prefer-cargo-shear-in-ci
Suggests running an unused-dependency check (cargo-shear or cargo-machete) in CI for Rust workspaces.
What it detects
- a
Cargo.tomlthat declares dependencies, for a single crate or a workspace ([dependencies],[dev-dependencies],[build-dependencies],[workspace.dependencies], or a workspace with members) - no visible unused-dependency check in CI: no
cargo shear/cargo machete/cargo udeps, nocargo-shear/cargo-macheteinstall step - no config in
Cargo.tomlmetadata (metadata.shear/metadata.machete) and no config file (.cargo-shear.toml,.shear.toml,shear.toml,.machete.toml)
A crate that declares no dependencies is skipped, since there is nothing to check.
Why it matters
Cargo compiles every dependency declared in Cargo.toml, even when the crate never imports it, so an unused dependency still costs build time and cache space. An unused dependency can also create a spurious synchronization point in the build graph that blocks later crates. Removing unused dependencies typically cuts a few percent of build time and shrinks the target directory, with larger wins in big workspaces or when a heavy dependency is declared but unused.
cargo-shear is the current best tool: it uses rust-analyzer’s parser to analyze sources statically (no compile needed), runs once on a single platform for all targets, and also detects misplaced dependencies and unlinked source files. It supports --fix, JSON output, and CI exit codes. cargo-machete is a faster but less precise alternative.
Suggested action
Add a CI step that runs cargo-shear on the workspace and fails on findings, for example:
cargo install cargo-shear && cargo shear --deny-warningsAdd ignore configuration for the few false positives: macro-generated imports (use --expand on nightly), dependencies used only through features or conditional compilation, and cargo-hakari workspace-hack crates.
Measurement
Compare cargo build / cargo check --workspace wall-clock time, dependency count, and target directory size before and after removing unused dependencies.