pytools/test/test_persistent_dict.py::test_{attrs,dataclass}_hashing failures on big-endian

[mgorny]

I'm sorry that I've missed this previously but looks like cee5027 introduced two test regressions on big-endian:

$ pytest pytools/test/test_persistent_dict.py::test_{attrs,dataclass}_hashing
========================================================= test session starts =========================================================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0
rootdir: /home/mgorny/pytools
configfile: pyproject.toml
plugins: xdist-3.6.1
collected 2 items                                                                                                                     

pytools/test/test_persistent_dict.py FF                                                                                         [100%]

============================================================== FAILURES ===============================================================
_________________________________________________________ test_attrs_hashing __________________________________________________________

    def test_attrs_hashing() -> None:
        attrs = pytest.importorskip("attrs")
    
        keyb = KeyBuilder()
    
        @attrs.define
        class MyAttrs:
            name: str
            value: int
    
>       assert (keyb(MyAttrs("hi", 1)) == "5b6c5da60eb2bd0f")  # type: ignore[call-arg]
E       AssertionError: assert '626f565387d8b006' == '5b6c5da60eb2bd0f'
E         
E         - 5b6c5da60eb2bd0f
E         + 626f565387d8b006

pytools/test/test_persistent_dict.py:593: AssertionError
_______________________________________________________ test_dataclass_hashing ________________________________________________________

    def test_dataclass_hashing() -> None:
        keyb = KeyBuilder()
    
        @dataclass
        class MyDC:
            name: str
            value: int
    
>       assert keyb(MyDC("hi", 1)) == "d1a1079f1c10aa4f"
E       AssertionError: assert 'e36c7d4306a3670a' == 'd1a1079f1c10aa4f'
E         
E         - d1a1079f1c10aa4f
E         + e36c7d4306a3670a

pytools/test/test_persistent_dict.py:569: AssertionError
======================================================= short test summary info =======================================================
FAILED pytools/test/test_persistent_dict.py::test_attrs_hashing - AssertionError: assert '626f565387d8b006' == '5b6c5da60eb2bd0f'
FAILED pytools/test/test_persistent_dict.py::test_dataclass_hashing - AssertionError: assert 'e36c7d4306a3670a' == 'd1a1079f1c10aa4f'
========================================================== 2 failed in 0.35s ==========================================================

Since we are using the system byte order now, the hashes are different on big-endian and little-endian systems. I suppose this can be fixed either by updating the test to have different expected values depending on sys.byteorder, or by mocking it to force little-endian:

diff --git a/pytools/test/test_persistent_dict.py b/pytools/test/test_persistent_dict.py
index b04d43a..cdddb37 100644
--- a/pytools/test/test_persistent_dict.py
+++ b/pytools/test/test_persistent_dict.py
@@ -1,6 +1,7 @@
 import shutil
 import sys
 import tempfile
+import unittest.mock
 from dataclasses import dataclass
 from enum import Enum, IntEnum
 from typing import Any, Dict, Optional
@@ -566,7 +567,8 @@ def test_dataclass_hashing() -> None:
         name: str
         value: int
 
-    assert keyb(MyDC("hi", 1)) == "d1a1079f1c10aa4f"
+    with unittest.mock.patch("sys.byteorder", "little"):
+        assert keyb(MyDC("hi", 1)) == "d1a1079f1c10aa4f"
 
     assert keyb(MyDC("hi", 1)) == keyb(MyDC("hi", 1))
     assert keyb(MyDC("hi", 1)) != keyb(MyDC("hi", 2))
@@ -590,7 +592,8 @@ def test_attrs_hashing() -> None:
         name: str
         value: int
 
-    assert (keyb(MyAttrs("hi", 1)) == "5b6c5da60eb2bd0f")  # type: ignore[call-arg]
+    with unittest.mock.patch("sys.byteorder", "little"):
+        assert (keyb(MyAttrs("hi", 1)) == "5b6c5da60eb2bd0f")  # type: ignore[call-arg]
 
     assert keyb(MyAttrs("hi", 1)) == keyb(MyAttrs("hi", 1))  # type: ignore[call-arg]
     assert keyb(MyAttrs("hi", 1)) != keyb(MyAttrs("hi", 2))  # type: ignore[call-arg]

Please let me know if I should make a PR doing either of the two.

[inducer]

Upon second thought, I think cee5027 was the wrong approach. numpy supports byteswapping, so we should just standardize its representation to a fixed byte order (and revert cee5027) IMO.

[inducer]

cc @matthiasdiener