Rust Compiler & Native CLI — Shipped

Status: shipped · Shipped: September 2026

The Rust compiler and native CLI are complete. Python is removed from the repository. This page is kept as the design record.

What shipped

• Oxc parser for .mx/.ts/.tsx — replaces tree-sitter; arena-allocated, spec-compliant, 3× faster than SWC
• lightningcss for CSS parsing — typed property values, browser-grade, parallel
• Full Rust CLI (morphc → cargo install morphc) — new, install, update, dev, build, run, check, doctor, cache, plus direct file morphing (morph <file> --to cpp|rust)
• Intent-based codegen via crates/morpher — escape analysis, type widening, unique_ptr/shared_ptr decisions at compile time, no GC
• Python removed — morph/, pyproject.toml, tests/unit, tests/integration, PyPI workflow all deleted September 2026

Final crate layout

morph/
├── crates/
│   ├── morphc/            # Native CLI binary → `morph`
│   ├── morpher/           # TS→C++/Rust translator (Oxc + escape analysis)
│   ├── morph-parser/      # .mx parsing (Oxc + lightningcss)
│   ├── morph-ir/          # Intermediate Representation
│   ├── morph-codegen/     # C++ / Rust code emission
│   ├── morph-build/       # Compilation, dev rt + IPC, packaging
│   ├── morph-cache/       # Global cache, runtime downloads
│   └── morph-config/      # Config + lock file parsing

What stayed identical for users

• .mx files, CSS, Tailwind, morphState/morphEffect, fetch()
• morph dev, morph build, morph run, morph check
• The JSON IR wire format
• Dev-mode hot reload

Original design notes

Why Python was used first

Morph didn't start with Rust because the goal at the start wasn't a fast compiler. The goal was to find out whether the concept even works — and Python was the right tool for that experiment:

• Validate the idea first. The core question was "can we parse .mx → IR → render natively?" Python answered that in days.
• No recompile on every change. The pipeline was changing daily. In Python, save → run → see is instant.
• Better tools for text data. Python's dicts/lists and standard library were purpose-built for shuffling structured text.

SWC vs Oxc (the parser choice)

Morph chose Oxc: 3–4× faster than SWC parsing, arena allocator, spec-compliant, powers Rolldown (Vite's future bundler), and the oxlint ecosystem provides a natural home for morph check.

Phases as shipped

Phase	What
1	Parser swap: Oxc + lightningcss producing MxSource
2	Walker + IR builder port
3	Codegen port (node emitter, logic emitter, feature set)
4	Build system + dev mode (CMake devrt, loopback TCP IPC)
5	CLI finalization + crates.io publishing
6	Python deletion — all Python files, tests, and workflows removed