Handle filesystem surrogate characters in test node ID hashes

CHANGELOG.rst

@@ -5,6 +5,9 @@ Changelog
 Unreleased
 ----------
 
+* Avoid crashing when test node IDs contain surrogate characters from undecodable filesystem names.
+  In that situation, pytest’s independent cache serialization limitation still requires ``-p no:cacheprovider``.
+
 * Drop Python 3.10 support.
 
 5.0.0 (2026-09-01)

src/pytest_randomly/__init__.py

@@ -287,7 +287,7 @@ def reduce_list_of_lists(lists: list[list[T]]) -> list[T]:
 
 @lru_cache
 def _crc32(string: str) -> int:
-    return crc32(string.encode())
+    return crc32(string.encode(errors="surrogatepass"))
 
 
 if have_faker:  # pragma: no branch

tests/test_pytest_randomly.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import shutil
+import sys
 from unittest import mock
 
 import pytest
@@ -884,3 +885,16 @@ def test_xdist(n, ourtester):
 
     # Can't make any assertion on the order, since output comes back from
     # workers non-deterministically
+
+
+@pytest.mark.skipif(
+    sys.platform == "win32", reason="Requires byte-based filesystem names"
+)
+def test_undecodable_filename(ourtester):
+    path = ourtester.path / "test_\udcff.py"
+    path.write_text("def test_one():\n    assert True\n", encoding="utf-8")
+
+    # The cache provider independently cannot persist surrogate-containing node IDs.
+    result = ourtester.runpytest("-p", "no:cacheprovider", "--randomly-seed=1")
+
+    result.assert_outcomes(passed=1)