python-libfaketime: fast date/time mocking

python-libfaketime is a wrapper of libfaketime for python. Some brief details:

• Linux and OS X, Pythons 3.8 through 3.12, pypy and pypy3
• Mostly compatible with freezegun.
• Microsecond resolution.
• Accepts datetimes and strings that can be parsed by dateutil.
• Not threadsafe.
• Will break profiling. A workaround: use libfaketime.{begin, end}_callback to disable/enable your profiler (nosetest example).

Installation

$ pip install libfaketime

Usage

import datetime
from libfaketime import fake_time, reexec_if_needed

# libfaketime needs to be preloaded by the dynamic linker.
# This will exec the same command, but with the proper environment variables set.
# You can also skip this and manually manage your env (see "How to avoid re-exec").
reexec_if_needed()

def test_datetime_now():

    # fake_time can be used as a context_manager
    with fake_time('1970-01-01 00:00:01'):

        # Every calls to a date or datetime function returns the mocked date
        assert datetime.datetime.utcnow() == datetime.datetime(1970, 1, 1, 0, 0, 1)
        assert datetime.datetime.now() == datetime.datetime(1970, 1, 1, 0, 0, 1)
        assert time.time() == 1


# fake_time can also be used as a decorator
@fake_time('1970-01-01 00:00:01', tz_offset=12)
def test_datetime_now_with_offset():

    # datetime.utcnow returns the mocked datetime without offset
    assert datetime.datetime.utcnow() == datetime.datetime(1970, 1, 1, 0, 0, 1)

    # datetime.now returns the mocked datetime with the offset passed to fake_time
    assert datetime.datetime.now() == datetime.datetime(1970, 1, 1, 12, 0, 1)

remove_vars

By default, reexec_if_needed removes the LD_PRELOAD variable after the re-execution, to keep your environment as clean as possible. You might want it to stick around, for example when using parallelized tests that use subprocess like pytest-xdist, and simply for tests where subprocess is called. To keep them around, pass remove_vars=False like:

reexec_if_needed(remove_vars=False)

quiet

To avoid displaying the informative text when re-executing, you can set the quiet parameter:

reexec_if_needed(quiet=True)

timestamp_file

A common time can be shared between several execution contexts by using a file to store the time to mock, instead of environment variables. This is useful to control the time of a running process for instance. Here is a schematized use case:

reexec_if_needed(remove_vars=False)

with fake_time("1970-01-01 00:00:00", timestamp_file="/tmp/timestamp"):
    subprocess.run("/some/server/process")

with fake_time("2000-01-01 00:00:00", timestamp_file="/tmp/timestamp"):
    assert request_the_server_process_date() == "2000-01-01 00:00:00"

Performance

libfaketime tends to be significantly faster than freezegun. Here's the output of a totally unscientific benchmark on my laptop:

$ python benchmark.py
re-exec with libfaketime dependencies
timing 1000 executions of <class 'libfaketime.fake_time'>
0.021755 seconds

$ python benchmark.py freezegun
timing 1000 executions of <function freeze_time at 0x10aaa1140>
6.561472 seconds

Use with py.test

The pytest-libfaketime plugin will automatically configure python-libfaketime for you:

$ pip install pytest-libfaketime

Alternatively, you can reexec manually from inside the pytest_configure hook:

# conftest.py
import os
import libfaketime

def pytest_configure():
    libfaketime.reexec_if_needed()
    _, env_additions = libfaketime.get_reload_information()
    os.environ.update(env_additions)

Use with tox

In your tox configuration file, under the testenv bloc, add the libfaketime environment variables to avoid re-execution:

setenv =
    LD_PRELOAD = {envsitepackagesdir}/libfaketime/vendor/libfaketime/src/libfaketime.so.1
    DONT_FAKE_MONOTONIC = 1
    FAKETIME_DID_REEXEC = true

Migration from freezegun

python-libfaketime should have the same behavior as freezegun when running on supported code. To migrate to it, you can run:

find . -type f -name "*.py" -exec sed -i 's/freezegun/libfaketime/g' "{}" \;

How to avoid re-exec

In some cases - especially when your tests start other processes - re-execing can cause unexpected problems. To avoid this, you can preload the necessary environment variables yourself. The necessary environment for your system can be found by running python-libfaketime on the command line:

$ python-libfaketime
export LD_PRELOAD="/home/foo/<snip>/vendor/libfaketime/src/libfaketime.so.1"
export DONT_FAKE_MONOTONIC="1"
export FAKETIME_NO_CACHE="1"
export FAKETIME_DID_REEXEC=true

You can easily put this in a script like:

$ eval $(python-libfaketime)
$ pytest  # ...or any other code that imports libfaketime

Contributing and testing

Contributions are welcome! You should compile libfaketime before running tests:

git submodule init --update
# For Linux:
env FAKETIME_COMPILE_CFLAGS="-UFAKE_STAT -UFAKE_UTIME -UFAKE_SLEEP" make -C libfaketime/vendor/libfaketime
# For macOS
env make -C libfaketime/vendor/libfaketime

Then you can install requirements with pip install -r requirements.txt and use pytest and tox to run the tests.

uuid1 deadlock

Calling uuid.uuid1() multiple times while in a fake_time context can result in a deadlock when an OS-level uuid library is available. To avoid this, python-libtaketime will monkeypatch uuid._uuid_generate_time (or similar, it varies by version) to None inside a fake_time context. This may slow down uuid1 generation but should not affect correctness.