07 February 2018
The primary changes are:
Add a default global event loop, eliminating the need for setting up and managing your own event loop in the vast majority of cases.
Decouple all task execution functionality from Tokio.
Up until today, creating an event loop was a manual process. Even though the vast majority of Tokio users would setup the reactor to do the same thing, everyone had to do it each time. This was partially due to the fact that there was a significant difference between running code on the Tokio reactor’s thread or from another thread (like a thread pool).
The key insight that allowed for the Tokio reform changes is that the Tokio reactor doesn’t actually have to be an executor. In other words, prior to these changes, the Tokio reactor would both power I/O resources and manage executing user submitted tasks.
Now, Tokio provides a reactor to drive I/O resources (like
UdpSocket) separately from the task executor. This means that it is easy to
create Tokio-backed networking types from any thread, making it easy to create
either single or multi threaded Tokio-backed apps.
For task execution, Tokio provides the
current_thread executor, which
behaves similarly to how the built-in tokio-core executor did. The plan is to
eventually move this executor into the
futures crate, but for now it is
provided directly by Tokio.
The Tokio reform changes have been released as 0.1. Dependencies (
mio, etc…) have not had their versions incremented. This
tokio crate to be released with minimal ecosystem disruption.
The plan is to let the changes made in this release get some usage before committing to them. Any fixes that require breaking changes will be able to be done at the same time as the release to all the other crates. The goal is for this to happen in 6-8 weeks. So please try out the changes released today and provide feedback.
This is just the beginning. Tokio has ambitious goals to provide additional functionality to get a great “out of the box” experience building asynchronous I/O applications in Rust.
In order to reach these goals as fast as possible without causing unnecessary ecosystem disruption, we will be taking a few steps.
First, similarly to the
futures 0.2 release, the
tokio crate will be
transitioned to be more of a facade. Traits and types will be broken up into a
number of sub crates and re-exported by
tokio. Application authors will be
able to depend directly on
tokio while library authors will pick and choose
the specific Tokio components that they wish to use as part of their libraries.
Each sub crate will clearly indicate its stability level. Obviously, there is an upcoming breaking change with the futures 0.2 release, but after that, fundamental building blocks will aim to remain stable for at least a year. More experimental crates will reserve the right to issue breaking changes at a quicker pace.
This means that the
tokio crate itself will be able to iterate at a faster
pace while the library ecosystem remains stable.
The pre 0.2 period will also be a period of experimentation. Additional functionality will be added to Tokio in an experimental capacity. Before an 0.2 release, an RFC will be posted covering the functionality that we would like to include in that release.
One remaining question is what to do about
tokio-proto. It was released as
part of the initial Tokio release. Since then, the focus has shifted and that
crate has not received enough attention.
I posted an issue to discuss what to do with that crate here
Please try out the changes released today. Again, the next couple of months are a period of experimentation before we commit on the next release. So, now is the time to try things out and provide feedback.