# Development Commands
Development commands and validation steps for the repository.
Agents must run appropriate checks after modifying code.
---
## Core Workflow
Typical development loop:
```bash
just check
just fix
just test
```
These commands ensure:
- formatting
- linting
- static analysis
- tests
## Justfile Usage
This repository standardizes development tasks through the `justfile`.
Agents should **prefer running `just` commands instead of invoking the underlying tools directly**. The justfile ensures the correct flags, configuration, and
tool ordering are used.
Examples:
- prefer `just fix` instead of running `cargo fmt` directly
- prefer `just check` instead of running `cargo clippy` directly
- prefer `just ci` instead of manually running multiple validation steps
Direct tool invocation should only be used when a corresponding `just` command does not exist.
Rust unit, integration, CLI, slow, example, and release test recipes run with `cargo nextest`. Documentation tests intentionally remain on `cargo test --doc`
because nextest does not run rustdoc doctests.
---
## Formatting
Rust formatting:
```bash
cargo fmt
```
Typically run through:
```bash
just fix
```
Formatting must always be applied before committing changes.
---
## Linting
Lint checks include:
```bash
cargo clippy
semgrep
```
Warnings are treated as errors in CI.
Run via:
```bash
just check
```
---
## Documentation Validation
Documentation must build successfully.
Verify with:
```bash
just doc-check
```
---
## Full CI Validation
Before large changes, run the full CI command:
```bash
just ci
```
This runs:
- formatting checks
- lint checks
- repository-owned Semgrep rules
- unit tests
- integration tests
- documentation builds
- validated example runs
- benchmark compilation
For heavier stabilization work, run the slow-test wrapper:
```bash
just ci-slow
```
This runs the normal CI command and then feature-gated slow/stress tests through the repository `perf` Cargo profile. The slow suite includes large-scale
toroidal debug probes, so optimized builds are part of the validation contract rather than an optional speed tweak.
## Semgrep
Repository-owned Semgrep rules live in `semgrep.yaml`. They encode focused project invariants that are not already covered by Rust, Clippy, Ruff, or ShellCheck.
When adding or changing a Semgrep rule, add a matching fixture under `tests/semgrep/` and keep `just semgrep-test` passing. Rules ported from
`markov-chain-monte-carlo` must be adapted to CDT naming, paths, and architecture constraints rather than copied mechanically.
Commands:
```bash
just semgrep # Run repository-owned rules
just semgrep-test # Verify Semgrep rule fixtures
```
## Coverage
Coverage is generated with `cargo-llvm-cov`, matching the Codecov workflow.
Commands:
```bash
just coverage # HTML report at target/llvm-cov/html/index.html
just coverage-ci # Cobertura XML at coverage/cobertura.xml
```
---
## Examples
Example programs and scripts live in:
```text
examples/
examples/scripts/
```
Validate with:
```bash
just examples
just examples-validate
```
Examples must:
- compile
- run successfully
- demonstrate correct API usage
`just examples-validate` additionally checks stable output markers for user-facing Cargo examples. Keep those markers semantic rather than exact numeric values
so simulation output can evolve without making the example contract brittle.
The example runner compiles all Cargo examples once with `cargo build --release --examples`, then executes the compiled binaries directly. This preserves
example coverage while avoiding repeated Cargo invocations for each example.
When adding or renaming a Cargo example, update `scripts/run_all_examples.sh` `validate_example_output()` with stable semantic output markers, or intentionally
document why success-only validation is sufficient for that example.
---
## Spell Checking
Documentation and comments are spell‑checked.
Run:
```bash
just spell-check
```
If a legitimate technical word fails, add it to `typos.toml` under:
```toml
[default.extend-words]
```
---
## TOML Formatting
TOML files should be validated and formatted using Taplo.
Commands:
```bash
just toml-check # Non-mutating formatting and lint checks
just toml-fix # Apply formatting fixes
```
Compatibility aliases remain available as granular recipes:
`just toml-lint`, `just toml-fmt-check`, and `just toml-fmt`.
---
## Markdown Formatting
Markdown files are checked and fixed with `rumdl`. The repository uses a
160-column Markdown width, and `just markdown-check` also runs a raw line-length
guard so constructs that `rumdl` exempts from MD013 still respect that limit.
Commands:
```bash
just markdown-check # Non-mutating check
just markdown-fix # Apply fixes
```
---
## Shell Script Validation
Shell scripts must pass:
```text
shfmt
shellcheck
```
Commands:
```bash
just shell-check # Lint (non-mutating)
just shell-fix # Format (mutating)
```
`just shell-fmt` remains as a compatibility alias for the formatter.
---
## YAML Validation
YAML and `CITATION.cff` files are checked with `yamllint` and formatted with
`dprint` using the Rust-native `pretty_yaml` plugin.
Commands:
```bash
just yaml-check # Non-mutating formatting and lint checks
just yaml-fix # Format (mutating)
```
Compatibility aliases remain available as granular recipes:
`just yaml-lint` and `just yaml-fmt-check`.
---
## CITATION.cff Validation
Citation metadata should pass both YAML style linting and CFF schema validation.
Run:
```bash
just citation-check
```
---
## JSON Validation
JSON files should be validated after edits.
```bash
just validate-json
```
Or directly:
```bash
jq empty file.json
```
---
## GitHub Actions Validation
Workflows must pass `actionlint`.
The repository has separate workflows for full CI, dependency audit, Codecov coverage, repository-rule SARIF upload, Clippy SARIF, performance checks, and
CodeQL analysis. Do not add another external analysis workflow unless it has a distinct signal and required secrets are configured for this repository.
External GitHub Actions must use full commit-SHA pins, stay within the
repository-owned allowlist in `semgrep.yaml`, and keep a readable version
comment next to each pin. Dependabot remains configured for the
`github-actions` ecosystem; its update PRs should preserve both the SHA pin and
the adjacent human-readable version comment.
Run with:
```bash
just action-lint
```
GitHub Actions security analysis runs with `zizmor` locally and in `.github/workflows/zizmor.yml`:
```bash
just zizmor
```
---
## Python Validation
Python scripts are linted and type-checked:
```bash
just python-lint # ruff format + ruff check
just python-fix # ruff check --fix + ruff format
just python-typecheck # ty check (blocking)
just test-python # pytest
```
## Benchmark And Release Hygiene
Benchmark harnesses can be smoke-tested without producing baseline-quality performance data:
```bash
just bench-smoke
```
The smaller CI regression benchmark contract runs through the `perf` Cargo profile:
```bash
just bench-ci
```
For manual large-scale toroidal 1+1 CDT move-kernel debugging, run the parameterized slow-test harness through the `perf` Cargo profile:
```bash
just debug-large-scale-1p1 512 16 10
```
Named scale probes are also available:
```bash
just debug-large-scale-1p1-512
just debug-large-scale-1p1-1024
```
The arguments are total vertices, timeslices, and sweeps. The umbrella recipe runs the curated large-scale debug case set:
```bash
just perf-large-scale-debug
```
The umbrella recipe currently runs `512/16/10` and `1024/32/1` cases. It is intentionally manual and long-running; use the parameterized
`debug-large-scale-1p1` recipe directly when you only need a smaller smoke probe.
These debug recipes do not enable volume fixing. Volume drift in the reported final vertex and simplex counts is expected when `(1,3)` and `(3,1)` moves are
accepted. Treat these runs as unfixed-volume Metropolis debug sweeps, not as fixed-volume CDT production ensembles. Future fixed-volume recipes should be
explicitly labeled because quadratic volume fixing samples a modified ensemble; see Ambjørn et al.,
[The Semiclassical Limit of Causal Dynamical Triangulations](https://arxiv.org/abs/1102.3929), and
[The phase structure of Causal Dynamical Triangulations with toroidal spatial topology](https://arxiv.org/abs/1802.10434).
For unfixed-volume Metropolis debug runs, each sweep is sized from the current number of top-dimensional simplices at the start of that sweep, then executed as
one checkpoint-preserving Metropolis chunk. The cosmological constant is the volume-control coupling. If a large-scale recipe grows or shrinks too aggressively,
tune the action parameters rather than interpreting the run as a failed fixed-volume simulation.
The recipes set per-case environment variables, and the Rust harness also accepts these variables directly:
| `CDT_LARGE_DEBUG_VERTICES` / `CDT_LARGE_DEBUG_VERTICES_1P1` | Total vertex count |
| `CDT_LARGE_DEBUG_TIMESLICES` / `CDT_LARGE_DEBUG_TIMESLICES_1P1` | Number of periodic time slices |
| `CDT_LARGE_DEBUG_SWEEPS` / `CDT_LARGE_DEBUG_SWEEPS_1P1` | Sweep count; one sweep attempts one move per current simplex |
| `CDT_LARGE_DEBUG_SEED` / `CDT_LARGE_DEBUG_SEED_1P1` | RNG seed, decimal or `0x` hexadecimal |
| `CDT_LARGE_DEBUG_MAX_RUNTIME_SECS` | Optional wall-clock cap checked between sweeps; `0` disables it |
To compile benchmarks and release-profile integration tests without running them:
```bash
just bench-test-compile
```
Before release preparation, optional Cargo hygiene checks are available:
```bash
just unused-deps
just publish-check
```
---
## Recommended Command Matrix
| Run lints | `just check` |
| Format code | `just fix` |
| Run unit tests | `just test` |
| Run integration tests | `just test-integration` |
| Run slow tests | `just test-slow` |
| Run all tests | `just test-all` |
| Run Python tests | `just test-python` |
| Run examples | `just examples` |
| Validate examples | `just examples-validate` |
| Run full CI | `just ci` |
| Pre-commit check | `just commit-check` |
---
## Testing by File Type
| `tests/` | `just test-integration` (or `just ci`) |
| `examples/` | `just examples-validate` |
| `benches/` | `just bench-compile` |
| `src/` | `just test` |
| `scripts/` | `just test-python` |
| Any Rust | `just doc-check` |
---
## CI Expectations
CI enforces:
- formatting
- clippy lints
- documentation build
- tests
- validated examples
All warnings are treated as errors.
Agents must ensure changes pass CI locally before proposing patches.
---
## Changelog
The changelog is **auto-generated**.
Never edit manually.
Regenerate with:
```bash
just changelog
```
This runs `git-cliff`, applies the Python postprocessor, archives completed minor release series under `docs/archive/changelog/`, and
formats generated changelog files with `rumdl`.
For release PRs, generate the changelog for a version before the final tag exists with:
```bash
just changelog-unreleased v0.1.0
```
Create annotated release tags from the generated changelog after the release PR is merged with:
```bash
just tag v0.1.0
```