Build & Test

Note: CI does not use --all-features -- it runs each crate with a curated, crate-specific feature combination (see .github/workflows/regression.yml). The Makefile's per-crate pattern targets (make test-<crate>, make lint-<crate>) apply the same feature combinations CI uses and are the closest local equivalent. CI also tests lindera-segmenter and lindera once more with every dictionary embedded (on Linux only), which runs the tests gated on the other dictionaries; to do the same locally, run cargo test -p lindera --features embed-ipadic,embed-ipadic-neologd,embed-unidic,embed-sudachidict,embed-ko-dic,embed-cc-cedict,embed-jieba (and likewise for lindera-segmenter).

Workspace-Wide Tests

cargo test

Quality Checks

Format Check

Verify code formatting matches the project style:

cargo fmt --all -- --check

To auto-fix formatting:

cargo fmt --all

Note: cargo fmt formats the Rust code only. For a language binding, make format-lindera-<binding> also formats the binding's Python, JavaScript, Ruby or PHP code, with the tools and versions make lint-lindera-<binding> checks it with (see below). make format-scripts does the same for the files outside the crates that make lint-scripts checks.

Lint

Run Clippy with warnings treated as errors:

cargo clippy -- -D warnings

Note: CI runs cargo clippy --all-targets -- -D warnings for each crate, with the features it tests that crate with (see .github/workflows/test-crate.yml, and the language binding jobs in .github/workflows/regression.yml), so the test code is linted too. make lint-<crate> runs the same clippy command. Run clippy locally (e.g. via make lint) before opening a PR.

For a language binding, make lint-lindera-<binding> then runs the binding's language-side linter, as the binding's CI job does: ruff for Python, Prettier for Node.js and WASM, RuboCop for Ruby, and PHP CS Fixer for PHP. Their versions are pinned (ruff and Prettier in the Makefile, RuboCop in lindera-ruby/Gemfile, PHP CS Fixer in composer.json), so a new release of a tool cannot fail CI on unchanged code. The targets install the pinned versions themselves, which needs the language's toolchain (Python 3, Node.js with npx, Bundler, Composer) and network access.

make lint-scripts lints the hand-written files that belong to no binding, as the Lint scripts CI job does: the benchmark harness in scripts/benchmarks/, which the bindings share, and the JavaScript the docs load (docs/mermaid-init.js). It runs the same pinned tools, and checks each file with the rules of the binding written in its language, so it needs all four toolchains. make lint runs it after the crates.

Documentation

API Documentation

Generate and open Rust API documentation:

cargo doc --no-deps --open

Check the crates' API documentation for rustdoc warnings, such as a broken intra-doc link or a public item linking to a private one:

make check-docs

Note: CI runs make check-docs and fails on any rustdoc warning. It documents every crate but the language bindings with all features: the libraries twice, once with public items only, as docs.rs publishes them, and once with private items too, whose doc comments only that second pass checks; then lindera-cli's binary, where the CLI's code lives, with its private items. It builds with dummy dictionaries (DOCS_RS=1) in its own target directory, target/check-docs, so it downloads no dictionary and leaves target/ alone.

The language bindings are checked one at a time, since building each needs its language's toolchain:

make check-docs-lindera-python   # also -nodejs, -ruby, -php, -wasm

Note: Each binding's CI job runs its target and fails on any rustdoc warning. The target documents the binding twice, public items only and then with private items too, with default features, which already enable every feature that adds or removes an item; no dictionary is built. lindera-ruby runs through the same RbConfig wrapper as make test-lindera-ruby, and lindera-wasm is documented for wasm32-unknown-unknown.

mdBook Documentation

Build the user-facing documentation:

mdbook build docs

Preview locally at http://localhost:3000:

mdbook serve docs

Markdown Lint

Check documentation for Markdown style issues:

markdownlint-cli2 "docs/src/**/*.md"

Rules are configured in .markdownlint.json at the repository root.

Lindera Documentation