Changelog¶

3.5.1 (2026-09-08)¶

Only pass the compiler option -mno-omit-leaf-frame-pointer when the compiler supports it, checked with a tiny probe program. The option is unsupported outside of x86 and ARM64, so this fixes building from source on other architectures, like PowerPC, which failed since the option was added in version 3.3.0.

PR #692. Thanks to Colin Watson for pointing this out and linking to the workaround in Debian Bug #1146407.
Fix the mocked datetime.date.today() and datetime.datetime.today() to be exact for all supported dates, like datetime.datetime.now(). Previously, they went through a floating-point timestamp, which could round the microseconds, or even the whole day, for dates far in the future.

PR #690.
Fix the class decorator to stop time travelling when tearDownClass() raises an exception, or when setUpClass() raises an exception not deriving from Exception, such as the skip outcome from pytest.skip(). Previously, time remained mocked for the rest of the process in these cases.

PR #678.
Fix Traveller.move_to() to keep the current timezone mocked when the given destination is unsupported. Previously, the timezone was restored before the destination was checked, leaving it unmocked whilst still time travelling.

PR #679.
Fix the time_machine pytest fixture to not try to stop a traveller that failed to start.

PR #680.
Fix the mocked time.strftime() to raise TypeError when passed None as its second argument, like the unmocked function, rather than treating it as the current time.

PR #681.
Fix the Migration CLI to keep the rewritten imports on the same line when a freezegun import shares its line with other code, like if TYPE_CHECKING: from freezegun import freeze_time, FakeDate. Previously, the remaining from freezegun import FakeDate was moved to a new line, outside the block.

PR #682.
Extend the Migration CLI to migrate uses of FrozenDateTimeFactory in string annotations, like freezer: "FrozenDateTimeFactory". Previously, the import was removed whilst such annotations were left referring to it.

PR #683.
Fix the Migration CLI to not rename a freezer argument when the function already has a time_machine argument, which produced invalid syntax.

PR #684.
Fix the Migration CLI to report positions in the rewritten file, rather than the original. Previously, positions could be off when rewrites earlier in the file changed the number of lines, or the length of the same line.

PR #686.
Fix the Migration CLI to not report a freezegun usage for migrated calls with a parenthesized module name, like (freezegun).freeze_time(...).

PR #687.
Fix the Migration CLI to leave a pass statement when removing every unused FrozenDateTimeFactory import in a block, rather than producing invalid syntax.

PR #688.
Make TimeMachineFixture importable even when pytest isn’t installed. This avoids triggering possibly-missing-attribute or possibly-missing-import violations in ty for test files with annotations.

Thanks to Nick Pope in PR #675.

3.5.0 (2026-08-25)¶

Support None as a destination for travel() and Traveller.move_to(), meaning the current time. Use this with tick=False to freeze time at the present moment, matching freezegun’s behaviour when freeze_time() is called with no arguments.

PR #666.
Extend the Migration CLI to migrate aliased imports: import freezegun as fg and from freezegun import freeze_time as ft, plus calls using such aliases.

PR #662.
Extend the Migration CLI to migrate the pytest.mark.freeze_time marker in module-level and class-level pytestmark assignments.

PR #663.
Extend the Migration CLI to migrate freeze_time() calls and markers that pass tz_offset with a literal zero value. The argument is dropped, since a zero offset has no effect.

PR #664.
Extend the Migration CLI to migrate freeze_time() calls and markers that pass real_asyncio. The argument is dropped, whatever its value, since time-machine does not mock time.monotonic(), so asyncio event loops always see real time.

PR #665.
Extend the Migration CLI to migrate freeze_time() calls and markers with no destination argument, which freeze at the current time. None is added as the destination, using the new support for None destinations (above).

PR #667.
Extend the Migration CLI to migrate freeze_time() calls and markers that pass ignore. The argument is dropped, since it works around problems with freezegun’s module patching, which time-machine’s C-level mocking doesn’t have.

PR #671.
Make the Migration CLI report freezegun-related usages that it recognizes but cannot migrate, with their positions, like:
```
example/tests.py:9:2: freeze_time usage not migrated
```
This makes it easier to find and fix the remaining usages manually.

PR #668 and PR #669.
Make the Migration CLI report files that cannot be parsed, rather than skipping them silently. This helps notice files using syntax from a newer Python version than the tool is run with.

PR #674.
Extend the Migration CLI to migrate “raw use” assignments of freeze_time() to variables or self. attributes used with start() and stop(), covering unittest setUp() / tearDown() patterns like self.freezer = freeze_time(...) with self.addCleanup(self.freezer.stop).

PR #670 and PR #672.
Extend the Migration CLI to migrate imports and uses of freezegun’s FrozenDateTimeFactory class, often used to annotate the freezer fixture argument, to time-machine’s newly-documented equivalent, time_machine.TimeMachineFixture.

PR #673.

3.4.0 (2026-08-10)¶

Publish wheels for Python 3.15.

Thanks to Edgar Ramírez Mondragón in PR #655.
Fix a reference leak in the patched datetime.datetime.utcnow() that occurred when its Python 3.12+ DeprecationWarning was raised as an error, such as under -W error. Each such call leaked a reference to the active traveller object.

PR #656.
Extend the Migration CLI to migrate patterns from freezegun’s pytest plugins, pytest-freezegun and pytest-freezer: the freezer fixture and the pytest.mark.freeze_time marker. Also make it migrate freeze_time() context managers that bind the result with as, including converting tick() method calls to shift().

PR #658. Thanks to Javier Buzzi for the initial work in PR #562.
Fix the Migration CLI to handle trailing commas in freeze_time() calls when adding tick=False. Previously, it produced invalid syntax, like travel("2023-01-01",, tick=False).

PR #658.
Fix the Migration CLI to not indent imports kept from a rewritten from-import when the import doesn’t start its line, such as after if True: on one line. Previously, the kept import gained leading whitespace, producing invalid syntax.

PR #658.

3.3.1 (2026-08-04)¶

Fix small rounding errors in destination calculations.

Previously, time-machine computed destinations in floating-point seconds, which could lead to small rounding errors: up to a few hundred nanoseconds for present-day destinations, growing to several microseconds by the year 2500. This change fixes the pipeline to use integer nanoseconds throughout, so time-travel is precise for any supported date.

PR #651.

3.3.0 (2026-07-31)¶

Support Python 3.15.

PR #631.
Mock datetime.date.today() directly, for Python 3.15 support.

Previously time-machine was mocked only indirectly, since CPython implemented it by calling cls.fromtimestamp(time.time()). Python 3.15 added a fast path that reads the system clock directly (CPython Issue #130980), so time travel no longer affected it, which the new mock fixes. Consequently, there are new escape hatch functions: escape_hatch.datetime.date.today() and escape_hatch.datetime.datetime.today().

Thanks to Miro Hrončok and Karolina Surma for the report in Issue #610, Lumír ‘Frenzy’ Balhar for the fix in PR #618, and Maurycy Pawłowski-Wieroński for review
Support isolated subinterpreters, as created by concurrent.interpreters on Python 3.14+. Previously, importing time-machine in an isolated subinterpreter failed with an ImportError.

PR #644.
Move the intermediary functions that patched functions call from Python to C. This change reduces the overhead of calling patched functions like time.time() while time travelling, making them around 10-20% faster.

PR #643.
Fix returning datetime subclasses from their now() and utcnow() methods while time travelling. Previously, time-machine would always return a plain datetime.datetime instance, rather than the subclass.

PR #643.
Fix hiding of the DeprecationWarning from datetime.utcnow() on Python 3.12+ while time travelling. Previously, the mocked version of the function did not raise the warning at all, so deprecated calls could pass unnoticed in tests. The warning is attributed to the calling code, like the real function does.

The escape hatch function escape_hatch.datetime.datetime.utcnow() also now attributes its warning to the calling code, rather than to a line within time-machine.

PR #649. Thanks to Tamir Duberstein for the report in Issue #445 and Anders Kaseorg for the initial implementation in PR #486.
Fix the standard uuid library functions uuid1(), uuid6(), and uuid7() to correctly generate values for the destination when time travels backwards. These functions cache the timestamp of the value they generated most recently, and never generate a value stamped before it. Previously, after time-travelling backwards, these functions would keep generating values stamped for the cached timestamp, rather than the new destination. The caches are now reset whenever time travel starts, stops, or moves backwards.

PR #648. Thanks to Diego Carrasco for the report in Issue #601 and initial work in PR #597.
Build with frame pointers enabled, preparation for PEP 831.

PR #627.
Stop shipping wheels for 32-bit Linux and Windows.

PR #605.
Stop shipping wheels for free-threaded Python 3.13 since cibuildwheel 4.0.0 dropped support for building them.
Extend the Migration CLI to update freeze_time() calls that pass tick.

Thanks to George-Cristian Birzan for the report in Issue #609 and tanren for the fix in PR #636.
Extend the Migration CLI to rewrite freeze_time() decorators on async functions.

Thanks to George-Cristian Birzan for the report in Issue #608 and Sanjay Santhanam for the fix in PR #640.
Fix the Migration CLI to leave relative imports called freezegun alone. Previously it rewrote from .freezegun import freeze_time, which imports from a local module that happens to share freezegun’s name.

PR #645.
Extend the Migration CLI to rewrite from-imports that import freeze_time alongside other names. For example, from freezegun import freeze_time, FakeDate now becomes import time_machine plus from freezegun import FakeDate.

PR #647.

3.2.0 (2025-12-17)¶

Add time_machine.naive_mode to control how time-machine interprets naive datetimes.

The default mode is MIXED, which preserves existing behaviour: naive datetime objects and date objects are interpreted as UTC, while naive datetime strings are interpreted as local time.

Three alternative modes are available:
- UTC: naive datetimes are always interpreted as UTC.
- LOCAL: naive datetimes are interpreted as local time, matching Python’s default semantics, and freezegun.
- ERROR: naive datetimes raise a RuntimeError, ensuring your tests are isolated from the current timezone.
Note

It’s recommended you use LOCAL or ERROR to avoid confusion around naive datetimes.

PR #591. Thanks to Paolo Melchiorre for review.

Thanks to PhML, Stefaan Lippens, Matthieu Rigal, Nikita Demir, Steve Mavens, Andy Freeland, and Paul Ganssle for their input on Issue #257.
Raise RuntimeError when attempting to start time travelling if freezegun is active.

This change should help avoid surprises when migrating complex test suites from freezegun to time-machine.

PR #590.

3.1.0 (2025-11-21)¶

Optimize patching of uuid module. By avoiding using unittest.mock, this small overhead from starting time_machine.travel() has been reduced about 20x, from ~600ns to ~30ns by one benchmark.

PR #585.

3.0.0 (2025-11-18)¶

Remove mocking of time.monotonic() and time.monotonic_ns().

This mocking caused too many issues, such as causing freezes in asyncio event loops (Issue #387), preventing pytest-durations from timing tests correctly (Issue #505), and triggering timeouts in psycopg (Issue #509). The root cause here is that mocking the monotonic clock breaks its contract, allowing it to move backwards when it’s meant to only move forwards.

As an alternative, use unittest.mock to mock the monotonic function for the specific tested modules that need it. That means that your code should import monotonic() or monotonic_ns() directly, so that your tests can mock it in those places only. For example, if your system under test looks like:
```
# example.py
from time import monotonic


def measurement():
    start = monotonic()
    ...
    end = monotonic()
    return end - start
```
…then your tests can mock monotonic() like this:
```
from unittest import TestCase, mock

import example


class MeasurementTests(TestCase):
    def test_success(self):
        with mock.patch.object(example, "monotonic", side_effect=[0.0, 1.23]):
            result = example.measurement()
        assert result == 1.23
```
PR #560.
Parse str destinations with datetime.fromisoformat() first, before falling back to dateutil if installed. datetime.fromisoformat() can parse most valid ISO 8601 formats, with better performance and no extra dependencies.

PR #578.
Make the dependency on dateutil optional. To include dateutil support, install with the dateutil extra:
```
python -m pip install time-machine[dateutil]
```
Beware that some of the formats that dateutil parses are ambiguous and may lead to unexpected results.

PR #576.
Rename the Coordinates class to Traveller, to match the recommended context manager variable name.

Thanks to Matt Wang in PR #535.
Drop Python 3.9 support.
Make the escape_hatch functions raise ValueError when called outside of time-travelling, rather than triggering segmentation faults.

PR #580.

2.19.0 (2025-08-19)¶

Add marker support to the pytest plugin. Decorate tests with @pytest.mark.time_machine(<destination>) to set time during a test, affecting function-level fixtures as well.

Thanks to Javier Buzzi in PR #499.
Add asynchronous context manager support to time_machine.travel(). You can now use async with time_machine.travel(...): in asynchronous code, per the documentation.

PR #556.
Import date and time functions once in the C extension.

This should improve speed a little bit, and avoid segmentation faults when the functions have been swapped out, such as when freezegun is in effect. (time-machine still won’t apply if freezegun is in effect.)

PR #555.

2.18.0 (2025-08-18)¶

Update the migration CLI to detect unittest classes based on whether they use self.assert* methods like self.assertEqual().
Fix free-threaded Python warning: RuntimeWarning: The global interpreter lock (GIL) has been enabled... as seen on Python 3.13+.

Thanks to Javier Buzzi in PR #531.
Add support to travel() for datetime destinations with tzinfo set to datetime.UTC (datetime.timezone.utc).

Thanks to Lawrence Law in PR #502.
Prevent segmentation faults in unlikely scenarios, such as if the time_machine module cannot be imported.

PR #543, PR #545.
Make travel() fully unpatch date and time functions when travel ends. This may fix certain edge cases.

Issue #532.

2.17.0 (2025-08-05)¶

Include wheels for Python 3.14.

Thanks to Edgar Ramírez Mondragón in PR #521.
Support free-threaded Python.

Thanks to Javier Buzzi in PR #500.
Add a new CLI for migrating code from freezegun to time-machine.

Install with pip install time-machine[cli] and run with python -m time_machine migrate.

See more in the documentation.
Move the documentation to Read the Docs, and add a retro-futuristic logo.

2.16.0 (2024-10-08)¶

Drop Python 3.8 support.

2.15.0 (2024-08-06)¶

Include wheels for Python 3.13.

2.14.2 (2024-06-29)¶

Fix SystemError on Python 3.13 and Windows when starting time travelling.

Thanks to Bernát Gábor for the report in Issue #456.

2.14.1 (2024-03-22)¶

Fix segmentation fault when the first travel() call in a process uses a timedelta.

Thanks to Marcin Sulikowski for the report in Issue #431.

2.14.0 (2024-03-03)¶

Fix utcfromtimestamp() warning on Python 3.12+.

Thanks to Konstantin Baikov in PR #424.
Fix class decorator for classmethod overrides.

Thanks to Pavel Bitiukov for the reproducer in PR #404.
Avoid calling deprecated uuid._load_system_functions() on Python 3.9+.

Thanks to Nikita Sobolev for the ping in CPython Issue #113308.
Support Python 3.13 alpha 4.

Thanks to Miro Hrončok in PR #409.

2.13.0 (2023-09-19)¶

Add support for datetime.timedelta to time_machine.travel().

Thanks to Nate Dudenhoeffer in PR #298.
Fix documentation about using local time for naive date(time) strings.

Thanks to Stefaan Lippens in PR #306.
Add shift() method to the time_machine pytest fixture.

Thanks to Stefaan Lippens in PR #312.
Mock time.monotonic() and time.monotonic_ns(). They return the values of time.time() and time.time_ns() respectively, rather than real monotonic clocks.

Thanks to Anthony Sottile in PR #382.

2.12.0 (2023-08-14)¶

Include wheels for Python 3.12.

2.11.0 (2023-07-10)¶

Drop Python 3.7 support.

2.10.0 (2023-06-16)¶

Support Python 3.12.

2.9.0 (2022-12-31)¶

Build Windows ARM64 wheels.
Explicitly error when attempting to install on PyPy.

Thanks to Michał Górny in PR #315.

2.8.2 (2022-09-29)¶

Improve type hints for time_machine.travel() to preserve the types of the wrapped function/coroutine/class.

2.8.1 (2022-08-16)¶

Actually build Python 3.11 wheels.

2.8.0 (2022-08-15)¶

Build Python 3.11 wheels.

2.7.1 (2022-06-24)¶

Fix usage of ZoneInfo from the backports.zoneinfo package. This makes ZoneInfo support work for Python < 3.9.

2.7.0 (2022-05-11)¶

Support Python 3.11 (no wheels yet, they will only be available when Python 3.11 is RC when the ABI is stable).

2.6.0 (2022-01-10)¶

Drop Python 3.6 support.

2.5.0 (2021-12-14)¶

Add time_machine.escape_hatch, which provides functions to bypass time-machine.

Thanks to Matt Pegler for the feature request in Issue #206.

2.4.1 (2021-11-27)¶

Build musllinux wheels.

2.4.0 (2021-09-01)¶

Support Python 3.10.

2.3.1 (2021-07-13)¶

Build universal2 wheels for Python 3.8 on macOS.

2.3.0 (2021-07-05)¶

Allow passing tick to Coordinates.move_to() and the pytest fixture’s time_machine.move_to(). This allows freezing or unfreezing of time when travelling.

2.2.0 (2021-07-02)¶

Include type hints.
Convert C module to use PEP 489 multi-phase extension module initialization. This makes the module ready for Python sub-interpreters.
Release now includes a universal2 wheel for Python 3.9 on macOS, to work on Apple Silicon.
Stop distributing tests to reduce package size. Tests are not intended to be run outside of the tox setup in the repository. Repackagers can use GitHub’s tarballs per tag.

2.1.0 (2021-02-19)¶

Release now includes wheels for ARM on Linux.

2.0.1 (2021-01-18)¶

Prevent ImportError on Windows where time.tzset() is unavailable.

2.0.0 (2021-01-17)¶

Release now includes wheels for Windows and macOS.
Move internal calculations to use nanoseconds, avoiding a loss of precision.
After a call to move_to(), the first function call to retrieve the current time will return exactly the destination time, copying the behaviour of the first call to travel().
Add the ability to shift timezone by passing in a ZoneInfo timezone.
Remove tz_offset argument. This was incorrectly copied from freezegun. Use the new timezone mocking with ZoneInfo instead.
Add pytest plugin and fixture time_machine.
Work with Windows’ different epoch.

1.3.0 (2020-12-12)¶

Support Python 3.9.
Move license from ISC to MIT License.

1.2.1 (2020-08-29)¶

Correctly return naive datetimes from datetime.utcnow() whilst time travelling.

Thanks to Søren Pilgård and Bart Van Loon for the report in Issue #52.

1.2.0 (2020-07-08)¶

Add move_to() method to move to a different time whilst travelling. This is based on freezegun’s move_to() method.

1.1.1 (2020-06-22)¶

Move C-level clock_gettime() and clock_gettime_ns() checks to runtime to allow distribution of macOS wheels.

1.1.0 (2020-06-08)¶

Add shift() method to move forward in time by a delta whilst travelling. This is based on freezegun’s tick() method.

Thanks to Alex Subbotin for the feature in PR #27.
Fix to work when either clock_gettime() or CLOCK_REALTIME is not present. This happens on some Unix platforms, for example on macOS with the official Python.org installer, which is compiled against macOS 10.9.

Thanks to Daniel Crowe for the fix in PR #30.

1.0.1 (2020-05-29)¶

Fix datetime.now() behaviour with the tz argument when not time-travelling.

1.0.0 (2020-05-29)¶

First non-beta release.
Added support for tz_offset argument.
tick=True will only start time ticking after the first method return that retrieves the current time.
Added nestability of travel().
Support for time.time_ns() and time.clock_gettime_ns().

1.0.0b1 (2020-05-04)¶

First release on PyPI.