nu_plugin_kcl/adr-001-kcl-cli-wrapper-architecture.md

# ADR-001: KCL Plugin CLI Wrapper Architecture\n\n## Status\n\n**Accepted** - 2025-12-15\n\n## Context\n\nThe nu_plugin_kcl project provides Nushell integration for KCL (KittyCAD Language). The core decision was whether to implement this as:\n\n1. **Pure Rust Implementation** (using `kcl-lib` crate directly)\n2. **CLI Wrapper** (using `Command::new("kcl")` to invoke external binary)\n\n### Technical Constraints\n\nKCL is a **configuration language with module system**:\n\n- Imports: `import provisioning.lib as lib`\n- Module resolution: `kcl.mod` files (similar to Go)\n- Standard library with external dependencies\n- Complex build context and validation\n\n### User Requirements\n\nAll KCL files in provisioning project use module system:\n\n```kcl\n# workspace_librecloud/infra/sgoyol/settings.k\nimport provisioning.lib as lib\nimport provisioning.settings as cfg\n\n# Requires full module resolution\n```\n\nThis means any KCL evaluation must:\n\n- ✅ Read and parse `kcl.mod` files\n- ✅ Resolve import paths (`provisioning.lib` → file path)\n- ✅ Access standard library\n- ✅ Handle all validation and compilation\n\n## Decision\n\nImplement nu_plugin_kcl as a **CLI Wrapper** that invokes the external `kcl` binary.\n\n### Architecture\n\n```plaintext\nNushell Script\n    ↓\nkcl-run (plugin command)\n    ↓\nhelpers.rs: run_kcl_command()\n    ↓\nstd::process::Command::new("kcl")\n    ↓\nKCL CLI (official binary)\n    ↓\nModule Resolution (guaranteed correct)\n    ↓\nJSON Output\n    ↓\nPlugin: serde_json::Value → nu_protocol::Value\n    ↓\nNushell Records/Lists\n```\n\n### Implementation Details\n\n**Core Functions** (`helpers.rs`):\n\n```rust\npub(crate) fn run_kcl_command(\n    file: &str,\n    format: &str,\n    output: Option<&str>,\n    defines: Option<&str>,\n) -> Result<String>\n```\n\n**Plugin Commands** (`main.rs`):\n\n1. `kcl-run` - Execute KCL files (input/output JSON, YAML)\n2. `kcl-eval` - Evaluate KCL with automatic caching\n3. `kcl-format` - Format KCL files\n4. `kcl-validate` - Validate KCL files/directories\n5. `kcl-cache-status` - Show cache information\n\n**Output Processing**:\n\n- Invokes: `kcl run /file.k --format json`\n- Captures: stdout (JSON string)\n- Parses: serde_json::Value\n- Converts: `json_value_to_nu_value()` recursive function\n- Returns: nu_protocol::Value (records/lists, not strings)\n\n**Caching** (non-blocking, graceful degradation):\n\n- SHA256 content-addressed cache\n- Location: `~/.cache/provisioning/config-cache/`\n- Key: SHA256(file_content + format + context)\n- Hit rate: Expected 80-90% in typical workflows\n\n### Type System\n\nCommand signatures declare `Type::Any` output:\n\n```rust\n.input_output_type(Type::Any, Type::Any)\n```\n\nThis allows:\n\n- Plugin returns: nu_protocol::Value::Record\n- Nushell receives: proper record (not string)\n- Cell path access works: `kcl-run /file.k | .servers | length`\n\n## Rationale\n\n### Why CLI Wrapper Over Pure Rust\n\n| Aspect | Pure Rust (kcl-lib) | CLI Wrapper (chosen) |\n|--------|-------------------|----------------------|\n| **Module resolution** | ❌ Unclear/Manual | ✅ Works automatically |\n| **kcl.mod support** | ❓ Undocumented | ✅ Guaranteed |\n| **Import resolution** | ❌ Need manual impl | ✅ Built-in |\n| **Standard library** | ❓ How to access? | ✅ Automatic |\n| **Maintenance** | ❌ Track CLI changes | ✅ No maintenance |\n| **Error handling** | ❌ Different from CLI | ✅ Same as CLI |\n| **Complexity** | 🔴 High | 🟢 Low |\n| **External CLI** | ✅ None needed | ✅ Requires kcl binary |\n\n### Why Not Pure Rust\n\nUsing `kcl-lib` directly would require:\n\n1. **Implement module resolution** from scratch:\n\n   ```rust\n   // Parse kcl.mod\n   let mod_data = fs::read("kcl.mod")?;\n   let config = parse_kcl_mod(&mod_data)?;\n\n   // Resolve each import\n   for import in file.imports {\n       let path = resolve_module_path(&import, &config)?;\n   }\n   ```\n\n2. **Handle standard library discovery**:\n\n   ```rust\n   let stdlib_path = find_kcl_stdlib()?;\n   // What version? Is it installed? Where?\n   ```\n\n3. **Replicate all CLI logic**:\n   - Error handling and messages\n   - Validation rules\n   - Module caching\n   - Build context setup\n   - Runtime options handling\n\nThis is essentially **re-implementing the KCL CLI** in Rust, which is:\n\n- ❌ Massive undertaking\n- ❌ Error-prone\n- ❌ High maintenance burden\n- ❌ Duplicates official implementation\n\n### Single Source of Truth\n\nDelegating to the CLI ensures:\n\n- ✅ Official implementation handles all edge cases\n- ✅ KCL updates automatically available\n- ✅ No maintenance as language evolves\n- ✅ Guaranteed correctness\n\n## Consequences\n\n### Positive\n\n- **Correctness**: Module resolution guaranteed correct by official CLI\n- **Simplicity**: No need to re-implement language internals\n- **Maintenance**: Updates to KCL automatically available\n- **Features**: All CLI features automatically supported\n- **Compatibility**: Works with all KCL versions\n- **Reliability**: Single point of truth (official implementation)\n- **Error Messages**: Same error handling as CLI users expect\n\n### Negative\n\n- **External Dependency**: Requires `kcl` binary in PATH\n- **Performance Overhead**: Process fork (~100-200ms vs direct call)\n- **Process Management**: Spawns subprocess for each execution\n- **Error Output**: Subprocess stderr handling required\n\n### Mitigations\n\n**For External Dependency**:\n\n- Clear documentation: setup guide with kcl installation\n- Error messages: helpful if `kcl` not found\n- Distribution: kcl included in provisioning distributions\n\n**For Performance Overhead**:\n\n- Caching: 80-90% hit rate in typical workflows\n- Cache hits: ~1-5ms (not 100-200ms)\n- Lazy evaluation: Only runs when needed\n\n## Alternatives Considered\n\n### Alternative 1: Pure Rust with kcl-lib\n\n**Rejected**: Module resolution undefined, high maintenance cost\n\n### Alternative 2: Pure Rust with manual module implementation\n\n**Rejected**: Duplicates official CLI, maintenance nightmare\n\n### Alternative 3: Hybrid (pure Rust + CLI fallback)\n\n**Rejected**: Adds complexity, two implementations to maintain\n\n### Alternative 4: WebAssembly (kcl compiled to WASM)\n\n**Rejected**: KCL WASM support unclear, adds complexity\n\n## Implementation Status\n\n### Completed\n\n- ✅ Plugin command infrastructure (5 commands)\n- ✅ CLI invocation via `Command::new("kcl")`\n- ✅ JSON output parsing (serde_json → nu_protocol)\n- ✅ Recursive value conversion (records, lists, primitives)\n- ✅ Caching system (SHA256, filesystem-based)\n- ✅ Error handling (CLI errors → Nushell errors)\n- ✅ Type system (Type::Any for proper output types)\n\n### Files\n\n- `src/main.rs` - Plugin commands and JSON parsing (95 lines of logic)\n- `src/helpers.rs` - CLI invocation and caching (300+ lines)\n- `tests/` - Test suite for all commands\n\n## Testing\n\n**Manual Testing**:\n\n```bash\n# Test basic execution\nkcl-run /path/to/file.k -f json\n\n# Test with imports\nkcl-run /workspace_librecloud/infra/sgoyol/settings.k -f json | .servers\n\n# Test cache\nkcl-cache-status\n```\n\n**Verification**:\n\n- ✅ Module imports resolve correctly\n- ✅ Output is proper records (not strings)\n- ✅ Cell path access works\n- ✅ Cache hits are fast\n- ✅ Error messages are helpful\n\n## References\n\n- [KCL Official Documentation](https://kcl-lang.io/)\n- [kcl-lib Crate](https://crates.io/crates/kcl-lib/)\n- [Module System Design](./MODULE_SYSTEM.md)\n- [Caching Strategy](./CACHING.md)\n- [JSON Output Format](./OUTPUT_FORMAT.md)\n\n---\n\n**Author**: Architecture Team\n**Date**: 2025-12-15\n**Decision Made By**: Technical Review