Summary

This RFC defines a mechanism for evolving the default Rust style over time without breaking backwards compatibility, via the Rust edition mechanism.

Motivation

The current Rust style, as defined in the Rust Style Guide and as implemented by rustfmt, has some stability expectations associated with it. In particular, many projects implement continuous integration hooks that verify the style of Rust code (such as with cargo fmt --check), and changes to the style would break the CI of such projects, in addition to causing churn.

This document proposes to evolve the current Rust style, without breaking backwards compatibility, by tying style evolution to Rust edition. Code in Rust 2015, 2018, or 2021 will use the existing default style. Code in future editions (Rust 2024 and onwards) may use a new style edition.

This RFC only defines the mechanism by which we evolve the Rust style; this RFC does not define any specific style changes. Future RFCs or style-guide PRs will define future style editions. This RFC does not propose or define any specific future style editions or other formatting changes.

Explanation

rustfmt, and cargo fmt, will format code according to the default Rust style. The default Rust style varies by Rust edition. (Not every edition changes the Rust style, and thus some editions have identical default styles; Rust 2015, 2018, and 2021 all have the same default style.)

Direct invocations of rustfmt obtain the edition used for parsing Rust code from the edition option in its configuration file (rustfmt.toml or .rustfmt.toml), or via the --edition command-line option; cargo fmt obtains the edition from the edition option in Cargo.toml and passes it to rustfmt. By default, rustfmt and cargo fmt will use the same edition for style as the Rust edition used for parsing.

However, when transitioning between editions, projects may want to separately make and commit the changes for 1) transitioning to a new Rust edition and 2) transitioning to a new style edition. Keeping formatting changes in a separate commit also helps tooling ignore that commit, such as with git's blame.ignoreRevsFile.

To allow for this, rustfmt also allows configuring the style edition directly, via a separate style_edition configuration option, or --style-edition command-line option. style_edition or --style-edition, if set, always overrides edition or --edition for the purposes of styling, though edition or --edition still determines the edition for the purposes of parsing Rust code.

Note that rustfmt may not necessarily support all combinations of Rust edition and style edition; in particular, it may not support using a style edition that differs by more than one step from the Rust edition. Similarly, rustfmt need not support every existing configuration option in new style editions.

New style editions will be initially introduced as nightly-only, to make them available for testing; such nightly-only editions will produce an error if requested in stable rustfmt. Nightly versions of style editions are subject to change and do not provide stability guarantees. New style editions will get stabilized contemporaneously with the corresponding Rust edition.

The current version of the style guide will describe the latest Rust edition. Each distinct past style will have a corresponding archived version of the style guide. Note that archived versions of the style guide may not necessarily document formatting for newer Rust constructs that did not exist at the time that version of the style guide was archived. However, each style edition will still format all constructs valid in that Rust edition, with the style of those constructs coming from the first subsequent style edition providing formatting rules for that construct.

Rationale and alternatives

We could have a completely separate configuration mechanism, unrelated to editions. This would increase the development and testing burden of rustfmt, and seems unlikely to provide commensurate value. This would also increase the complexity for end users, who would have to understand two separate mechanisms for handling forwards-compatibility and wonder how they differ. We feel that since we're providing a mechanism similar to editions, we should make it clear to users that it works like editions.

We could allow style edition to vary completely independently of Rust edition. This would, for instance, allow projects to stay on old style editions indefinitely. However, this would substantially increase the development and testing burden for formatting tooling, and require more complex decisions about how old style editions format constructs that didn't exist in the corresponding Rust edition. In general, while the Rust edition mechanism allows projects to stay on old Rust editions, and projects doing so can similarly stay on the corresponding old style editions, the style edition mechanism does not exist to facilitate staying on old styles indefinitely while still moving forward to newer Rust editions.

We could leave out the separate configuration of style edition, and keep style edition in lockstep with Rust edition. This would be easier to develop and test, but would mean larger and noisier commits in projects transitioning from one edition to another.

We could keep the Rust style static forever, and never change it.

We could evolve the Rust style without a backwards-compatibility mechanism. This would result in churn in people's repositories if collaborating developers have different versions of Rust, and would break continuous-integration checks that check formatting.

Prior art

The Rust edition mechanism itself serves as prior art, as does the mechanism of nightly features remaining subject to change until stabilization.

rustfmt has a still-unstable option version = "Two" to opt into new formatting, though the exact changes this makes are not documented.

rustfmt's stability guarantees are documented in RFC 2437.

Future possibilities

Actual changes to the Rust style for Rust 2024 or future editions.

6.4 KiB Raw Permalink Blame History