⏰ Whenever#

https://img.shields.io/pypi/v/whenever.svg?style=flat-square&color=blue https://img.shields.io/pypi/pyversions/whenever.svg?style=flat-square https://img.shields.io/pypi/l/whenever.svg?style=flat-square&color=blue https://img.shields.io/badge/mypy-strict-forestgreen?style=flat-square https://img.shields.io/badge/coverage-100%25-forestgreen?style=flat-square https://img.shields.io/github/actions/workflow/status/ariebovenberg/whenever/tests.yml?branch=main&style=flat-square https://img.shields.io/readthedocs/whenever.svg?style=flat-square

Strict, predictable, and typed datetimes

Do you cross your fingers every time you work with datetimes, hoping that you didn’t mix naive and aware? or that you converted to UTC everywhere? or that you avoided the many pitfalls of the standard library? There’s no way to be sure…

✨ Until now! ✨

Whenever is a datetime library designed from the ground up to enforce correctness. Mistakes become red squiggles in your IDE, instead of bugs in production.


  • 🎯 Distinct classes with well-defined behavior

  • 🔧 Fixes pitfalls that arrow and pendulum don’t

  • ✅ Enforce correctness without runtime checks

  • 🔌 Based on familiar concepts and standards

  • 💎 Simple and obvious; no frills or surprises

  • 📖 Thoroughly documented and tested

  • 🎁 One file; no third-party dependencies


>>> from whenever import (
...    # Explicit types for different use cases
...    UTCDateTime,     # -> Great for codebases that normalize to UTC
...    OffsetDateTime,  # -> Localized times without ambiguities
...    ZonedDateTime,   # -> Full-featured IANA timezone support
...    NaiveDateTime,   # -> Detached from any timezone
... )

>>> py311_release = UTCDateTime(2022, 10, 24, hour=17)
UTCDateTime(2022-10-24 17:00:00Z)
>>> pycon23_start = OffsetDateTime(2023, 4, 21, hour=9, offset=-6)
OffsetDateTime(2023-04-21 09:00:00-06:00)

# Simple, explicit conversions
>>> py311_release.as_zoned("Europe/Paris")
ZonedDateTime(2022-10-24 19:00:00+02:00[Europe/Paris])
>>> pycon23_start.as_local()  # example: system timezone in NYC
LocalDateTime(2023-04-21 11:00:00-04:00)

# Comparison and equality across aware types
>>> py311_release > pycon23_start
>>> py311_release == py311_release.as_zoned("America/Los_Angeles")

# Naive type that can't accidentally mix with aware types
>>> hackathon_invite = NaiveDateTime(2023, 10, 28, hour=12)
>>> # Naïve/aware mixups are caught by typechecker
>>> hackathon_invite - py311_release
>>> # Only explicit assumptions will make it aware
>>> hackathon_start = hackathon_invite.assume_zoned("Europe/Amsterdam")
ZonedDateTime(2023-10-28 12:00:00+02:00[Europe/Amsterdam])

# DST-aware operators
>>> hackathon_end = hackathon_start.add(hours=24)
ZonedDateTime(2022-10-29 11:00:00+01:00[Europe/Amsterdam])

# Lossless round-trip to/from text (useful for JSON/serialization)
>>> py311_release.canonical_format()
>>> ZonedDateTime.from_canonical_format('2022-10-24T19:00:00+02:00[Europe/Paris]')
ZonedDateTime(2022-10-24 19:00:00+02:00[Europe/Paris])

# Conversion to/from common formats
>>> py311_release.rfc2822()  # also: from_rfc2822()
"Mon, 24 Oct 2022 17:00:00 GMT"
>>> pycon23_start.rfc3339()  # also: from_rfc3339()

# Basic parsing
>>> OffsetDateTime.strptime("2022-10-24+02:00", "%Y-%m-%d%z")
OffsetDateTime(2022-10-24 00:00:00+02:00)

# If you must: you can access the underlying datetime object
>>> pycon23_start.py_datetime().ctime()
'Fri Apr 21 09:00:00 2023'

Read more in the full overview or API reference.

Why not…?#

The standard library#

The standard library is full of quirks and pitfalls. To summarize the detailed blog post:

  1. Incompatible concepts of naive and aware are squeezed into one class

  2. Operators ignore Daylight Saving Time (DST)

  3. The meaning of “naive” is inconsistent (UTC, local, or unspecified?)

  4. Non-existent datetimes pass silently

  5. It guesses in the face of ambiguity

  6. False negatives on equality of ambiguous times between timezones

  7. False positives on equality of ambiguous times within the same timezone

  8. datetime inherits from date, but behaves inconsistently

  9. datetime.timezone isn’t enough for full-featured timezones.

  10. The local timezone is DST-unaware


Pendulum is full-featured datetime library, but it’s hamstrung by the decision to inherit from the standard library datetime. This means it inherits most of the pitfalls mentioned above, with the notable exception of DST-aware addition/subtraction.


Arrow is probably the most historically popular datetime library. Pendulum did a good write-up of the issues with Arrow. It addresses fewer of datetime’s pitfalls than Pendulum.


DateType mostly fixes the issue of mixing naive and aware datetimes, and datetime/date inheritance during type-checking, but doesn’t address the other pitfalls. The type-checker-only approach also means that it doesn’t enforce correctness at runtime, and it requires developers to be knowledgeable about how the ‘type checking reality’ differs from the ‘runtime reality’.


Dateutil attempts to solve some of the issues with the standard library. However, it only adds functionality to work around the issues, instead of removing the pitfalls themselves. This still puts the burden on the developer to know about the issues, and to use the correct functions to avoid them. Without removing the pitfalls, it’s still very likely to make mistakes.


It’s unmaintained, but does have an interesting approach. By enforcing UTC, it bypasses a lot of issues with the standard library. To do so, it sacrifices the ability to represent offset, zoned, and local datetimes. So in order to perform any timezone-aware operations, you need to convert to the standard library datetime first, which reintroduces the issues.


This library is a lot more explicit about the different types of datetimes, addressing issue of naive/aware mixing with UTC, local, and zoned datetime subclasses. It doesn’t address the other datetime pitfalls though.


  • 🧪 0.x: get to feature-parity, process feedback, and tweak the API:

    • ✅ Datetime classes

    • ✅ Durations and periods

    • 🚧 Date and time of day (separate from datetime)

    • 🚧 Parsing and formatting

    • ❓ Different calendar systems

  • 🔒 1.0: API stability and backwards compatibility

  • ⚡️ 2.0: Reimplement in Rust for performance

  • 🐍 future: Inspire a standard library improvement


Why isn’t it a drop-in replacement for the standard library?

Fixing the issues with the standard library requires a different API. Keeping the same API would mean that the same issues would remain.

Why not inherit from datetime?

Not only would this keep most of the issues with the standard library, it would result in brittle code: many popular libraries expect datetime exactly, and don’t work with subclasses.

Why so many classes?

The different classes correspond to different ways of representing time. By having separate classes, you can be more explicit about what you mean, and the API can better enforce correctness.

What is the performance impact?

Because whenever wraps the standard library, head-to-head performance will always be slightly slower. However, because whenever removes the need for many runtime checks, it may result in a net performance gain in real-world applications.

Why not a C or Rust extension?

It actually did start out as a Rust extension. But since the wrapping code is so simple, it didn’t make much performance difference. Since it did make the code a lot more complex, a simple pure-Python implementation was preferred. Once the API is stable, a Rust implementation will be considered again.

Is this production-ready?

The core functionality is complete and stable and the goal is to reach 1.0 soon. The API may change slightly until then. Of course, it’s still a relatively young project, so the stability relies on you to try it out and report any issues!

Versioning and compatibility policy#

Whenever follows semantic versioning. Until the 1.0 version, the API may change with minor releases. Breaking changes will be avoided as much as possible, and meticulously explained in the changelog. Since the API is fully typed, your typechecker and/or IDE will help you adjust to any API changes.

⚠️ Note: until 1.x, pickled objects may not be unpicklable across versions. After 1.0, backwards compatibility of pickles will be maintained as much as possible.


This project is inspired by the following projects. Check them out!


Contributions are welcome! Please open an issue or a pull request.

⚠️ Note: big changes should be discussed in an issue first. This is to avoid wasted effort if the change isn’t a good fit for the project.

⚠️ Note: Some tests are skipped on Windows. These tests use unix-specific features to set the timezone for the current process. As a result, Windows isn’t able to run certain tests that rely on the system timezone. It appears that this functionality is not available on Windows.

Setting up a development environment#

You’ll need poetry installed. An example of setting up things up:

poetry install

# To run the tests with the current Python version

# if you want to build the docs
pip install -r docs/requirements.txt

# Various checks
mypy src/ tests/
flake8 src/ tests/

# autoformatting
black src/ tests/
isort src/ tests/

# To run the tests with all supported Python versions
# Alternatively, let the github actions on the PR do it for you
pip install tox
tox -p auto


Indices and tables#