Contributing

This page details the guidelines to follow when contributing to EpiAwareADTools.jl.

Getting started

Before contributing, please: 2. New to Julia development? Start from the EpiAware site

3. Install Task for the development workflows below:

• macOS: brew install go-task/tap/go-task

• Linux: Download from releases or use your package manager

• Windows: winget install Task.Task or download from releases

4. Check out the developer documentation for advanced workflows

5. Review the project structure and development commands below

Project structure

EpiAwareADTools.jl uses several environments for different purposes.

EpiAwareADTools.jl/
├── Project.toml            # Main package environment
├── src/                    # Package source
│   ├── EpiAwareADTools.jl  # Module file: exports and centralised imports
│   ├── docstrings.jl       # DocStringExtensions @template registration
│   ├── primal.jl           # Tape-strip helpers (primal, primal_distribution)
│   ├── gamma_ad.jl         # Analytic gamma-CDF derivative machinery
│   └── ad_safe.jl          # The AD-safe evaluation hooks
├── ext/                    # Per-backend rules (ChainRulesCore, Enzyme,
│                           # ForwardDiff, Mooncake, ReverseDiff)
├── test/
│   ├── Project.toml        # Test environment
│   ├── runtests.jl         # Main test entry (TestItemRunner)
│   ├── unit/               # Unit tests (hooks)
│   ├── package/            # Quality gates
│   ├── ad/                 # AD gradient harness (own environment)
│   ├── jet/                # JET static analysis (isolated environment)
│   ├── formatter/          # JuliaFormatter check (isolated environment)
│   └── ADFixtures/         # AD gradient scenario registry
├── docs/
│   ├── Project.toml        # Documentation environment
│   ├── make.jl             # Documentation build script
│   ├── pages.jl            # Documentation navigation tree
│   └── src/                # Documentation source files
└── benchmark/
    ├── Project.toml        # Benchmark environment
    └── benchmarks.jl       # Benchmark suite definition

Files carrying a MANAGED by EpiAwarePackageTools.scaffold header are owned by the shared kit and rewritten on every sync. See EpiAwarePackageTools for which files are managed and which are yours to edit.

Development commands

This project includes a Taskfile for the development workflows.

Quick start with tasks

# Discover all available tasks
task --list

# Common development workflow
task setup      # One-time environment setup (all sub-environments)
task dev        # Format + pre-commit + fast tests + fast docs
task precommit  # Pre-commit validation

# Individual workflows
task test-fast  # Quick testing (skips quality checks)
task docs       # Full documentation build
task benchmark  # Run benchmarks

Detailed commands

For advanced usage, or when tasks don't cover a specific need, use the underlying Julia commands:

# Full test suite (recommended for CI and final checks)
julia --project=. -e 'using Pkg; Pkg.test()'

# Run tests directly (faster during development)
julia --project=test test/runtests.jl

# Skip quality tests for faster development iteration
julia --project=test test/runtests.jl skip_quality

# Run only the quality tests
julia --project=test test/runtests.jl quality_only

# Build complete documentation
julia --project=docs docs/make.jl

# Execute the benchmark suite
task benchmark

Testing strategy

Test organisation

• Unit tests: In test/unit/ (hooks.jl), covering the AD-safe hook dispatch and the tape-strip helpers on primal values

• Quality tests: Located in test/package/, tagged :quality

• AD gradient tests: Located in test/ad/, tagged :ad, with their own environment and dedicated per-backend CI

Tests are @testitems discovered with TestItemRunner. The main entry (test/runtests.jl) accepts skip_quality and quality_only arguments to scope discovery, and excludes :ad-tagged items (they run from test/ad/runtests.jl).

Quality gates

The quality gates in test/package/ guard the package's health:

• Aqua.jl: Common package issues (stale deps, ambiguities, piracy)

• ExplicitImports.jl: No implicit or stale imports, and import centralisation

• JET.jl: Static analysis for type stability (run from the isolated test/jet environment)

• JuliaFormatter: Formatting check (run from the isolated test/formatter environment)

• Docstring format: Every docstring matches the src/docstrings.jl template

• Doctest and README section checks, plus an extension-ambiguity check

Run them all with task test-quality, or a single isolated gate directly:

task test-jet         # JET static analysis
task test-formatting  # JuliaFormatter check

AD gradient harness

The AD-safe hooks exist to differentiate cleanly across every backend the ecosystem uses. The harness in test/ad/ sweeps the scenarios registered in test/ADFixtures/ across six backends: ForwardDiff, ReverseDiff, Enzyme (reverse and forward), and Mooncake (reverse and forward), each an @testitem tagged for per-backend CI selection. The scenarios differentiate each hook (and the internal _gamma_cdf) with respect to a Gamma's shape and scale.

task test-ad                             # all backends
task test-ad-backend TAG=enzyme_reverse  # a single backend

Each scenario carries a ForwardDiff reference gradient the other backends are checked against. When you add or change a hook or a derivative rule, add a matching scenario to test/ADFixtures/src/ADFixtures.jl so the sweep covers it.

Style guide

This project follows the SciML style guide.

Key points:

• Use descriptive variable names

• Follow Julia naming conventions (snake_case for variables, CamelCase for types)

• Write docstrings for exported functions

• Keep lines under 92 characters where possible

• Use consistent indentation (4 spaces)

Documentation standards

All docstrings use the DocStringExtensions.jl @template conventions registered in src/docstrings.jl. That file is included near the top of the module, before any docstrings are defined, so the templates apply everywhere. The DocStringExtensions import itself lives in the module file (src/EpiAwareADTools.jl) rather than in docstrings.jl, because the kit's import-centralisation gate requires every import to sit in the module file.

Functions: Use $(TYPEDSIGNATURES) for automatic signature generation:

@doc "
$(TYPEDSIGNATURES)

Brief description of the function.

# Arguments
- `param1`: Description (no type annotations needed)
- `param2`: Description
"
function my_function(param1, param2; kwarg1 = default)
    # implementation
end

Key rules:

• Never use @doc raw" with DocStringExtensions macros, as it prevents expansion (reach for it only when the docstring carries LaTeX math and no macros)

• Don't repeat type information in argument descriptions, since $(TYPEDSIGNATURES) shows them

• Use @doc " (not @doc """) to allow macro expansion, unless the docstring carries LaTeX math

• Document argument purpose, not types

Benchmarks

The suite in benchmark/benchmarks.jl reads the cost of the AD-safe hooks against the bare Distributions.jl functions and times the per-backend gradients.

task benchmark                 # benchmark the current state
task benchmark-compare         # compare main vs current
task benchmark -- --filter=Hooks   # filter to specific benchmarks

Benchmark results recorded on main are rendered on the Benchmarks documentation page.

Code quality

Pre-commit checklist

Before submitting a pull request: 2. Run pre-commit checks (recommended):

task precommit

3. Or run individual checks:

task test        # Full test suite
task docs-fast   # Build documentation

Adding new features

2. Write tests first: Add a @testitem under test/unit/

3. Implement the fix: Add the method in src/, following Adding a workaround

4. Add an AD scenario: Register a gradient scenario in test/ADFixtures/

5. Document the feature: Add docstrings and a Charter row for the new entry

6. Test thoroughly: Run the full test suite

Getting help

• Questions: Open a GitHub discussion

• Bugs: File a GitHub issue with a minimal reproducible example

• Feature requests: Open a GitHub issue with rationale and use case

• General Julia help: See Julia Discourse or Julia Slack

Thank you for contributing to EpiAwareADTools.jl!