Migration plan to Rust (M4)

This document validates the migration lens on CodArch itself. The numbers come from codarch analyze . and change with the code; they are not a commitment, they are a snapshot for prioritising.

Current snapshot

Signal	Value
Files analysed	211
TS/JS	211 files · 33,621 rloc
Rust	0
WASM/Node	0
Bridges	2
FFI edges	0
Migration debt	5,871

Debt is (maxComplexity + 1) × (1 + fanIn) per TS/JS file. It does not count lines: a large, trivial file weighs less than a complex one with many consumers. Test fixtures no longer enter: this repo’s .codarch.yml excludes them with ignore (ADR 0038), so the snapshot measures only production code.

Priority by debt

Debt	RLOC	File
357	499	packages/render/src/model.ts
210	281	packages/parsers/src/migration.ts
195	682	packages/metrics/src/diff.ts
185	313	packages/parsers/src/analyze-core.ts
132	234	apps/cli/src/commands/compare.ts
126	2,588	apps/web/src/App.tsx
117	359	packages/schema/src/legacy.ts
112	732	apps/mcp/src/store.ts
108	237	packages/parsers/src/graph.ts
105	147	packages/versions/src/index.ts
104	256	packages/addons/src/index.ts
102	535	apps/cli/src/commands/serve.ts
90	1,910	apps/desktop/src/renderer/App.tsx
90	95	packages/metrics/src/report.ts
90	124	packages/parsers/src/walk.ts

Proposed phases

1. Pure contracts. schema/identity, schema/stats, schema/tree and the data structures. They are deterministic code, with no I/O or tree-sitter; the first Rust crate can be validated against the same fixtures and checksums.
2. Graph and metrics. graph, diff, risk and rules. They are pure transformations over data; ideal for property tests and for comparing JSON byte by byte with the TS implementation.
3. TS parser. parsers/walk, resolve and the tree-sitter parser. It is the riskiest phase: import resolution, grammars and metrics. Keeping a napi boundary with the current parser allows migrating file by file and comparing results.
4. Render and adapters. The render can stay in TS or move to WASM; the CLI and MCP are thin adapters. They are not the priority: they change more and do not concentrate the debt.
5. GitHub Action. YAML and render-comment.mjs; do not migrate. It consumes the published CLI.

Boundary

The natural strategy is a Rust crate with a stable API exposed through napi, and TS adapters that replace module by module. Each replacement must compare snapshots and checksums of the same commit. The migration lens measures the progress: rust goes up, migrationDebt goes down, and ffiCrossings says how much new coupling the bridge introduces.

Reproducing it

npm run analyze -- .
npm run inspect -- .codarch/snapshot.cc.json --metric migrationDebt --top 15
npm run render -- .codarch/snapshot.cc.json -o .codarch/sitio

Limits

• Rust is measured, not parsed: there are no Rust imports or complexity in v1.
• napi edges are not observable without the generated glue; only the bridge is marked.
• This repo’s test fixtures are excluded in .codarch.yml; they are not production debt.
• Debt prioritises; it does not estimate weeks or people.