crap4rust

crap4rust computes CRAP scores for Rust functions by combining complexity and test coverage.

Definition

What is a CRAP score? CRAP (Change Risk Anti-Patterns) combines cognitive complexity and test coverage: CRAP(m) = comp(m)² × (1 − cov(m))³ + comp(m). Functions above a score of 30 are flagged as crappy — they are complex enough that their lack of test coverage makes them a maintenance risk.

Full derivation of every term — how cognitive complexity is scored construct by construct, how coverage is matched and duplicate records resolved, and how a project-level verdict is computed from every function's own — is in docs/FORMULA.md.

It is published as the Cargo subcommand package cargo-crap4rust, so the command is cargo crap4rust.

Documentation

Install

cargo install cargo-crap4rust
cargo install cargo-llvm-cov
rustup component add llvm-tools

crap4rust scores complexity against coverage, and it measures that coverage by running cargo-llvm-cov, which in turn needs the llvm-tools rustup component. Neither comes with crap4rust, and neither is optional: without them any run that has to generate coverage fails.

Passing --coverage <PATH> with a report you already produced is the one way to run without them.

Development

just stage1
just stage2

Both must be green before a change is complete. Stage 1 is formatting, clippy and tests; stage 2 is cargo xtask stage2, which runs, in order: cargo stern4rust (house coding rules), cargo crap4rust (complexity against coverage), cargo twin4rust (every source file has a mirrored test file) and cargo iceberg4rust (file risk).

Stage 1 is not cargo built-ins alone here, which is the one way this repository differs from its siblings: crap4rust is the coverage tool, so its own validation tests drive the built binary through the path that shells out to cargo llvm-cov. A checkout without that install fails stage 1, before ever reaching the gate.

Everything the two stages need, none of which ships with cargo:

cargo-llvm-cov is needed by stage 1 as well, not only by the gate: this repository is the coverage tool, so its own validation tests drive the built binary down the path that shells out to it.

The CRAP gate measures this repository with the published cargo-crap4rust rather than the binary in the working tree, so a regression in the tree cannot excuse itself.

CI (.github/workflows/ci.yml) runs both stages on Ubuntu, Windows and macOS for every pull request and every push to main.

License

Licensed under the MIT License.

What It Does

• Computes a CRAP score for each discovered Rust function
• Generates coverage automatically with cargo llvm-cov when --coverage is omitted
• Prints a single report to the console
• Supports multiple --package flags for one aggregated report
• Defaults to analysing all workspace members when --package is omitted in a multi-package workspace
• Does not count try-operator propagation with ? as cognitive complexity
• Supports --output-format json for structured CI-friendly output
• Supports --warn-threshold to set the warning level independently from the crappy threshold
• Supports --warn-only to report without failing the exit code even when thresholds are exceeded
• Supports --all-features to activate all Cargo features during the coverage build
• Excludes test-only code from discovery whether it's an inline #[cfg(test)] mod tests { ... } block or a file-based #[cfg(test)] mod tests; submodule
• Cognitive complexity scoring lives in its own dedicated module

Examples

Analyse the default scope for a manifest:

• single-package manifest: analyses the root package
• multi-package workspace: analyses all workspace members unless --package is provided

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml

Analyse one specific package in a workspace and override the default all-members selection:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core

Analyse multiple packages and produce one combined console report:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --package app-validation

Use a precomputed coverage export instead of generating coverage automatically:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --coverage C:\Projects\my-workspace\target\coverage.json

Use stricter project thresholds:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --threshold 25 --project-threshold 3.0 --strict

Pass Cargo feature flags to the coverage build:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --features host-tests

Disable default features and enable specific ones:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --no-default-features --features host-analysis

Activate all Cargo features during the coverage build:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --all-features

Include test targets in the analysis:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-validation --include-test-targets

Exclude specific source paths from analysis:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --package app-core --exclude-path src/scenarios

Output as JSON for CI pipelines:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --output-format json

Set a custom warning threshold:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --warn-threshold 15

Report without failing the exit code even when thresholds are exceeded:

cargo crap4rust --manifest-path C:\Projects\my-workspace\Cargo.toml --warn-only

Real Workspace Example

Example run against the etheram-raft workspace (all four workspace members — node, node-infra, validation, system-tests — analysed together, since --package is omitted):

cargo crap4rust --manifest-path C:\Projects\etheram\etheram-raft\Cargo.toml

Report excerpt:

crap4rust report for node, node-infra, validation, system-tests

Summary: total_functions=419, crappy_functions=3, crappy_percent=0.7%, threshold=30.0, project_threshold=5.0%, verdict=warn. Production functions only — test code and generated code excluded by default.

The report above is abbreviated to the highest-scoring rows, with function names and file paths shortened for readability. When coverage is generated automatically, cargo llvm-cov also emits normal build and test output before the final crap4rust report.

Try-operator propagation with ? is treated as error forwarding rather than decision-making complexity, so CRAP scoring reflects branching and control-flow structure instead of penalising straightforward Result propagation.

See docs/IMPLEMENTED-FEATURES.md for the shipped feature set and docs/ROADMAP.md for the broader plan.