BMB - Bare-Metal-Banter

A contract-verified systems programming language.

Hard to write. Hard to get wrong. And that's what AI prefers.

Why BMB?

Traditional languages prioritize human convenience—readable syntax, flexible conventions, implicit behaviors. The cost: ambiguity, guesswork, runtime surprises.

BMB takes a different approach. Contracts are mandatory. Specifications are explicit. Invariants are verified at compile time.

fn binary_search(arr: &[i64], target: i64) -> i64
  pre is_sorted(arr)
  post ret == -1 || (0 <= ret && ret < len(arr))
  post ret != -1 implies arr[ret] == target
= {
    var lo = 0;
    var hi = len(arr) - 1;
    while lo <= hi
      invariant 0 <= lo && hi < len(arr)
    {
        let mid = lo + (hi - lo) / 2;
        if arr[mid] == target { mid }
        else if arr[mid] < target { lo = mid + 1; }
        else { hi = mid - 1; }
    };
    -1
};

Priorities

Priority	Principle
P0	Performance — No syntax that constrains optimization. Target: exceed C/Rust.
P0	Correctness — If it can be verified at compile time, it must be.

Quick Start

# Build
cargo build --release

# Run
bmb run examples/hello.bmb

# Type check
bmb check examples/simple.bmb

# Contract verification (requires Z3)
bmb verify examples/contracts.bmb

# Native compile (requires LLVM)
bmb build examples/hello.bmb -o hello

# REPL
bmb repl

Current Status: v0.46 (Independence)

Category	Status
Language Core	✅ Complete
Type System	✅ Complete
Contract System	✅ Complete
Bootstrap Compiler	✅ 30K LOC
Test Suite	✅ 1,753+ tests
Documentation	✅ Complete
CI/CD	✅ Complete
Performance	✅ 0.89x-0.99x vs C
Self-Compile	✅ 0.56s
v1.0.0-beta	🎯 Target

Features

Completed

• Types: i8-i128, u8-u128, f64, bool, char, String
• Generics: <T>, <K, V>, bounds, where clauses
• Contracts: pre, post, invariant, where, pure, @trust
• Control Flow: if-else, match, while, for-in, loop
• Operators: Arithmetic, overflow-safe (+%, +|, +?), bitwise (band, bor), shift (<<, >>)
• Collections: Vec, Box, HashMap (stdlib)
• Tooling: Package manager (gotgan), VS Code, formatter, LSP

In Progress

• 3-Stage self-hosting verification (WSL)
• Performance Gate #3.2, #3.3 (Benchmarks Game, Contract optimization)
• Ecosystem packages (14+ target, 12 complete)

Project Structure

lang-bmb/
├── bmb/           # Rust compiler (being replaced)
├── bootstrap/     # Self-hosted BMB compiler (30K LOC)
├── stdlib/        # Standard library
├── examples/      # Example programs
├── ecosystem/     # Tools & extensions
│   ├── gotgan/           # Package manager
│   ├── vscode-bmb/       # VS Code extension
│   ├── tree-sitter-bmb/  # Syntax highlighting
│   ├── playground/       # Online editor
│   └── benchmark-bmb/    # Performance suite
└── docs/          # Documentation

Requirements

Requirement	Purpose	Required
Rust 1.70+	Build compiler	Yes (until v0.45)
LLVM 21+	Native codegen	Optional
Z3	Contract verification	Optional

Documentation

Document	Description
SPECIFICATION.md	Language specification
LANGUAGE_REFERENCE.md	Complete reference
ARCHITECTURE.md	Compiler internals
ROADMAP.md	Development roadmap
API_STABILITY.md	API guarantees
BENCHMARK_COMPARISON.md	C/Rust/BMB performance
tutorials/	Getting started guides

Roadmap to v1.0.0-beta

v0.46 (Current) ─── Independence (3-Stage verification)
     │
v0.47 ─────────── Performance Gates (C parity verified)
     │
v0.48 ─────────── Ecosystem (14+ core packages)
     │
v0.49 ─────────── Samples & scenarios
     │
v0.50 ─────────── Final verification
     │
v1.0.0-beta ───── Complete programming language ★

See ROADMAP.md for detailed phases.

Performance

BMB matches or exceeds C/Rust performance on compute-intensive workloads:

                     C         Rust      BMB       Winner
─────────────────────────────────────────────────────────
fibonacci(45)        1.65s     1.66s     1.63s     ★ BMB (0.99x)
fibonacci(40)        177ms     180ms     150ms     ★ BMB (0.85x)
mandelbrot           42ms      42ms      39ms      ★ BMB (0.93x)
spectral_norm        44ms      44ms      39ms      ★ BMB (0.89x)
self-compile         -         -         0.56s     ✅ (30K LOC)

See BENCHMARK_COMPARISON.md for detailed methodology and results.

License

MIT