2019 was a transformative year for urllib3 and I'm hoping to keep up the pace with Python's most used HTTP client. Here are a list of ideas, some easier achieved than others, that would be great to ship this year!
We're excited to be working with Tidelift for another year as one of the original lifted Python projects. They've enabled us to provide a comprehensive security disclosure workflow and help me and another maintainer to "keep the lights on" so to speak for the project.
However for some of the larger goals mentioned in this post we'll likely require dedicated developer-hours to achieve. In 2019 ~78% of urllib3's $23,580 in funding came from grants. What urllib3 was able to achieve with the grants has increased the security of the project and benefited millions of people, your generous organization could help us do the same in 2020!
If your organization benefits from urllib3 being secure, maintained, and performant, and maintained please consider contacting me to discuss a grant opportunity or subscribing to Tidelift to keep urllib3 and many more Python projects secure and maintained.
Python's HTTP stack has seen a surge of activity and will need even more helping hands to accomplish our goals and keep Python on the bleeding edge in the HTTP space and keep our dependable existing libraries up-to-date.
If you're looking to get started in Python Open Source and have an interest in making high-impact contributions, learn a lot about networking, or bolster your resume reach out to me.
Python's HTTP clients all rely on a trust store named
which is the Mozilla CA Certificate Store,
packaged and released on PyPI. This works well for the Internet
at large as it's the same CA certificate store that powers Firefox
and curl. However many organizations and IT machine configurations
only ensure that the operating system CA bundle is up to date
certifi be on separate means organizations can't
control the safety of their users the same way that they control
safety of browsers and other tools.
We'd also want this work to be usable by other projects in the Python ecosystem as this is an important feature for any library that uses TLS.
in particular looks like a promising start.
Thanks to the security trail-blazing of many organizations TLS 1.2 is becoming the new minimum security level for the Internet. CDNs and cloud services mean almost all services deployed there use TLS 1.2+ whether their users know it or not by taking configuration of TLS termination away from users.
Browsers are also doing their part by deprecating and eventually removing TLS 1.0 and 1.1 from future versions. Firefox, Edge, Safari, and Chrome have all committed to removing TLS <1.2 in 2020.
This means 2020 is a great time for us to also push the better-security-by-force envelope. :)
Setting TLS 1.2+ means that we'll be secure against TLS downgrade attacks or if a vulnerability is found in TLS 1.0 or 1.1 our users will already be protected. It also means that we can restrict our default cipher suites to rock-solid secure ones instead of ones that are there only to be compatible with older TLS versions.
Even with all you'll still be able to use TLS 1.0 or 1.1 if you
specify those versions manually via
ssl_version in the case you have to connect
to an older server.
Python 2.7.9 is the first Python version that supported
Currently urllib3 supports all Python 2.7 versions but maintains a large chunk of
code that allows us to work with Python versions that don't have proper
objects. This means that we're forced to do hostname verification ourselves rather
than rely on OpenSSL's own implementation of certificate verification.
We'll continue to support Python interpreters that don't have a
ssl module, however HTTPS connections won't be possible
as is the case today.
We will also continue to in general support Python 2.7 for the forseeable future and won't consider removing support via a breaking change until Pip and other upstream projects also drop support.
For some data on what percentage of downloads would be effected I've created a small data table. The number of downloads effected by the change would be ~4%.
Zstandard defined in
RFC 8478 is looking
to be a great replacement for the long-dominant
gzip as a content encoding.
With both great compression ratios and quick parallelizable CPU usage I'm
guessing this technology will soon take the Internet by storm, especially at
the edge / CDN layer.
When Zstandard is standardized and accepted it'd be great to already have the feature in place to immediately start giving our users the new performance benefits.
The number of Python versions and configurations that urllib3 supports is quite large, testing against Linux, macOS, and Windows to ensure we're compatible with all major OS flavors. With Travis supporting 5 and AppVeyor continuing to support only 1 concurrent job we're feeling cornered into long CI durations.
We're also wondering how the faster Python release cadence will impact our CI matrix of versions we test against and support.
Fortunately we've been experiencing success with GitHub Actions as a CI platform in other projects with the 20 concurrent jobs as well as great flexibility and access to all platforms from one service and would like to switch wholesale to GA from Travis and AppVeyor.
If you have questions or there are projects you think should be prioritized contact me on email or Twitter!