CONTRIBUTING.md

Rama main · réplica de lectura del repositorio original.

# Contributing to Sounio

Thank you for your interest in contributing to Sounio! Sounio is a self-hosted, safety-critical systems and scientific programming language for epistemic computing. This document provides guidelines and instructions for contributing to the compiler, standard library, and documentation.

## Code of Conduct

Be respectful. Be constructive. Be patient. Sounio is an academic and active research project with high-fidelity, safety-critical standards. We are building software that communicates the quality of its own knowledge, and we carry that ethos of precision and integrity into our community.

---

## Getting Started

### Prerequisites

Sounio compiles itself (self-hosted). It does not rely on Cargo or Rust for production builds. To build Sounio from source, you need a **Linux x86_64** environment with standard Unix tools:

- `gcc` or `clang` — standard C compiler (required only for Stage 0 bootstrap)
- `make` — build automation utility
- `bash` — standard shell shell
- `sha256sum` — to verify the bootstrap fixed point

### Building from Source (The Bootstrap Ceremony)

The Sounio compiler is written entirely in Sounio. To build the compiler from a clean checkout, we run a multi-stage bootstrap ceremony that compiles a Stage 0 C compiler, compiles successive generations of the Sounio compiler, and verifies that the compiler has reached a stable fixed point (where Stage N and Stage N+1 produce bit-identical ELF binaries):

```bash
# Clone the repository
git clone https://github.com/Sounio-lang/sounio.git
cd sounio

# Execute the self-hosted bootstrap chain and fixed-point verification
make build

# Verify the compiler version
./bin/souc --version
```

### Running the Test Suite

```bash
# Run the entire standard test suite
bash scripts/run_sio_test_suite.sh

# Run a specific subset of tests matching a pattern (e.g. vancomycin)
bash scripts/run_sio_test_suite.sh vancomycin --verbose
```

### Generated snapshots: do not commit them in your PR

`docs/governance/topic-registry.v1.json`, `docs/governance/DOCS_AUTHORITY_MATRIX.md`
and `datasets/sounio-code-examples/{train,validation}.jsonl,manifest.json` are
functions of the whole tree. Committing them from feature PRs made any two PRs
conflict, so CI's *Generated snapshot guard* refuses them in every PR except a
`snapshot/refresh-*` branch. The docs checkers validate against the registry
rebuilt in memory, so a stale snapshot hides nothing.

- `node scripts/docs/sync_governance_metadata.mjs` updates the metadata headers in
  your documents (commit those) and leaves the snapshot alone.
- To refresh the snapshot, open a PR from `snapshot/refresh-<date>` that runs
  `node scripts/docs/sync_governance_metadata.mjs --snapshot` and
  `python3 scripts/dev/export_hf_dataset.py`.

### Local Git Setup (one-time)

Some tracked files are generated — notably the `docs/governance/` metadata
produced by `scripts/docs/sync_governance_metadata.mjs`. A textual 3-way merge
of these is meaningless and conflicts on nearly every merge from an active
`main`. After cloning, install the repo's git merge driver once so they
auto-resolve:

```bash
bash scripts/dev/install-git-merge-drivers.sh
```

This registers the `governance-regen` merge driver plus a `post-merge` hook
that regenerate the generated `docs/governance/` artifacts from the merged doc
set instead of conflicting on them. It is opt-in per clone (git keeps
merge-driver commands in local config, which is not committed) and will not
overwrite a pre-existing `post-merge` hook.

---

## Development Workflow

### 1. Create a Branch

Always create a descriptive branch for your work:

```bash
git checkout -b feature/your-feature-name
```

Branch naming conventions:
- `feature/` — New compiler features or stdlib modules
- `fix/` — Compiler or runtime bug fixes
- `docs/` — Documentation translations or expansions
- `refactor/` — Code refactoring or cleanup
- `test/` — Enhancing test coverage or adding compiler compile-fail gates

### 2. Sounio Coding Guidelines (Crucial)

Sounio is **NOT Rust**. The compiler will reject common Rust syntax. Follow these rules strictly:

1. **Mutability**: Use `var` for mutable bindings, never `let mut`.
2. **Exclusive References**: Use `&!` for exclusive references, never `&mut`.
3. **No Rust Macros**: Use `println("text")` and `assert(cond)` instead of macro syntax.
4. **No Closure Literals**: Closure syntax like `|x| x + 1` does not exist. Use named function references.
5. **Explicit Self**: All struct method implementations must declare `self` explicitly as `self: &Type` or `self: &!Type`.
6. **Algebraic Effects System**: All functions with side effects must declare them using the `with` keyword:
   - `IO`: printing, file access, terminal operations
   - `Mut`: exclusive reference mutation or reassignment
   - `Div`: mathematical division or modulo operations
   - `Panic`: array indexing, assertions, or type casting
   - `Alloc`: heap memory allocation

Example of correct Sounio:
```sio
struct Tracker {
    value: f64
}

impl Tracker {
    fn update(self: &!Tracker, input: f64) with Mut {
        self.value = input
    }
}

fn process_measurement(val: f64) -> Knowledge[f64] with Div, Panic {
    let base = Knowledge { value: val, epsilon: 0.02 }
    base * 1.5
}
```

### 3. Testing and Verification

Before submitting any Pull Request:
- Type-check your Sounio files with `./bin/souc check file.sio`
- Compile and run your Sounio files with `./bin/souc run file.sio`
- Ensure all automated checks pass: `bash scripts/run_sio_test_suite.sh`
- Ensure documentation registry is aligned: `node scripts/docs/check_docs_registry.mjs` (sync document headers with `node scripts/docs/sync_governance_metadata.mjs` if needed; do not commit the regenerated snapshot)

The repository's default compiler surface is `bin/souc` as the official entrypoint (Madaros by default).

### 4. Commit Message Guidelines

We follow a strict semantic and component-scoped commit convention. Commit messages must be structured as:

```
[component] Brief description

- Bulleted list of changes and why they were made
```

**Allowed Components**: `lexer`, `parser`, `ast`, `check`, `types`, `effects`, `hir`, `sir`, `hlir`, `codegen`, `native`, `cli`, `docs`, `stdlib`, `tests`, `epistemic`, `units`, `website`.

*Example*:
```
[check] Reject variable assignment when confidence threshold is breached

- Adds compile-time constraint validation during bidirectional type checking
- Throws TypeMismatch error when Knowledge[T, epsilon] violates required threshold
```

### 5. Choosing and Claiming Work

Work is planned on a project board (Notion; the maintainer grants access on request). The board mirrors three files in this repository, which remain the source of truth:

- [`docs/planning/LANGUAGE_BACKLOG_2026-10-22.md`](docs/planning/LANGUAGE_BACKLOG_2026-10-22.md) — what is open, in priority order: **silent wrong values first**, then refusals without a location, then decisions owed, then capacity, then research.
- [`docs/decisions/PENDING.md`](docs/decisions/PENDING.md) — decisions only the maintainer can take. Do not implement a choice that is still listed there.
- [`docs/decisions/`](docs/decisions/README.md) — decisions already taken (ADRs). Each implementation row of an ADR is one board task with its own acceptance test.

Each card carries these fields:

- **Agent**: who owns it. One of Claude, Codex, Cursor, Warp, Kimi, Grok PM, Nightly Audit, or Human. A card with Agent = Human is a decision that only the maintainer can take.
- **Train**: the release it ships in, `3.0`, `3.1` or `Backlog` (see [ADR-018](docs/decisions/adr-018-release-cycle-and-versioning.md)).
- **Epic**: the parent EPIC card. Themes such as Bootstrap gen7, Marco 0 or the AI track are EPIC cards, not fields.
- **Acceptance**: the criterion that makes the card *Done*.

`Milestone` and `Bot` are deprecated; do not set them on new cards.

Rules for every contributor, human or automated:

1. **Claim before you start.** Take a task in *Not started* whose blocking tasks are all *Done*. Set **Agent** to yourself, move the card to *In progress*, and record the same claim with `bin/sounio-coord claim`. A card may be *In progress* only when Agent is set. Do not work on a task someone else has claimed. Before pushing to a branch that had a commit in the last 24 hours, message its author on the coord bus.
2. **Every task has an acceptance criterion** (a probe, a test, a byte-identical hash, a proof). A task is *Done* only when that criterion holds and the PR is merged.
3. **Blocked?** Move the task to *Blocked* and say why on the task.
4. **Open the PR, link it on the task, move the task to *In review*.** A draft PR keeps the task *In progress*; it moves to *In review* when the PR leaves draft. When the PR merges, the task is *Done*.
5. **Fix what you find.** A defect found while working is fixed in the same change, or recorded as a new board task. Please do not open GitHub issues for it.
6. **Decisions are not taken in PRs.** If your change needs a semantic choice, stop and add it to `PENDING.md` with the question, the options and a discriminating test.

### 6. Numbering ADRs

ADR numbers are append-only and assigned at merge time. Before adding `adr-NNN-*.md`, check both `main` and the open pull requests for the next free number; the index in [`docs/decisions/README.md`](docs/decisions/README.md) lists numbers reserved by open PRs.

### 7. Changelog fragments and release trains

Do not edit `CHANGELOG.md`. If your PR touches the compiler, runtime, stdlib or CI scripts, add `changelog.d/<slug>.<kind>.md`, a single paragraph written for release notes. Use `kind=internal` when nothing is user-visible. Kinds and examples are in [`changelog.d/README.md`](changelog.d/README.md). The Impact job checks this.

Choose the kind carefully: it decides which release can carry your change. A change that makes existing programs fail to compile, or changes what they do, is `breaking` and waits for the next major. A change that makes a program refuse instead of printing a wrong number is `refusal` and may ship in a minor. A change to a printed number is `numerical` or `numerical-correction`, judged by the epistemic consistency test of [ADR-018](docs/decisions/adr-018-release-cycle-and-versioning.md) §3.

Releases are feature trains with a freeze date (see [`docs/RELEASE_POLICY.md`](docs/RELEASE_POLICY.md)). After the freeze, a fix reaches the release branch only if it carries the label `backport:X.Y`.

---

## Standard Library Structure

The Sounio standard library (`stdlib/`) is organized by domain and is the core surface for scientific workflows:

| Domain Module | Purpose |
| :--- | :--- |
| `stdlib/epistemic/` | `Knowledge<T>` uncertainty container, GUM propagation, and provenance tracking. |
| `stdlib/units/` | Dimensional analysis, physical quantities, and `VAR_UNIT_DIM` check. |
| `stdlib/clinical/` | PK/PD modeling, clinical trial analytics, and drug dosing algorithms. |
| `stdlib/math/` | Special functions, fractional calculus, autograd, and linear algebra. |
| `stdlib/gpu/` | PTX and GPU-accelerated computing backends. |
| `stdlib/ontology/` | Semantic ontology stores, subclass relations, and clinical terminologies (LOINC/SNOMED). |

---

## License

By contributing to Sounio, you agree that your contributions will be licensed under the **Apache License, Version 2.0** (with copyright assigned to Sounio Language Project).

Thank you for contributing to the future of epistemic computing! 🏛️