dcrowe/time-machine

Travel through time in your tests.

time-machine

Travel through time in your tests.

A quick example:

import datetime as dt
import time_machine

@time_machine.travel("1955-11-05 01:22")
def test_delorean():
    assert dt.date.today().isoformat() == "1955-11-05"

For a bit of background, see the introductory blog post.

Installation

Use pip:

python -m pip install time-machine

Python 3.6 to 3.8 supported (CPython only).

Usage

travel(destination, *, tick=True, tz_offset=None)

travel() is a class that allows time travel, to the datetime specified by destination. It does so by mocking all functions from Python's standard library that return the current date or datetime. It can be used independently, as a function decorator, or as a context manager.

destination specifies the datetime to move to. It may be:

• A datetime.datetime. If it is naive, it will be assumed to have the UTC timezone.
• A datetime.date. This will be converted to a UTC datetime with the time 00:00:00.
• A float or int specifying a Unix timestamp
• A string, which will be parsed with dateutil.parse and converted to a timestamp.

Additionally, you can provide some more complex types:

• A generator, in which case next() will be called on it, with the result treated as above.
• A callable, in which case it will be called with no parameters, with the result treated as above.

tick defines whether time continues to "tick" after travelling, or is frozen. If True, the default, successive calls to mocked functions return values increasing by the elapsed real time since the first call. So after starting travel to 0.0 (the UNIX epoch), the first call to any datetime function will return its representation of 1970-01-01 00:00:00.000000 exactly. The following calls "tick," so if a call was made exactly half a second later, it would return 1970-01-01 00:00:00.500000.

tz_offset allows you to offset the given destination. It may be a timedelta or a number of seconds, which will be added to destination. It may be negative.

Mocked Functions

All datetime functions in the standard library are mocked to move to the destination current datetime:

• datetime.datetime.now()
• datetime.datetime.utcnow()
• time.gmtime()
• time.localtime()
• time.clock_gettime() (only for CLOCK_REALTIME)
• time.clock_gettime_ns() (only for CLOCK_REALTIME)
• time.strftime()
• time.time()
• time.time_ns()

The mocking is done at the C layer, replacing the function pointers for these built-ins. Therefore, it automatically affects everywhere those functions have been imported, unlike use of unittest.mock.patch().

Usage with start() / stop()

To use independently, create and instance, use start() to move to the destination time, and stop() to move back. For example:

import datetime as dt
import time_machine

traveller = time_machine.travel(dt.datetime(1955, 11, 5))
traveller.start()
# It's the past!
assert dt.date.today() == dt.date(1955, 11, 5)
traveller.stop()
# We've gone back to the future!
assert dt.date.today() > dt.date(2020, 4, 29)

travel() instances are nestable, but you'll need to be careful when manually managing to call their stop() methods in the correct order, even when exceptions occur. It's recommended to use the decorator or context manager forms instead, to take advantage of Python features to do this.

Usage with shift() method

The start() method and entry of the context manager both return a Coordinates object that corresponds to the time travel. This has a shift() method that takes one argument, delta, which moves the current time. delta may be a timedelta or a number of seconds, which will be added to destination.

For example:

import datetime as dt
import time_machine

with time_machine.travel(0, tick=False) as traveller:
    assert time.time() == 0

    traveller.shift(dt.timedelta(seconds=100))
    assert time.time() == 100

Function Decorator

When used as a function decorator, time is mocked during the wrapped function's duration:

import time
import time_machine

@time_machine.travel("1970-01-01 00:00 +0000")
def test_in_the_deep_past():
    assert 0.0 < time.time() < 1.0

You can also decorate asynchronous functions (coroutines):

import time
import time_machine

@time_machine.travel("1970-01-01 00:00 +0000")
async def test_in_the_deep_past():
    assert 0.0 < time.time() < 1.0

Beware: time is a global state - see below.

Context Manager

When used as a context manager, time is mocked during the with block:

def test_in_the_deep_past():
    with time_machine.travel(0.0):
        assert 0.0 < time.time() < 1.0

Class Decorator

Only unittest.TestCase subclasses are supported. When applied as a class decorator to such classes, time is mocked from the start of setUpClass() to the end of tearDownClass():

import time
import time_machine
import unittest

@time_machine.travel(0.0)
class DeepPastTests(TestCase):
    def test_in_the_deep_past(self):
        assert 0.0 < time.time() < 1.0

Note this is different to unittest.mock.patch()'s behaviour, which is to mock only during the test methods.

Caveats

Time is a global state. Any concurrent threads or asynchronous functions are also be affected. Some aren't ready for time to move so rapidly or backwards, and may crash or produce unexpected results.

Also beware that other processes are not affected. For example, if you use SQL datetime functions on a database server, they will return the real time.

Comparison

There are some prior libraries that try to achieve the same thing. They have their own strengths and weaknesses. Here's a quick comparison.

unittest.mock

The standard library's unittest.mock can be used to target imports of datetime and time to change the returned value for current time. Unfortunately, this is fragile as it only affects the import location the mock targets. Therefore, if you have several modules in a call tree requesting the date/time, you need several mocks. This is a general problem with unittest.mock - see Why Your Mock Doesn't Work.

It's also impossible to mock certain references, such as function default arguments:

def update_books(_now=time.time):  # set as default argument so faster lookup
    for book in books:
        ...

Although this is rare, it's often used to optimize repeat loops.

freezegun

Steve Pulec's freezegun library is a popular solution. It provides a clear API which was much of the inspiration for time-machine.

The main drawback is its slow implementation. It essentially does a find-and-replace mock of all the places that the datetime and time modules have been imported. This gets around the problems with using unittest.mock, but it means the time it takes to do the mocking is proportional to the number of loaded modules. In large projects, this can take several seconds, an impractical overhead for an individual test.

It's also not a perfect search, since it searches only module-level imports. Such imports are definitely the most common way projects use date and time functions, but they're not the only way. freezegun won’t find functions that have been “hidden” inside arbitrary objects, such as class-level attributes.

It also can't affect C extensions that call the standard library functions, including (I believe) Cython-ized Python code.

python-libfaketime

Simon Weber's python-libfaketime wraps the libfaketime library. libfaketime replaces all the C-level system calls for the current time with its own wrappers. It's therefore a "perfect" mock for the current process, affecting every single point the current time might be fetched, and performs much faster than freezegun.

Unfortunately python-libfaketime comes with the limitations of LD_PRELOAD. This is a mechanism to replace system libraries for a program as it loads (explanation). This causes two issues in particular when you use python-libfaketime.

First, LD_PRELOAD is only available on Unix platforms, which prevents you from using it on Windows. This can be a complete blocker for many teams.

Second, you have to help manage LD_PRELOAD. You either use python-libfaketime's reexec_if_needed() function, which restarts (re-execs) your test process while loading, or manually manage the LD_PRELOAD environment variable. Neither is ideal. Re-execing breaks anything that might wrap your test process, such as profilers, debuggers, and IDE test runners. Manually managing the environment variable is a bit of overhead, and must be done for each environment you run your tests in, including each developer's machine.

time-machine

time-machine is intended to combine the advantages of freezegun and libfaketime. It works without LD_PRELOAD but still mocks the standard library functions everywhere they may be referenced. Its weak point is that other libraries using date/time system calls won't be mocked. Thankfully this is rare. It's also possible such python libraries can be added to the set mocked by time-machine.

One drawback is that it only works with CPython, so can't be used with other Python interpreters like PyPy. However it may possible to extend it to support other interpreters through different mocking mechanisms.

Migrating from libfaketime or freezegun

freezegun has a useful API, and python-libfaketime copies some of it, with a different function name. time-machine also copies some of freezegun's API, in travel()'s destination, tick, and tz_offset arguments, and the shift() method. There are a few differences:

• time-machine's tick argument defaults to True, because code tends to make the (reasonable) assumption that time progresses between function calls, and should normally be tested as such. Testing with time frozen can make it easy to write complete assertions, but it's quite artificial.
• freezegun's tick() method has been implemented as shift(), to avoid confusion with the tick argument. It also requires an explicit delta rather than defaulting to 1 second.

Some features aren't supported like the auto_tick_seconds argument, or the move_to() method. These may be added in a future release.

If you are only fairly simple function calls, you should be able to migrate by replacing calls to freezegun.freeze_time() and libfaketime.fake_time() with time_machine.travel().