Use small negative sentinels for int

Author: tomwhite

sgkit/io/utils.py

@@ -8,12 +8,7 @@
 from ..typing import ArrayLike, DType
 from ..utils import encode_array, max_str_len
 
-# These values are based on BCF conventions
-INT32_MISSING, INT32_FILL = np.array(
-    # Equivalent to 0x80000000, 0x80000001
-    [np.iinfo(np.int32).min, np.iinfo(np.int32).min + 1],
-    dtype=np.int32,
-)
+INT32_MISSING, INT32_FILL = -1, -2
 
 FLOAT32_MISSING, FLOAT32_FILL = np.array([0x7F800001, 0x7F800002], dtype=np.int32).view(
     np.float32

sgkit/io/vcf/vcf_reader.py

@@ -25,7 +25,7 @@
 
 from sgkit import variables
 from sgkit.io.dataset import load_dataset
-from sgkit.io.utils import FLOAT32_FILL, INT32_FILL, STR_FILL
+from sgkit.io.utils import FLOAT32_FILL, INT32_FILL, INT32_MISSING, STR_FILL
 from sgkit.io.vcf import partition_into_regions
 from sgkit.io.vcf.utils import build_url, chunks, temporary_directory, url_filename
 from sgkit.io.vcfzarr_reader import (
@@ -262,6 +262,11 @@ def truncate_array(self, length: int) -> None:
         self.array = self.array[:length]
 
     def update_dataset(self, ds: xr.Dataset) -> None:
+        # cyvcf2 represents missing Integer values as the minimum int32 value
+        # so change these to be the missing value
+        if self.array.dtype == np.int32:
+            self.array[self.array == np.iinfo(np.int32).min] = INT32_MISSING
+
         ds[self.variable_name] = (self.dims, self.array)
         if len(self.description) > 0:
             ds[self.variable_name].attrs["comment"] = self.description