Add support for NumPy v2.0

pds4_tools/reader/data.py

@@ -8,7 +8,7 @@
 
 from .data_types import pds_to_numpy_name
 
-from ..utils.compat import OrderedDict
+from ..utils.compat import OrderedDict, np_issubclass
 from ..extern import six
 
 
@@ -537,7 +537,7 @@ def view(self, dtype=None, type=None, fill_value=None):
             obj = super(PDS_marray, self).view(dtype=dtype, type=type, fill_value=fill_value)
 
             # Fix bug in NumPy < v1.10, which resets fill value on ``view`` if mask is not nomask
-            if ((dtype is None) or ((type is None) and np.issubclass_(dtype, np.ma.MaskedArray))) and \
+            if ((dtype is None) or ((type is None) and np_issubclass(dtype, np.ma.MaskedArray))) and \
                 (fill_value is None):
 
                 obj._fill_value = self._fill_value

pds4_tools/reader/data_types.py

@@ -7,6 +7,7 @@
 
 import numpy as np
 
+from ..utils.compat import np_unicode
 from ..utils.deprecation import rename_parameter
 from ..utils.logging import logger_init
 
@@ -135,7 +136,7 @@ def pds_to_numpy_type(data_type=None, data=None, field_length=None, decode_strin
         # Get dtype for character data (from data)
         if is_character_data:
             unicode_requested = decode_strings and not is_bitstring_data
-            dtype = 'U' if (np.issubdtype(data.dtype, np.unicode_) or unicode_requested) else 'S'
+            dtype = 'U' if (np.issubdtype(data.dtype, np_unicode) or unicode_requested) else 'S'
 
             if field_length is not None:
                 dtype += str(field_length)
@@ -248,7 +249,7 @@ def pds_to_builtin_type(data_type=None, data=None, decode_strings=False, decode_
 
         if is_character_data:
             unicode_requested = decode_strings and not is_bitstring_data
-            _type = six.text_type if (np.issubdtype(data.dtype, np.unicode_) or unicode_requested
+            _type = six.text_type if (np.issubdtype(data.dtype, np_unicode) or unicode_requested
                                       ) else six.binary_type
         else:
             _type = type(np.asscalar(data[0]))
@@ -311,7 +312,7 @@ def numpy_to_pds_type(dtype, ascii_numerics=False):
     """
 
     # For string dtypes
-    if np.issubdtype(dtype, np.unicode_):
+    if np.issubdtype(dtype, np_unicode):
         data_type = 'UTF8_String'
 
     elif np.issubdtype(dtype, np.string_):
@@ -460,6 +461,8 @@ def data_type_convert_table_ascii(data_type, data, mask_nulls=False, decode_stri
         data[mask_array] = six.ensure_binary(str(fill_value))
 
     # Special handling for boolean due to e.g. bool('false') = True
+    # and that in NumPy 2.0+ string arrays cast to bool set
+    # all non-empty strings as True.
     if data_type == 'ASCII_Boolean':
 
         # Replace 'true' and 'false' with 1 and 0
@@ -469,9 +472,10 @@ def data_type_convert_table_ascii(data_type, data, mask_nulls=False, decode_stri
         data = data.split(b'@')
 
         try:
-            data = np.asarray(data).astype(dtype, copy=False)
+            data = np.asarray(data).astype(np.uint8, copy=False) \
+                                   .astype(dtype, copy=False)
         except TypeError:
-            data = np.asarray(data).astype(dtype)
+            data = np.asarray(data).astype(np.uint8).astype(dtype)
 
     # Convert ASCII numerics into their proper data type
     elif not np.issubdtype(dtype, np.character):
@@ -770,11 +774,23 @@ def apply_scaling_and_value_offset(data, scaling_factor=None, value_offset=None,
     data = adjust_array_data_type(data, scaling_factor, value_offset)
 
     # Apply scaling factor and value offset
+    # (workaround inability of NumPy 2.0+ to add / multiply in cases where
+    #  value on right of operand is out-of-bounds of dtype, see PR #99)
     if not no_scaling:
-        data *= scaling_factor
+
+        scale_dtype = None
+        if isinstance(scaling_factor, six.integer_types):
+            scale_dtype = get_min_integer_numpy_type([scaling_factor])
+
+        data *= np.array(scaling_factor, dtype=scale_dtype)
 
     if not no_offset:
-        data += value_offset
+
+        offset_dtype = None
+        if isinstance(value_offset, six.integer_types):
+            offset_dtype = get_min_integer_numpy_type([value_offset])
+
+        data += np.array(value_offset, dtype=offset_dtype)
 
     # Restore the original mask if necessary, removing any additional mask applied above for Special_Constants
     if (special_constants is not None) and (mask is not None):

pds4_tools/utils/compat.py

@@ -6,6 +6,8 @@
 import inspect
 from xml.etree import ElementTree as ET
 
+import numpy as np
+
 from ..extern import six
 
 # OrderedDict compat (Python 2.7+ and 3.1+)
@@ -26,6 +28,16 @@
 ET_Element_iter = ET_Element.iter if hasattr(ET_Element, 'iter') else ET_Element.getiterator
 ET_ParseError = ET.ParseError if hasattr(ET, 'ParseError') else None
 
+# NumPy compat (NumPy 2.0+)
+try:
+    np_unicode = np.unicode_
+except AttributeError:
+    np_unicode = np.str_
+
+try:
+    np_issubclass = np.issubclass_
+except AttributeError:
+    np_issubclass = issubclass
 
 # signature.bind(...).arguments compat (Python 3.3+)
 def bind_arguments(func, *args, **kwargs):

tox.ini

@@ -1,10 +1,11 @@
 [tox]
 envlist =
-    {py310}-numpy{1_21_0,latest}
-    {py39}-numpy{1_19_3,latest}
-    {py38,py37}-numpy{1_15_0,latest}
-    {py27,py35,py36}-numpy{1_13_0,1_8_0,latest}
-    {py34}-numpy{1_13_0,1_8_0,latest}-attrs_py34
+    {py310}-numpy{1_21_0,1_26_4,latest}
+    {py39}-numpy{1_19_3,1_26_4,latest}
+    {py37,py38}-numpy{1_15_0,latest}
+    {py35,py36}-numpy{1_13_0,latest}
+    {py34}-numpy{1_8_0,1_13_0,latest}-attrs_py34
+    {py27}-numpy{1_8_0,1_13_0,latest}
 recreate = True
 
 [gh-actions]
@@ -29,10 +30,12 @@ basepython =
 deps =
     pytest
     attrs_py34:  attrs==20.3.0
+    numpy1_8_0:  numpy==1.8.0
     numpy1_13_0: numpy==1.13.0
     numpy1_15_0: numpy==1.15.0
     numpy1_19_3: numpy==1.19.3
     numpy1_21_0: numpy==1.21.0
+    numpy1_26_4: numpy==1.26.4
     numpylatest: numpy
 commands = py.test