Carbon

Purpose

Guide agents through the Carbon programming language: syntax fundamentals, bidirectional C++ interoperability via Carbon.h, building from source with carbon-toolchain, the experimental memory safety model (checked borrows), current pre-1.0 limitations, and when to evaluate Carbon versus staying on C++.

When to Use

• Evaluating Carbon for a greenfield project with heavy C++ dependencies
• Calling C++ libraries from Carbon or exposing Carbon to C++
• Experimenting with Carbon toolchain from GitHub source
• Understanding Carbon's migration path from C++ codebases
• Assessing readiness for production (currently experimental)
• Comparing Carbon ergonomics to modern C++ (C++20/23)

Workflow

1. Project status awareness

Carbon (2026 state)
├── Experimental — no 1.0 release
├── Focus: C++ interoperability and migration
├── No production stdlib equivalent to C++ yet
└── Toolchain under active development

Use Carbon for exploration and migration prototyping, not production systems without team acceptance of instability.

2. Build carbon-toolchain

git clone https://github.com/carbon-language/carbon-lang
cd carbon-lang

# Prerequisites: LLVM, Clang, CMake, Ninja
./scripts/run_bazelisk.py build \
  //toolchain:install \
  --symlink_prefix=carbon/

# Add to PATH
export PATH="$PWD/carbon/bin:$PATH"

carbon --version

# Online explorer (Compiler Explorer instance)
# http://carbon.compiler-explorer.com/

3. Basic syntax

package Sample api;

fn Add(a: i32, b: i32) -> i32 {
    return a + b;
}

fn Main() -> i32 {
    var x: i32 = 3;
    var y: i32 = 4;
    Print(ToString(Add(x, y)));
    return 0;
}

4. C++ interop — calling C++ from Carbon

// math.h (C++ header)
#pragma once
namespace Math {
    int Add(int a, int b);
}

// Carbon imports C++ via library directive
package MathInterop api;

import Math;

fn Main() -> i32 {
    var sum: i32 = Math.Add(3, 4);
    return sum;
}

Interop uses Clang to parse C++ headers and generate Carbon bindings.

5. Carbon.h — exposing Carbon to C++

// Exported for C++ consumption
package MyLib api;

export fn PublicApi() -> i32 {
    return 42;
}

// C++ side includes generated Carbon headers
#include "carbon/generated/MyLib.h"

int main() {
    return MyLib::PublicApi();
}

Build system links Carbon object files with C++ via the Carbon toolchain driver.

6. Memory safety — checked borrows (experimental)

// Conceptual — API evolving
fn Process(data: i32*) -> i32 {
    // Checked borrow: compiler tracks lifetime
    var ref: i32* = data;
    return *ref;
}

Carbon aims for memory safety without garbage collection — checked pointers and ownership semantics. Feature maturity varies; check latest design docs.

7. Current limitations

• No stable standard library matching C++ <vector>, <thread>, etc.
• Toolchain APIs change between releases
• IDE support limited compared to C++/Rust
• Cross-compilation story immature
• Community and package ecosystem small

8. Carbon vs C++ decision tree

Choose Carbon exploration when
├── Large C++ codebase to gradually migrate
├── Team wants improved syntax with C++ interop
└── Can tolerate toolchain churn

Stay on C++ when
├── Production stability required now
├── Need full stdlib, Boost, mature tooling
├── Heavy template metaprogramming investment
└── Platform support beyond Carbon targets

9. Build a Carbon file

# Compile and link (toolchain evolving — check docs)
carbon compile sample.carbon --output=sample.o
carbon link sample.o --output=sample

# Typecheck only
carbon compile --phase=check sample.carbon

# Or build from source with Bazel in carbon-lang repo
./scripts/run_bazelisk.py build //toolchain:install --symlink_prefix=carbon/

Common Problems

Related Skills

• skills/compilers/cpp-templates — C++ being migrated from
• skills/compilers/cpp-modules — C++20 modules comparison
• skills/compiler-internals/compiler-frontend — Carbon compiler architecture
• skills/languages/hare-lang — alternative systems language
• skills/compilers/clang — Clang backend for Carbon
• skills/compilers/llvm — LLVM IR pipeline