No. The plan for Cargo is to use crates.io, like npm or Rubygems do with npmjs.org and rubygems.org.
We plan to support git repositories as a source of packages forever, because they can be used for early development and temporary patches, even when people use the registry as the primary source of packages.
We think that it’s very important to support multiple ways to download packages, including downloading from GitHub and copying packages into your package itself.
That said, we think that crates.io offers a number of important benefits, and will likely become the primary way that people download packages in Cargo.
For precedent, both Node.js’s npm and Ruby’s bundler support both a central registry model as well as a Git-based model, and most packages are downloaded through the registry in those ecosystems, with an important minority of packages making use of git-based packages.
Some of the advantages that make a central registry popular in other languages include:
- Discoverability. A central registry provides an easy place to look for existing packages. Combined with tagging, this also makes it possible for a registry to provide ecosystem-wide information, such as a list of the most popular or most-depended-on packages.
- Speed. A central registry makes it possible to easily fetch just the metadata for packages quickly and efficiently, and then to efficiently download just the published package, and not other bloat that happens to exist in the repository. This adds up to a significant improvement in the speed of dependency resolution and fetching. As dependency graphs scale up, downloading all of the git repositories bogs down fast. Also remember that not everybody has a high-speed, low-latency Internet connection.
Cargo handles compiling Rust code, but we know that many Rust packages link against C code. We also know that there are decades of tooling built up around compiling languages other than Rust.
Our solution: Cargo allows a package to specify a script
(written in Rust) to run before invoking
rustc. Rust is leveraged to
implement platform-specific configuration and refactor out common build
functionality among packages.
Indeed. While we intend Cargo to be useful as a standalone way to compile Rust packages at the top-level, we know that some people will want to invoke Cargo from other build tools.
We have designed Cargo to work well in those contexts, paying attention to things like error codes and machine-readable output modes. We still have some work to do on those fronts, but using Cargo in the context of conventional scripts is something we designed for from the beginning and will continue to prioritize.
Rust itself provides facilities for configuring sections of code based
on the platform. Cargo also supports platform-specific
dependencies, and we plan to support more per-platform
Cargo.toml in the future.
In the longer-term, we’re looking at ways to conveniently cross-compile packages using Cargo.
We support environments through the use of profiles to support:
- environment-specific flags (like
-g --opt-level=0for development and
- environment-specific dependencies (like
hamcrestfor test assertions).
All commits to Cargo are required to pass the local test suite on Windows. If, however, you find a Windows issue, we consider it a bug, so please file an issue.
The purpose of a
Cargo.lock is to describe the state of the world at the time
of a successful build. It is then used to provide deterministic builds across
whatever machine is building the package by ensuring that the exact same
dependencies are being compiled.
This property is most desirable from applications and packages which are at the
very end of the dependency chain (binaries). As a result, it is recommended that
all binaries check in their
For libraries the situation is somewhat different. A library is not only used by
the library developers, but also any downstream consumers of the library. Users
dependent on the library will not inspect the library’s
Cargo.lock (even if it
exists). This is precisely because a library should not be deterministically
recompiled for all users of the library.
If a library ends up being used transitively by several dependencies, it’s
likely that just a single copy of the library is desired (based on semver
compatibility). If Cargo used all of the dependencies'
then multiple copies of the library could be used, and perhaps even a version
In other words, libraries specify semver requirements for their dependencies but cannot see the full picture. Only end products like binaries have a full picture to decide what versions of dependencies should be used.
As of January 22nd, 2016, crates.io rejects all packages (not just libraries) with wildcard dependency constraints.
While libraries can, strictly speaking, they should not. A version requirement
* says “This will work with every version ever,” which is never going
to be true. Libraries should always specify the range that they do work with,
even if it’s something as general as “every 1.x.y version.”
As one of the most frequent interactions with Cargo, the question of why the
configuration file is named
Cargo.toml arises from time to time. The leading
C was chosen to ensure that the manifest was grouped with other
similar configuration files in directory listings. Sorting files often puts
capital letters before lowercase letters, ensuring files like
Cargo.toml are placed together. The trailing
.toml was chosen to emphasize
the fact that the file is in the TOML configuration
Cargo does not allow other names such as
emphasize the ease of how a Cargo repository can be identified. An option of
many possible names has historically led to confusion where one case was handled
but others were accidentally forgotten.
Cargo is often used in situations with limited or no network access such as airplanes, CI environments, or embedded in large production deployments. Users are often surprised when Cargo attempts to fetch resources from the network, and hence the request for Cargo to work offline comes up frequently.
Cargo, at its heart, will not attempt to access the network unless told to do so. That is, if no crates comes from crates.io, a git repository, or some other network location, Cargo will never attempt to make a network connection. As a result, if Cargo attempts to touch the network, then it's because it needs to fetch a required resource.
Cargo is also quite aggressive about caching information to minimize the amount
of network activity. It will guarantee, for example, that if
cargo build (or
an equivalent) is run to completion then the next
cargo build is guaranteed to
not touch the network so long as
Cargo.toml has not been modified in the
meantime. This avoidance of the network boils down to a
and a populated cache of the crates reflected in the lock file. If either of
these components are missing, then they're required for the build to succeed and
must be fetched remotely.
As of Rust 1.11.0, Cargo understands a new flag,
--frozen, which is an
assertion that it shouldn't touch the network. When passed, Cargo will
immediately return an error if it would otherwise attempt a network request.
The error should include contextual information about why the network request is
being made in the first place to help debug as well. Note that this flag does
not change the behavior of Cargo, it simply asserts that Cargo shouldn't touch
the network as a previous command has been run to ensure that network activity
shouldn't be necessary.
--offline flag was added in Rust 1.36.0. This flag tells Cargo to not
access the network, and try to proceed with available cached data if possible.
You can use
cargo fetch in one project to download dependencies before
going offline, and then use those same dependencies in another project with
--offline flag (or configuration value).
For more information about vendoring, see documentation on source replacement.