@@ -101,44 +101,69 @@ def test_float_measurements_have_fill_value(snapshot: dict) -> None:
101101
102102
103103def test_fill_value_masking_roundtrip (tmp_path : pathlib .Path ) -> None :
104- """End-to-end: a float array written with our ``_FillValue`` convention
105- must round-trip through xarray's ``use_zarr_fill_value_as_mask=True``.
104+ """End-to-end converter check: float arrays produced by
105+ ``create_geozarr_dataset`` must round-trip through xarray's
106+ ``use_zarr_fill_value_as_mask=True`` so NaN cells come back masked.
106107
107- Mirrors the rio-tiler reader expectation: NaN-filled cells become masked
108- when re-opened, all other cells remain unmasked.
108+ Mirrors the rio-tiler reader expectation. Builds a minimal float
109+ DataTree, runs the real converter, then reopens the output and
110+ asserts masking semantics on a measurement band.
109111 """
110112 import numpy as np
111113 import xarray as xr
112114
113- # Build a small float dataset following the converter's convention:
114- # - attrs["_FillValue"] = np.nan (xarray encodes via FillValueCoder)
115- # - encoding["fill_value"] = "NaN"
116- data = np .arange (50 * 50 , dtype = "float32" ).reshape (50 , 50 )
117- data [0 :5 , 0 :5 ] = np .nan # nodata patch
118- da = xr .DataArray (data , dims = ("y" , "x" ), name = "band" )
119- da .attrs ["_FillValue" ] = np .nan
120-
121- ds = da .to_dataset ()
122- store = tmp_path / "roundtrip.zarr"
123- ds .to_zarr (
124- store ,
125- zarr_format = 3 ,
126- encoding = {"band" : {"fill_value" : "NaN" , "chunks" : (25 , 25 )}},
127- consolidated = True ,
115+ from eopf_geozarr .conversion import create_geozarr_dataset
116+
117+ epsg_code = 32632
118+ x_min , x_max = 600000 , 605120 # 5.12 km
119+ y_min , y_max = 5090000 , 5095120
120+ nx = ny = 512
121+
122+ # Float reflectance band with a nodata patch (NaN) — what the converter
123+ # would normally see after CF decoding of scale_factor/add_offset.
124+ data = np .random .default_rng (42 ).uniform (0.0 , 1.0 , size = (ny , nx )).astype ("float32" )
125+ data [0 :32 , 0 :32 ] = np .nan
126+
127+ ds = xr .Dataset (
128+ {"b04" : (["y" , "x" ], data , {"long_name" : "Red band (B04)" })},
129+ coords = {
130+ "x" : np .linspace (x_min , x_max , nx , endpoint = False ),
131+ "y" : np .linspace (y_max , y_min , ny , endpoint = False ),
132+ },
133+ ).rio .write_crs (f"EPSG:{ epsg_code } " )
134+
135+ dt = xr .DataTree ()
136+ dt ["measurements" ] = xr .DataTree ()
137+ dt ["measurements/reflectance" ] = xr .DataTree ()
138+ dt ["measurements/reflectance/r10m" ] = ds
139+
140+ output_path = tmp_path / "fill_value_roundtrip.zarr"
141+ create_geozarr_dataset (
142+ dt_input = dt ,
143+ groups = ["/measurements/reflectance/r10m" ],
144+ output_path = str (output_path ),
145+ spatial_chunk = 256 ,
146+ min_dimension = 128 ,
147+ max_retries = 1 ,
128148 )
129149
150+ band_group = output_path / "measurements" / "reflectance" / "r10m"
130151 reopened = xr .open_dataset (
131- store ,
152+ band_group ,
132153 engine = "zarr" ,
133154 zarr_format = 3 ,
134155 consolidated = True ,
135156 decode_times = False ,
136157 decode_coords = False ,
137158 use_zarr_fill_value_as_mask = True ,
138159 )
139- masked = reopened ["band" ].to_masked_array ()
140-
141- assert np .ma .is_masked (masked ), "NaN cells should be masked when opened with mask=True"
142- assert masked .mask [0 , 0 ], "nodata corner cell must be masked"
143- assert not masked .mask [- 1 , - 1 ], "valid cell must not be masked"
144- reopened .close ()
160+ try :
161+ masked = reopened ["b04" ].to_masked_array ()
162+ assert np .ma .is_masked (masked ), (
163+ "NaN cells in converter output should be masked when opened with "
164+ "use_zarr_fill_value_as_mask=True"
165+ )
166+ assert masked .mask [0 , 0 ], "nodata corner cell must be masked"
167+ assert not masked .mask [- 1 , - 1 ], "valid cell must not be masked"
168+ finally :
169+ reopened .close ()
0 commit comments