From 18e38436c0f357f592ba41ebb27726cf8b77334f Mon Sep 17 00:00:00 2001 From: Alessandro Amici Date: Sun, 15 Nov 2026 19:27:02 +0100 Subject: [PATCH] Use explicit int32 for all the integer variables and coordinates The line, pixel, degree and count values and the noise azimuth first and last line and sample used the platform default integer, int64 on 64-bit and int32 on 32-bit architectures. Set them to int32 everywhere, as the CPM 3.0 mapping does for line, pixel and the noise azimuth values; it leaves degree and count undeclared. Drop the 32-bit xfails added in #203, the dtypes no longer depend on the architecture. Co-Authored-By: Claude Opus 5.5 --- tests/data/eopf/S01SIWSLC_target.json | 30 +++++++++++++-------------- tests/test_15_eopf_sentinel1.py | 7 +------ tests/test_30_eopf_product.py | 4 ---- xarray_sentinel/eopf_sentinel1.py | 27 +++++++++++++----------- xarray_sentinel/sentinel1.py | 22 ++++++++++---------- 5 files changed, 42 insertions(+), 48 deletions(-) diff --git a/tests/data/eopf/S01SIWSLC_target.json b/tests/data/eopf/S01SIWSLC_target.json index 7696522..23b6a8b 100644 --- a/tests/data/eopf/S01SIWSLC_target.json +++ b/tests/data/eopf/S01SIWSLC_target.json @@ -3,7 +3,7 @@ "conditions": {}, "conditions/antenna_pattern": { "azimuth_time": [["azimuth_time"], "datetime64[ns]"], - "count": [["count"], "int64"], + "count": [["count"], "int32"], "elevation_angle": [["polarization", "azimuth_time", "count"], "float32"], "incidence_angle": [["polarization", "azimuth_time", "count"], "float32"], "polarization": [["polarization"], " None: assert res.data_dc_rms_error_above_threshold.dtype == bool -@pytest.mark.xfail(IS_32_BIT, reason="int32 instead of int64 on 32bit architectures.") def test_open_coordinate_conversion() -> None: # dims and dtypes from the CPM 3.0 S01SIWGRD mapping expected = { "azimuth_time": [["azimuth_time"], "datetime64[ns]"], - "degree": [["degree"], "int64"], + "degree": [["degree"], "int32"], "gr0": [["azimuth_time"], "float64"], "grsr_coefficients": [["azimuth_time", "degree"], "float64"], "slant_range_time": [["azimuth_time"], "float64"], diff --git a/tests/test_30_eopf_product.py b/tests/test_30_eopf_product.py index 3408a0f..a2d1dab 100644 --- a/tests/test_30_eopf_product.py +++ b/tests/test_30_eopf_product.py @@ -1,7 +1,6 @@ import json import pathlib import shutil -import sys import warnings from typing import Any @@ -11,8 +10,6 @@ from xarray_sentinel import eopf_product, sentinel1 -IS_32_BIT = sys.maxsize == 0x7FFFFFFF - DATA_FOLDER = pathlib.Path(__file__).parent / "data" SLC_IW = ( @@ -315,7 +312,6 @@ def test_open_datatree_groups( assert slc_iw_node_structure.keys() == EXPECTED_SLC_IW_STRUCTURE.keys() -@pytest.mark.xfail(IS_32_BIT, reason="int32 instead of int64 on 32bit architectures.") @pytest.mark.parametrize("group", sorted(EXPECTED_SLC_IW_STRUCTURE)) def test_open_datatree_group_structure( slc_iw_node_structure: dict[str, dict[str, list[Any]]], group: str diff --git a/xarray_sentinel/eopf_sentinel1.py b/xarray_sentinel/eopf_sentinel1.py index 773d021..24001ac 100644 --- a/xarray_sentinel/eopf_sentinel1.py +++ b/xarray_sentinel/eopf_sentinel1.py @@ -60,7 +60,7 @@ def open_antenna_pattern(annotation: dict[str, Any]) -> xr.Dataset: "roll": pattern["roll"], "terrain_height": pattern["terrainHeight"], } - coords = {"count": np.arange(slant_range_time.size)} + coords = {"count": np.arange(slant_range_time.size, dtype=np.int32)} pattern_dataset = xr.Dataset(data_vars=data_vars, coords=coords) azimuth_time = [np.datetime64(pattern["azimuthTime"], "ns")] pattern_datasets.append(pattern_dataset.expand_dims(azimuth_time=azimuth_time)) @@ -132,7 +132,7 @@ def open_doppler_centroid(annotation: dict[str, Any]) -> xr.Dataset: coords = { "azimuth_time": [np.datetime64(e["azimuthTime"], "ns") for e in estimates], # the polynomial powers, not reversed as in EOPF, see quirk 2 in docs/DATATREE.md - "degree": np.arange(data_dc_polynomial[0].size), + "degree": np.arange(data_dc_polynomial[0].size, dtype=np.int32), } return xr.Dataset(data_vars=data_vars, coords=coords) @@ -155,8 +155,8 @@ def open_gcp(annotation: dict[str, Any]) -> xr.Dataset: # the annotated times of the first column, see quirk 1 in docs/DATATREE.md "azimuth_time": grids["azimuth_time_gcp"][:, 0], range_name: range_values, - "line": ("azimuth_time", lines), - "pixel": (range_name, pixels), + "line": ("azimuth_time", np.array(lines, dtype=np.int32)), + "pixel": (range_name, np.array(pixels, dtype=np.int32)), } dims = ("azimuth_time", range_name) data_vars = {name: (dims, grid) for name, grid in grids.items()} @@ -190,7 +190,7 @@ def open_coordinate_conversion(annotation: dict[str, Any]) -> xr.Dataset | None: } coords = { "azimuth_time": [np.datetime64(c["azimuthTime"], "ns") for c in conversions], - "degree": np.arange(srgr_coefficients[0].size), + "degree": np.arange(srgr_coefficients[0].size, dtype=np.int32), } return xr.Dataset(data_vars=data_vars, coords=coords) @@ -235,7 +235,7 @@ def open_range_vectors( The GRD vectors change their pixels at the sub-swath boundaries, which move in azimuth, so the shorter vectors are padded with NaN and pixel -1. """ - pixels = [np.fromstring(v["pixel"]["$"], dtype=int, sep=" ") for v in vectors] + pixels = [np.fromstring(v["pixel"]["$"], dtype=np.int32, sep=" ") for v in vectors] pixel = stack_vectors(pixels, -1) range_name, range_values = range_coordinate(pixel[0], annotation) dims = ("azimuth_time", range_name) @@ -249,7 +249,10 @@ def open_range_vectors( raise ValueError("Unable to organise vectors in a regular line-pixel grid") coords[range_name] = range_values pixel_coordinate = (range_name, pixel[0]) - coords["line"] = ("azimuth_time", [v["line"] for v in vectors]) + coords["line"] = ( + "azimuth_time", + np.array([v["line"] for v in vectors], dtype=np.int32), + ) coords["pixel"] = pixel_coordinate data_vars: dict[str, Any] = {} for name, tag in variables.items(): @@ -302,16 +305,16 @@ def open_noise_azimuth( vectors = vector_list["noiseAzimuthVector"] vector_datasets = [] for vector in vectors: - line = np.fromstring(vector["line"]["$"], dtype=int, sep=" ") + line = np.fromstring(vector["line"]["$"], dtype=np.int32, sep=" ") noise_azimuth_lut = np.fromstring( vector["noiseAzimuthLut"]["$"], dtype=np.float32, sep=" " ) data_vars = { "noise_azimuth_lut": ("line", noise_azimuth_lut), - "first_azimuth_line": vector["firstAzimuthLine"], - "last_azimuth_line": vector["lastAzimuthLine"], - "first_range_sample": vector["firstRangeSample"], - "last_range_sample": vector["lastRangeSample"], + "first_azimuth_line": np.int32(vector["firstAzimuthLine"]), + "last_azimuth_line": np.int32(vector["lastAzimuthLine"]), + "first_range_sample": np.int32(vector["firstRangeSample"]), + "last_range_sample": np.int32(vector["lastRangeSample"]), } vector_dataset = xr.Dataset(data_vars=data_vars, coords={"line": line}) vector_datasets.append(vector_dataset.expand_dims(swath=[vector["swath"]])) diff --git a/xarray_sentinel/sentinel1.py b/xarray_sentinel/sentinel1.py index 9815656..977ce78 100644 --- a/xarray_sentinel/sentinel1.py +++ b/xarray_sentinel/sentinel1.py @@ -92,7 +92,7 @@ def open_calibration_dataset( for vector in calibration_vectors: azimuth_time_list.append(vector["azimuthTime"]) line_list.append(vector["line"]) - pixel = np.fromstring(vector["pixel"]["$"], dtype=int, sep=" ") + pixel = np.fromstring(vector["pixel"]["$"], dtype=np.int32, sep=" ") pixel_list.append(pixel) sigmaNought = np.fromstring( vector["sigmaNought"]["$"], dtype=np.float32, sep=" " @@ -117,7 +117,7 @@ def open_calibration_dataset( "gamma": (("line", "pixel"), gamma_list), "dn": (("line", "pixel"), dn_list), } - coords = {"line": line_list, "pixel": pixel_list[0]} + coords = {"line": np.array(line_list, dtype=np.int32), "pixel": pixel_list[0]} return xr.Dataset(data_vars=data_vars, coords=coords, attrs=attrs) @@ -248,7 +248,7 @@ def open_noise_range_dataset( for vector in noise_vectors: azimuth_time_list.append(vector["azimuthTime"]) line_list.append(vector["line"]) - pixel = np.fromstring(vector["pixel"]["$"], dtype=int, sep=" ") + pixel = np.fromstring(vector["pixel"]["$"], dtype=np.int32, sep=" ") pixel_list.append(pixel) noiseRangeLut = np.fromstring( vector["noiseRangeLut"]["$"], dtype=np.float32, sep=" " @@ -264,7 +264,7 @@ def open_noise_range_dataset( "azimuth_time": ("line", [np.datetime64(dt, "ns") for dt in azimuth_time_list]), "noiseRangeLut": (("line", "pixel"), noiseRangeLut_list), } - coords = {"line": line_list, "pixel": pixel_list[0]} + coords = {"line": np.array(line_list, dtype=np.int32), "pixel": pixel_list[0]} return xr.Dataset(data_vars=data_vars, coords=coords, attrs=attrs) @@ -278,16 +278,16 @@ def open_noise_azimuth_dataset( vector_datasets = [] for vector in noise_vectors: - line = np.fromstring(vector["line"]["$"], dtype=int, sep=" ") + line = np.fromstring(vector["line"]["$"], dtype=np.int32, sep=" ") noise_azimuth_lut = np.fromstring( vector["noiseAzimuthLut"]["$"], dtype=np.float32, sep=" " ) data_vars = { "noiseAzimuthLut": ("line", noise_azimuth_lut), - "firstAzimuthLine": vector["firstAzimuthLine"], - "lastAzimuthLine": vector["lastAzimuthLine"], - "firstRangeSample": vector["firstRangeSample"], - "lastRangeSample": vector["lastRangeSample"], + "firstAzimuthLine": np.int32(vector["firstAzimuthLine"]), + "lastAzimuthLine": np.int32(vector["lastAzimuthLine"]), + "firstRangeSample": np.int32(vector["firstRangeSample"]), + "lastRangeSample": np.int32(vector["lastRangeSample"]), } vector_ds = xr.Dataset(data_vars=data_vars, coords={"line": line}) vector_datasets.append(vector_ds.expand_dims(swath=[vector["swath"]])) @@ -662,8 +662,8 @@ def open_pol_dataset( ] = azimuth_time_burst coords = { - "pixel": np.arange(0, number_of_samples, dtype=int), - "line": np.arange(0, number_of_lines, dtype=int), + "pixel": np.arange(0, number_of_samples, dtype=np.int32), + "line": np.arange(0, number_of_lines, dtype=np.int32), # We don't need top set the "units" as Xarray now detects the best resolution. # See: https://github.com/pydata/xarray/issues/4183#issuecomment-685200043 # and https://github.com/pydata/xarray/pull/4684