cargo-guppy: track and query dependency graphs

This repository contains the source code for:

• guppy: a library for performing queries on Cargo dependency graphs
• libraries used by guppy:
  • guppy-summaries: a library for managing build summaries listing packages and features
  • target-spec: an evaluator for Cargo.toml target specifications
• tools built on top of guppy:
  • determinator: figure out what packages changed between two revisions
  • hakari: manage workspace-hack packages
  • cargo-guppy: a command-line frontend for the guppy library
• and a number of internal tools and test fixtures used to verify that guppy behaves correctly.

Use cases

guppy and cargo-guppy can be used to solve many practical problems related to dependency graphs in large Rust codebases. Some examples -- all of these are available through the guppy library, and will eventually be supported in the cargo-guppy CLI as well:

• track existing dependencies for a crate or workspace
• query direct or transitive dependencies of a subset of packages — useful when some packages have greater assurance or reliability requirements
• figure out what's causing a particular crate to be included as a dependency
• iterate over reverse dependencies of a crate in topological order
• iterate over some or all links (edges) in a dependency graph, querying if the link is a build, dev or regular dependency
• filter out dev-only dependencies while performing queries
• perform queries based on Cargo features
• simulate Cargo builds and return what packages and features would be built by it
• evaluate target specs for platform-specific dependencies
• generate summary files for Cargo builds, which can be used to:
  • receive CI feedback if a dependency is added, updated or removed, or if new features are added
  • receive CI feedback if a package is added to a high-assurance subset, or if any new features are enabled in an existing package in that subset. This can be used to flag those changes for extra scrutiny.
• print out a dot graph for a subset of crates, for formatting with graphviz

Still to come:

• a command-line query language

Development status

The core guppy code in this repository is considered mostly complete and the API is mostly stable.

We're building a number of tools on top of guppy, and those are still are under active development. Tool requirements may cause further changes in the API, but the goal will be to avoid extensive overhauls.

guppy's simulation of Cargo builds is extensively tested against upstream Cargo, and there are no known differences. Comparison testing has found a number of bugs in upstream Cargo, for example:

• v2 resolver: different handling for inactive, optional dependencies based on how they're specified
• v2 resolver: a proc macro being specified with the key "proc_macro" vs "proc-macro" causes different results
• specifying different versions in unconditional and target-specific dependency sections causes "multiple rmeta candidates" error

Production users

cargo-guppy is extensively used by the Diem Core project.

guppy is used for several lint checks. This includes basic rules that look at every workspace package separately:

• every package has fields like author and license specified
• crate names and paths should use - instead of _

to more complex rules about the overall dependency graph, such as:

• some third-party dependencies are banned from the workspace entirely, or only from default builds
• every workspace package depends on a workspace-hack crate (similar to rustc-workspace-hack)
• for any given third-party dependency, the workspace only depends on one version of it directly (transitive dependencies to other versions are still allowed)
• every workspace package is categorized as either production or test-only, and the linter checks that test-only crates are not included in production builds
• support for overlay features, which allow test-only code to be:
  • included in crates (similar to the #[cfg(test)] annotation)
  • depended on by test-only code in other crates (#[cfg(test)] does not allow this)
  • but guaranteed to be excluded from production builds

In addition, guppy-summaries is used to generate build summaries of packages and features (particularly for high-security subsets of the codebase), and changes to these sets are flagged by Diem's CI (example).

Design

guppy is written on top of the excellent petgraph library. It is a separate codebase from cargo, depending only on the stable cargo metadata format. (Some other tools in this space like cargo-tree use cargo internals directly.)

Minimum supported Rust version

The minimum supported Rust version (MSRV) is 1.48.

While a crate is pre-release status (0.x.x) it may have its MSRV bumped in a patch release. Once a crate has reached 1.x, any MSRV bump will be accompanied with a new minor version.

Contributing

See the CONTRIBUTING file for how to help out.

License

This project is available under the terms of either the Apache 2.0 license or the MIT license.