gh-143959: Split datetime tests requiring _datetime by VanshAgarwal24036 · Pull Request #143997 · python/cpython

VanshAgarwal24036 · 2026-01-18T13:40:00Z

This PR splits the datetime tests into two groups:

Pure-Python tests that do not require _datetime
Fast C-accelerated tests that require _datetime

The _Fast tests are only loaded when _datetime is available, allowing
the test suite to run correctly on builds where _datetime is not present.

No test behavior is changed.

Issue: test_sys globally requires _datetime for cpython_only test #143959

VanshAgarwal24036 · 2026-01-18T13:41:24Z

Test-only change; no user-visible impact, so no NEWS entry is needed.

serhiy-storchaka · 2026-01-20T12:13:40Z

- fast_tests = import_fresh_module(TESTS,
- fresh=['datetime', '_strptime'],
- blocked=['_pydatetime'])
+ pure_tests = import_fresh_module(


You can now restore the old formatting.

Thanks — I’ve restored the original formatting.

It is not restored yet.

I’ve now restored the original structure by keeping the two import_fresh_module() calls adjacent and moved the _datetime availability check outside that block.

vstinner · 2026-01-21T10:20:06Z

It's not easy to test this change on CPython since _datetime is now a built-in module.

I tried to build _datetime as a shared extension:

diff --git a/Modules/Setup b/Modules/Setup
index 7d816ead843..828368cf7c2 100644
--- a/Modules/Setup+++ b/Modules/Setup@@ -126,14 +126,14 @@ PYTHONPATH=$(COREPYTHONPATH)
# modules are to be built as shared libraries (see above for more
# detail; also note that *static* or *disabled* cancels this effect):
-#*shared*+*shared*
# Modules that should always be present (POSIX and Windows):
#_asyncio _asynciomodule.c
#_bisect _bisectmodule.c
#_csv _csv.c
-#_datetime _datetimemodule.c+_datetime _datetimemodule.c
#_decimal _decimal/_decimal.c
#_heapq _heapqmodule.c
#_interpchannels _interpchannelsmodule.c
diff --git a/Modules/Setup.bootstrap.in b/Modules/Setup.bootstrap.in
index 65a1fefe72e..3e7344382da 100644
--- a/Modules/Setup.bootstrap.in+++ b/Modules/Setup.bootstrap.in@@ -13,7 +13,7 @@ _signal signalmodule.c
_tracemalloc _tracemalloc.c
_suggestions _suggestions.c
# needs libm and on some platforms librt
-_datetime _datetimemodule.c+#_datetime _datetimemodule.c
# modules used by importlib, deepfreeze, freeze, runpy, and sysconfig
_codecs _codecsmodule.c

But the build fails with:

/usr/bin/ld: Python/pylifecycle.o: in function `pycore_init_types':
/home/vstinner/python/main/Python/pylifecycle.c:779:(.text+0x9db): undefined reference to `_PyDateTime_InitTypes'

I tested the change with:

diff --git a/Lib/test/test_datetime.py b/Lib/test/test_datetime.py
index 32556bcb184..5b6d52859e2 100644
--- a/Lib/test/test_datetime.py+++ b/Lib/test/test_datetime.py@@ -9,6 +9,7 @@
def load_tests(loader, tests, pattern):
try:
try:
+ raise ImportError
import _datetime
except ImportError:
has_datetime_ext = False

Output:

...
test_tricky (test.datetimetester.TestTimezoneConversions_Pure.test_tricky) ... ok
test_folds (test.datetimetester.ZoneInfoTest_Pure.test_folds) ... ok
test_gaps (test.datetimetester.ZoneInfoTest_Pure.test_gaps) ... ok
test_system_transitions (test.datetimetester.ZoneInfoTest_Pure.test_system_transitions) ... ok
setUpClass (test.datetimetester.CapiTest_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.ExtensionModuleTests_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.IranTest_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.Oddballs_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestDate_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestDateOnly_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestDateTime_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestDateTimeTZ_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestLocalTimeDisambiguation_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestModule_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestSubclassDateTime_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTZInfo_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTime_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTimeDelta_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTimeTZ_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTimeZone_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.TestTimezoneConversions_Fast) ... skipped 'requires _datetime module'
setUpClass (test.datetimetester.ZoneInfoTest_Fast) ... skipped 'requires _datetime module'
...

VanshAgarwal24036 · 2026-01-21T10:27:14Z

Thanks for testing this, as well as the explanation.

Yes, the intention is just this behavior: when _datetime is notimportable, all Fast tests are skipped once at the class level, while Pure tests are imported. To identify tests are still run. I agree that the simulation of the IMPORTERROR, or at least what would raise an IMPORTERROR practical way to exercise this right.

vstinner · 2026-01-21T10:32:51Z

I hacked Python to build again _datetime as an extension module and I removed ./build/lib.linux-x86_64-3.15/_datetime.cpython-315d-x86_64-linux-gnu.so.

Result: test_datetime fails.

0:00:00 load avg: 1.15 [1/1] test_datetime
Failed to call load_tests:
Traceback (most recent call last):
File "/home/vstinner/python/main/Lib/unittest/loader.py", line 113, in loadTestsFromModule
return load_tests(self, tests, pattern)
File "/home/vstinner/python/main/Lib/test/test_datetime.py", line 38, in load_tests
for name, cls in module.__dict__.items():
^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute '__dict__'. Did you mean: '__dir__'?
test test_datetime crashed -- Traceback (most recent call last):
File "/home/vstinner/python/main/Lib/test/libregrtest/single.py", line 210, in _runtest_env_changed_exc
_load_run_test(result, runtests)
~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
File "/home/vstinner/python/main/Lib/test/libregrtest/single.py", line 165, in _load_run_test
regrtest_runner(result, test_func, runtests)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/vstinner/python/main/Lib/test/libregrtest/single.py", line 118, in regrtest_runner
test_result = test_func()
File "/home/vstinner/python/main/Lib/test/libregrtest/single.py", line 162, in test_func
return run_unittest(test_mod, runtests)
File "/home/vstinner/python/main/Lib/test/libregrtest/single.py", line 38, in run_unittest
raise Exception("errors while loading tests")
Exception: errors while loading tests
0:00:00 load avg: 1.15 [1/1/1] test_datetime failed (uncaught exception)

I'm not convinced that this change works as expected.

vstinner · 2026-01-21T10:33:30Z

My hack to build _datetime as a shared extension:

Details

diff --git a/Modules/Setup b/Modules/Setup
index 7d816ead843..828368cf7c2 100644
--- a/Modules/Setup+++ b/Modules/Setup@@ -126,14 +126,14 @@ PYTHONPATH=$(COREPYTHONPATH)
# modules are to be built as shared libraries (see above for more
# detail; also note that *static* or *disabled* cancels this effect):
-#*shared*+*shared*
# Modules that should always be present (POSIX and Windows):
#_asyncio _asynciomodule.c
#_bisect _bisectmodule.c
#_csv _csv.c
-#_datetime _datetimemodule.c+_datetime _datetimemodule.c
#_decimal _decimal/_decimal.c
#_heapq _heapqmodule.c
#_interpchannels _interpchannelsmodule.c
diff --git a/Modules/Setup.bootstrap.in b/Modules/Setup.bootstrap.in
index 65a1fefe72e..3e7344382da 100644
--- a/Modules/Setup.bootstrap.in+++ b/Modules/Setup.bootstrap.in@@ -13,7 +13,7 @@ _signal signalmodule.c
_tracemalloc _tracemalloc.c
_suggestions _suggestions.c
# needs libm and on some platforms librt
-_datetime _datetimemodule.c+#_datetime _datetimemodule.c
# modules used by importlib, deepfreeze, freeze, runpy, and sysconfig
_codecs _codecsmodule.c
diff --git a/Modules/_datetimemodule.c b/Modules/_datetimemodule.c
index 8f64e572bd6..754e7c7da22 100644
--- a/Modules/_datetimemodule.c+++ b/Modules/_datetimemodule.c@@ -7690,6 +7690,12 @@ static PyModuleDef datetimemodule = {
PyMODINIT_FUNC
PyInit__datetime(void)
{
+ PyInterpreterState *interp = PyInterpreterState_Get();+ PyStatus status = _PyDateTime_InitTypes(interp);+ if (_PyStatus_EXCEPTION(status)) {+ Py_ExitStatusException(status);+ }+
return PyModuleDef_Init(&datetimemodule);
}
diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c
index 88dbdb6d139..c7fdd3dfd27 100644
--- a/Python/pylifecycle.c+++ b/Python/pylifecycle.c@@ -776,11 +776,6 @@ pycore_init_types(PyInterpreterState *interp)
return status;
}
- status = _PyDateTime_InitTypes(interp);- if (_PyStatus_EXCEPTION(status)) {- return status;- }-
return _PyStatus_OK();
}

serhiy-storchaka · 2026-01-21T10:35:15Z

It's not easy to test this change on CPython since _datetime is now a built-in module.

This is why I have not merged this PR yet.

I consider also the idea of adding '_datetime' to the fresh list. Then the result is None and this can be easily tested. But I am not sure that it is safe to reload the _datetime module, especially in older Python versions.

VanshAgarwal24036 · 2026-01-21T10:41:14Z

Thanks for investigating this and for conducting such an in-depth experiment.

My plan would be to decide whether there should be Fast tests in the function load_tests().
time, not later during execution. If _datetime is not importable, the Fast test module is simply not created or iterated over at all, and only the Pure tests are loaded.

This prevents us from having to use the skip() within setUpClass(). It load_tests() never sees a partially or incompletely initialized module, Which is what resulted in the NoneType error you encountered. Other
words, presence of _datetime determines whether test suite of Fast is executed otherwise is generated in the first place.

Does this approach look OK to you?

VanshAgarwal24036 · 2026-01-23T15:38:16Z

I think this CI failure unrelated to this change. The PR only modifies Lib/test/test_datetime.py, while the failure is a
C-level assertion in test_external_inspection . I did not observe this failure locally.

vstinner · 2026-01-26T16:11:24Z

"Tests / Windows (free-threading) / Build and test (arm64)": test_external_inspection failed with Assertion failed: (res_o != NULL) ^ (_PyErr_Occurred(tstate) != NULL), file C:\a\cpython\cpython\Python\generated_cases.c.h, line 3917. It's an unrelated issue.

serhiy-storchaka

Thank you for your patch, @vstinner. I used it to test an alternative PR #145248.

serhiy-storchaka · 2026-02-26T10:21:41Z


- test_modules = [pure_tests, fast_tests]
- test_suffixes = ["_Pure", "_Fast"]
+ test_modules = [pure_tests]


Now the accelerated module is not tested at all.

github-actions · 2026-05-05T06:44:44Z

This PR is stale because it has been open for 30 days with no activity.

vstinner · 2026-05-05T10:17:17Z

#145248 has been merged instead, so I close this PR.

pythongh-143959: Split datetime tests requiring _datetime
df22dbc

VanshAgarwal24036 requested review from abalkin and pganssle as code owners January 18, 2026 13:40

bedevere-appBot added the tests Tests in the Lib/test dir label Jan 18, 2026

bedevere-appBot mentioned this pull request Jan 18, 2026
test_sys globally requires _datetime for cpython_only test #143959
Open

bedevere-appBot added the awaiting review label Jan 18, 2026

pythongh-143959: Split datetime tests requiring _datetime
3b42bcd

serhiy-storchaka added skip news needs backport to 3.13 bugs and security fixes needs backport to 3.14 bugs and security fixes labels Jan 20, 2026

serhiy-storchaka reviewed Jan 20, 2026
View reviewed changes

Comment threadLib/test/test_datetime.py Outdated
Comment threadLib/test/test_datetime.py

VanshAgarwal24036 added 2 commits January 20, 2026 16:51

pythongh-143959: Skip datetime Fast tests when _datetime absent
ecfbbbd

Removed whitespaces
b227ee9

VanshAgarwal24036 requested a review from serhiy-storchaka January 20, 2026 11:27

Correct Lint error
281077d

serhiy-storchaka reviewed Jan 20, 2026
View reviewed changes

VanshAgarwal24036 added 2 commits January 20, 2026 18:05

pythongh-143959: Skip Fast datetime tests if _datetime is unavailable
9fafddb

restore
f21b5de

VanshAgarwal24036 requested a review from serhiy-storchaka January 20, 2026 13:02

serhiy-storchaka reviewed Jan 20, 2026
View reviewed changes

Comment threadLib/test/test_datetime.py Outdated

Restore Changes
20b6944

VanshAgarwal24036 requested a review from serhiy-storchaka January 21, 2026 04:00

vstinner reviewed Jan 21, 2026
View reviewed changes

Comment threadLib/test/test_datetime.py Outdated

Renamed Variable
b589068

VanshAgarwal24036 requested a review from vstinner January 21, 2026 10:04

Debug
73a161a

serhiy-storchaka reviewed Feb 26, 2026
View reviewed changes

serhiy-storchaka mentioned this pull request Mar 22, 2026
gh-143959: Fix test_datetime if _datetime is unavailable #145248
Merged

github-actionsBot added the stale Stale PR or inactive for long period of time. label May 5, 2026

vstinner closed this May 5, 2026

Uh oh!

Uh oh!

Conversation

VanshAgarwal24036 commented Jan 18, 2026• edited by bedevere-app BotLoading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

VanshAgarwal24036 commented Jan 18, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

serhiy-storchakaJan 20, 2026

Choose a reason for hiding this comment

Uh oh!

VanshAgarwal24036Jan 20, 2026

Choose a reason for hiding this comment

Uh oh!

serhiy-storchakaJan 20, 2026

Choose a reason for hiding this comment

Uh oh!

VanshAgarwal24036Jan 20, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

vstinner commented Jan 21, 2026

Uh oh!

VanshAgarwal24036 commented Jan 21, 2026

Uh oh!

vstinner commented Jan 21, 2026

Uh oh!

vstinner commented Jan 21, 2026

Uh oh!

serhiy-storchaka commented Jan 21, 2026

Uh oh!

VanshAgarwal24036 commented Jan 21, 2026

Uh oh!

VanshAgarwal24036 commented Jan 23, 2026

Uh oh!

vstinner commented Jan 26, 2026

Uh oh!

serhiy-storchaka left a comment

Choose a reason for hiding this comment

Uh oh!

serhiy-storchakaFeb 26, 2026

Choose a reason for hiding this comment

Uh oh!

github-actionsBot commented May 5, 2026

Uh oh!

vstinner commented May 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

VanshAgarwal24036 commented Jan 18, 2026•
edited by bedevere-app Bot
Loading