clap 3.1 is here! Clap is a CLI argument parser for Rust and the v3.1 releases focuses on API cleanup slated for clap 4.0. See the CHANGELOG for details.

clap 3.0 was in development for 4 years and though we saw comparisons to Half-life 3 in response to the release, we also saw people who cited the long gaps between breaking releases as a motivation for using it. For clap to stay relevant we feel we need to avoid the stagnation of long release cycles while keeping things smooth for the users where clap is already "good enough". The v3.1 release is a major step in trying to strike that balance.

clap's Model of Evolution

Traditionally, all features have been self-contained within the clap::App. Customization has been done through runtime flags and cargo feature flags. When a breaking change is needed, we batched them up into a large release with just a CHANGELOG to help you making it easy to miss the more subtle changes among basic transformations of "X was renamed to Y".

This has led to an ever growing API, a bloated code base, and gated evolution of user applications.

For example, clap's --help flag can:

Its unlikely the compiler can detect which you are using and compile out the other implementations. So even though you will only ever use one of these, we have to compile support for all of them and include all of them in the final binary. Its no wonder clap takes ~3x as long to compile as other parsers and results in ~20x the binary size. This frequently gets excused as "clap provides a lot for you" but I feel there needs to be a better way, that a CLI parser can be as powerful as you need without paying for what you don't need.

A better way?

We want to be able to continue to extend clap while:

Towards the end of clap 3's development, we started to look into how to minimize upgrade impact. We decided to leverage the compiler's deprecation messages to take care of the more mechanical changes so we can focus the CHANGELOG on the more subtle changes that might be missed otherwise. We spent a lot of time adding back in removed APIs, providing a smooth transition path from structopt, etc. I feel like this was an overall success, both from my own experience of upgrading my applications to clap 3 and from feedback we got from users about how smooth the process was.

As an example of new feature development, we had a request for response-file support. Following the existing patterns, this would have been built directly into clap's parser. We might satisfy the immediate request but it opens the door for people wanting it to fill their needs. Some prior art shows there are uses for customizing the prefix character. We'd also need to adapt to support different syntaxes in the wild, including whether:

Instead, we created the argfile crate.

clap Moving Forward

Our new guidelines in clap's CONTRIBUTING.md:

Our releases fall into one of:

If your change does not fit within a "patch" release, please coordinate with the clap maintainers for how to handle the situation.

Some practices to avoid breaking changes

While we expect this to evolve over time, we feel this is a good start to helping meet the needs from earlier.

What this means for clap 3.1

In v3.1, you'll find that we've changed a large swath of the API with deprecations. We looked to our breaking-change issues for v4.0 and focused on a common theme of changes that can be implemented now by introducing a new API and deprecating an old one. This allows us to develop iteratively and allow people to migrate in piecemeal rather than relying on large flag-day upgrades.

One catch I've found with this is how I've been turning warnings into errors in CI. I've been running something like:

$ cargo clippy --workspace --all-features --all-targets -- -deny warnings

and didn't pay attention to the fact that my CI will start failing if a dependency marks an item as deprecated. Without a lock file, your PRs might see failures unrelated to the change at hand. With a lock file, you will be forced to respond to all deprecations just to upgrade. Some see this as a good thing but I find the higher the cost of an upgrade, the more people avoid them and the slower you move. Being able to decouple the upgrade and the migration can be a big help for evolving your code.

I'd recommend changing your CI to:

$ cargo clippy --workspace --all-features --all-targets -- --deny warnings --allow deprecated

Participating in the Conversation

If you'd like to look over what we are hoping to accomplish in the future, check out our milestones. They represent our areas of focus and aren't exclusive of other work happening). We are also doing a cross-issue brainstorming of what clap's future API might look like. Please pipe in with any feedback; we recognize that hearing your ideas is likely the only way to meet our ambitious goals!