Niko Matsakis: Battery packs: Let's talk about crates, baby

This blog post describes an idea I’ve been kicking around called battery packs. Battery packs are a curated set of crates arranged around a common theme. For example, there’s a CLI battery pack that has , and ). We’ve also got some smaller ones, such as the repository.

Let’s talk about you and me

One of the key ideas from battery packs is that anybody can publish one. They are just a crate named X-battery-pack; the dependencies of that crate are your recommendations. Features are designations of common sets of crates frequently used together. The examples are your templates. And so forth.

Letting anybody create a battery pack is in contrast to the previous ideas for an “extended standard library for Rust” ^{and publishing a battery pack with libraries focused on no-std and binary size optimization.}

Being open-ended also addresses the “who decides?” question. To my mind, the best people to recommend what libraries you ought to use are other people building systems like yours. This is why I mentioned the Embedded Working Group publishing an Embedded battery pack, for example, as I think they are clearly a set of people who know their space well. But even within the embedded space there are yet smaller groups, and I imagine that sometimes it’ll make sense to get narrower. For example, perhaps a battery pack targeted , which bundles a template:

cargo bp new battery-pack

This will prompt you for the name of the battery pack you want to create and a few other things and make your crate. Then you can just use cargo add dependencies to represent the libraries you want to recommend and publish.

“Batteries” are more than dependencies

The “batteries” that you can add to your project aren’t always dependencies. They can also be “recipes” or templates. For example, the CI battery pack^{, but I think we’re going to want to expand on this.}

Giving it some structure

Battery Packs also support more than just a flat listing of dependencies/features/templates. You can group dependencies and features into categories and then, for each category, distinguish between “pick at most one” or “pick any number”. For a fun example, try cargo bp add embedded, which is derived from the banding together to publish battery packs and recommendations. These would cover the dependencies that they’re actually using.

Supporting maintainers

One of the reasons I want to have RCN-recognized battery packs is that they are a natural focal point to then prompt RCN members to fund the maintenance of those crates. I am imagining that for each sponsored battery pack vended within the RCN, there is an associated “ecosystem fund”. Companies or individuals could sponsor this fund to get access to early patches, security disclosures, etc or other perks. The money would be used to support the maintainers of those crates, to implement missing features, and so forth.

Fostering interoperability

Another value-add from battery packs is the ability to drive interop efforts. I think that as soon as we start talking about standardizing, we’re also going to recognize that there are some places where standardization is hard. For example, early conversations within the , some major companies are using their own runtimes internally. It’s not like the need for “async runtime interop” is was a popular way to parse command-line options. Then

“Always bet on the ecosystem”

I see always bet on the ecosystem as a key Rust design axiom. It’s the reason we chose a small standard library and a package manager in the first place. It’s also why battery packs are designed to be published by anyone.

But just like plants sometimes need a trellis to grow taller, any successful ecosystem reaches a point where it needs another layer of structure to help it keep growing. Without that, you have this “layer of tacic knowledge” (in idea we floated in 2016!