Remove VLenUTF8 from filters to avoid double encoding error pydata/xa…

…rray#3476

sgkit/io/dataset.py

@@ -2,6 +2,7 @@
 from typing import Any, Dict, MutableMapping, Optional, Union
 
 import fsspec
+import numcodecs
 import xarray as xr
 from xarray import Dataset
 
@@ -38,6 +39,15 @@ def save_dataset(
     for v in ds:
         # Workaround for https://github.com/pydata/xarray/issues/4380
         ds[v].encoding.pop("chunks", None)
+
+        # Remove VLenUTF8 from filters to avoid double encoding error https://github.com/pydata/xarray/issues/3476
+        filters = ds[v].encoding.get("filters", None)
+        var_len_str_codec = numcodecs.VLenUTF8()
+        if filters is not None and var_len_str_codec in filters:
+            filters = list(filters)
+            filters.remove(var_len_str_codec)
+            ds[v].encoding["filters"] = filters
+
     ds.to_zarr(store, **kwargs)
 
 

sgkit/tests/io/vcf/test_vcf_reader.py

@@ -7,14 +7,15 @@
 from numcodecs import Blosc, PackBits, VLenUTF8
 from numpy.testing import assert_allclose, assert_array_equal
 
-from sgkit import load_dataset
+from sgkit import load_dataset, save_dataset
 from sgkit.io.utils import FLOAT32_FILL, INT_FILL, INT_MISSING
 from sgkit.io.vcf import (
     MaxAltAllelesExceededWarning,
     partition_into_regions,
     vcf_to_zarr,
 )
 from sgkit.io.vcf.vcf_reader import zarr_array_sizes
+from sgkit.tests.io.test_dataset import assert_identical
 
 from .utils import path_for_test
 
@@ -95,6 +96,14 @@ def test_vcf_to_zarr__small_vcf(shared_datadir, is_path, tmp_path):
     assert_array_equal(ds["call_genotype_mask"], call_genotype < 0)
     assert_array_equal(ds["call_genotype_phased"], call_genotype_phased)
 
+    # save and load again to test https://github.com/pydata/xarray/issues/3476
+    with pytest.warns(xr.coding.variables.SerializationWarning):
+        path2 = tmp_path / "ds2.zarr"
+        if not is_path:
+            path2 = str(path2)
+        save_dataset(ds, path2)
+        assert_identical(ds, load_dataset(path2))
+
 
 @pytest.mark.parametrize(
     "is_path",