Release Notes#


  • Support for old serialization formats has been removed.

  • Move the serialization implementation into own method.

  • Drop support for Python older than 3.7.



The project has been moved to the PSF organization.

  • Discard the strict attribute when serializing and deserializing responses.

  • Fix the IncompleteRead error thrown by urllib3 2.0.

  • Remove usage of utcnow in favor of timezone-aware datetimes.

  • Remove the compat module.

  • Use Python’s unittest.mock library instead of mock.

  • Add type annotations.

  • Exclude the tests directory from the wheel.


  • Revert the change “switch lockfile to filelock” to fix the compatibility issue.


  • Discard the strict attribute when serializing and deserializing responses.

  • Fix the IncompleteRead error thrown by urllib3 2.0.

  • Exclude the tests directory from the wheel.



  • Dropped support for Python 2.7, 3.4, 3.5.

  • Reduced memory usage when caching large files.


Rather than using compressed JSON for caching values, we are now using MessagePack ( MessagePack has the advantage that that serialization and deserialization is faster, especially for caching large binary payloads.


This release introduces the cachecontrol.heuristics.LastModified heuristic. This uses the same behaviour as many browsers to base expiry on the Last-Modified header when no explicit expiry is provided.


The biggest change is the introduction of using compressed JSON rather than pickle for storing cached values. This allows Python 3.4 and Python 2.7 to use the same cache store. Previously, if a cache was created on 3.4, a 2.7 client would fail loading it, causing an invalid cache miss. Using JSON also avoids the exec call used in pickle, making the cache more secure by avoiding a potential code injection point. Finally, the compressed JSON is a smaller payload, saving a bit of space.

In order to support arbitrary binary data in the JSON format, base64 encoding is used to turn the data into strings. It has to do some encoding dances to make sure that the bytes/str types are correct, so please open a new issue if you notice any issues.

This release also introduces the cachecontrol.heuristics.ExpiresAfter heuristic. This allows passing in arguments like a datetime.timedelta in order to configure that all responses are cached for the specific period of time.


This is an important release as it changes what is actually cached. Rather than caching requests’ Response objects, we are now caching the underlying urllib3 response object. Also, the response will not be cached unless the response is actually consumed by the user.

These changes allowed the reintroduction of .raw support.

Huge thanks goes out to @dstufft for these excellent patches and putting so much work into CacheControl to allow cached responses to behave exactly as a normal response.

  • FileCache Updates (via @dstufft)

    • files are now hashed via sha-2

    • files are stored in a namespaced directory to avoid hitting os limits on the number of files in a directory.

    • use the io.BytesIO when reading / writing (via @alex)

  • #19 Allow for a custom controller via @cournape

  • #17 use highest protocol version for pickling via @farwayer

  • #16 FileCache: raw field serialization via @farwayer


  • #16: All cached responses get None for a raw attribute.

  • #13 Switched to md5 encoded keys in file cache (via @mxjeff)

  • #11 Fix timezones in tests (via @kaliko)