 
     # test altering the mask
-    mask[np.divide(mask.shape, 2).astype('int')] = False
+    mask[(np.divide(mask.shape, 2) + 1).astype('int')] = False
     ra_byte.mask = mask
     assert (ra_byte.mask == mask).all()
 
@@ -192,6 +199,23 @@ def test_to_rio_dataset(ra_byte, tmp_path: Path):
     assert (test_array == ra_byte.array).all()
 
 
+def test_to_rio_dataset_nodata_none(ra_byte, tmp_path: Path):
+    """ Test writing raster array to dataset with nodata=None writes an internal mask. """
+    ds_filename = tmp_path.joinpath('temp.tif')
+    profile = ra_byte.profile
+    profile.update(nodata=None)
+    with rio.open(ds_filename, 'w', driver='GTiff', **profile) as ds:
+        ra_byte.to_rio_dataset(ds)
+
+    with rio.open(ds_filename, 'r') as ds:
+        assert ds.nodata is None
+        assert ds.mask_flag_enums[0] == [MaskFlags.per_dataset]
+        test_mask = ds.dataset_mask().astype('bool')
+        test_array = ds.read(indexes=1)
+    assert (test_mask == ra_byte.mask).all()
+    assert (test_array[test_mask] == ra_byte.array[ra_byte.mask]).all()
+
+
 def test_to_rio_dataset_crop(ra_rgb_byte, tmp_path: Path):
     """ Test writing a raster array to a dataset where the dataset & raster array sizes differ. """
     ds_filename = tmp_path.joinpath('temp.tif')
@@ -290,7 +314,7 @@ def test_reprojection(ra_rgb_byte):
     ('int64', 1, 'float64', float('nan')),
     ('float32', float('nan'), 'float32', None),  # nodata unchanged
 ])
-def test_convert_dtype(profile_100cm_float: dict, src_dtype: str, src_nodata: float, dst_dtype: str, dst_nodata: float):
+def test_convert_array_dtype(profile_100cm_float: dict, src_dtype: str, src_nodata: float, dst_dtype: str, dst_nodata: float):
     """ Test dtype conversion with combinations covering rounding, clipping (with and w/o type promotion) and
     re-masking.
     """
@@ -307,32 +331,36 @@ def test_convert_dtype(profile_100cm_float: dict, src_dtype: str, src_nodata: fl
         array, crs=profile_100cm_float['crs'], transform=profile_100cm_float['transform'], nodata=src_nodata
     )
 
-    # convert src_ra to dtype
-    test_ra = src_ra.copy()
-    test_ra.convert_dtype(dst_dtype, nodata=dst_nodata)
+    # convert to dtype
+    src_copy_ra = src_ra.copy()
+    test_array = src_copy_ra._convert_array_dtype(dst_dtype, nodata=dst_nodata)
+
+    # test converting did not change src_copy_ra
+    assert utils.nan_equals(src_copy_ra.array, src_ra.array).all()
+    assert (src_copy_ra.mask == src_ra.mask).all()
 
     # create rounded & clipped array in src_dtype to test against
-    test_array = array
+    ref_array = array
     if np.issubdtype(dst_dtype, np.integer):
-        test_array = np.clip(np.round(test_array), dst_dtype_info.min, dst_dtype_info.max)
+        ref_array = np.clip(np.round(ref_array), dst_dtype_info.min, dst_dtype_info.max)
     elif np.issubdtype(src_dtype, np.floating):
         # don't clip float but set out of range vals to +-inf (as np.astype does)
-        test_array[test_array < dst_dtype_info.min] = float('-inf')
-        test_array[test_array > dst_dtype_info.max] = float('inf')
-        assert np.any(test_array[src_ra.mask] % 1 != 0)  # check contains decimals
+        ref_array[ref_array < dst_dtype_info.min] = float('-inf')
+        ref_array[ref_array > dst_dtype_info.max] = float('inf')
+        assert np.any(ref_array[src_ra.mask] % 1 != 0)  # check contains decimals
 
-    assert test_ra.dtype == dst_dtype
+    assert test_array.dtype == dst_dtype
     if dst_nodata:
-        assert utils.nan_equals(test_ra.nodata, dst_nodata)
-    assert np.any(test_ra.mask)
-    assert np.all(test_ra.mask == src_ra.mask)
+        test_mask = ~utils.nan_equals(test_array, dst_nodata)
+        assert np.any(test_mask)
+        assert (test_mask == src_ra.mask).all()
     # use approx test for case of (expected) precision loss e.g. float64->float32 or int64->float32
-    assert test_ra.array[test_ra.mask] == pytest.approx(test_array[src_ra.mask], rel=1e-6)
+    assert test_array[src_ra.mask] == pytest.approx(ref_array[src_ra.mask], rel=1e-6)
 
 
-def test_convert_dtype_error(ra_100cm_float: RasterArray):
+def test_convert_array_dtype_error(ra_100cm_float: RasterArray):
     """ Test dtype conversion raises an error when the nodata value cannot be cast to the conversion dtype. """
     test_ra = ra_100cm_float.copy()
     with pytest.raises(ValueError) as ex:
-        test_ra.convert_dtype('uint8', nodata=float('nan'))
+        test_ra._convert_array_dtype('uint8', nodata=float('nan'))
     assert 'cast' in str(ex.value)