rusty-cpp: Making C++ Rusty

This project aims to make C++ safer and more reliable by adopting Rust's proven safety principles, especially its borrow-checking system. We provide a static analyzer that enforces Rust-like ownership and borrowing rules through compile-time analysis.

---

Borrow checking and lifetime analysis

🎯 Vision

This project aims to catch memory safety issues at compile-time by applying Rust's proven ownership model to C++ code. It helps prevent common bugs like use-after-move, double-free, and dangling references before they reach production.

Though C++ is flexible enough to mimic Rust's idioms in many ways, implementing a borrow-checking without modifying the compiler system appears to be impossible, as analyzed in document.

We provide rusty-cpp-checker, a standalone static analyzer that enforces Rust-like ownership and borrowing rules for C++ code, bringing memory safety guarantees to existing C++ codebases without runtime overhead. rusty-cpp-checker does not bringing any new grammar into c++. Everything works through simple annoations such as adding // @safe enables safety checking on a function.

Note: two projects that (attempt to) implement borrow checking in C++ at compile time are Circle C++ and Crubit. As of 2025, Circle is not open sourced, and its design introduces aggressive modifications, such as the ref pointer ^. Crubit is not yet usable on this feature.

Example

Here's a simple demonstration of how const reference borrowing works:

// @safe
void demonstrate_const_ref_borrowing() {
    int value = 42;
    
    // Multiple const references are allowed (immutable borrows)
    const int& ref1 = value;  // First immutable borrow - OK
    const int& ref2 = value;  // Second immutable borrow - OK  
    const int& ref3 = value;  // Third immutable borrow - OK
    
    // All can be used simultaneously to read the value
    int sum = ref1 + ref2 + ref3;  // OK - reading through const refs
}

// @safe
void demonstrate_const_ref_violation() {
    int value = 42;
    
    const int& const_ref = value;  // Immutable borrow - OK
    int& mut_ref = value;          // ERROR: Cannot have mutable borrow when immutable exists
    
    // This would violate the guarantee that const_ref won't see unexpected changes
    mut_ref = 100;  // If allowed, const_ref would suddenly see value 100
}

Analysis Output:

error: cannot borrow `value` as mutable because it is also borrowed as immutable
  --> example.cpp:6:5
   |
5  |     const int& const_ref = value;  // Immutable borrow - OK
   |                            ----- immutable borrow occurs here
6  |     int& mut_ref = value;          // ERROR
   |          ^^^^^^^ mutable borrow occurs here

✨ Features

Core Capabilities

• 🔄 Borrow Checking: Enforces Rust's borrowing rules (multiple readers XOR single writer)
• 🔒 Ownership Tracking: Ensures single ownership of resources with move semantics
• ⏳ Lifetime Analysis: Validates that references don't outlive their data

Detected Issues

• Use-after-move violations
• Multiple mutable borrows
• Dangling references
• Lifetime constraint violations
• RAII violations
• Data races (through borrow checking)

📦 Installation

⚠️ Build Requirements (IMPORTANT)

This tool requires the following native dependencies to be installed before building from source or installing via cargo:

• Rust: 1.70+ (for building the analyzer)
• LLVM/Clang: 14+ (for parsing C++ - required by clang-sys)
• Z3: 4.8+ (for constraint solving - required by z3-sys)

Note: These dependencies must be installed system-wide before running cargo install rusty-cpp or building from source. The build will fail without them.

Installing from crates.io

Once you have the prerequisites installed:

# macOS: Set environment variable for Z3
export Z3_SYS_Z3_HEADER=/opt/homebrew/include/z3.h

# Linux: Set environment variable for Z3
export Z3_SYS_Z3_HEADER=/usr/include/z3.h

# Install from crates.io
cargo install rusty-cpp

# The binary will be installed as 'rusty-cpp-checker'
rusty-cpp-checker --help

Building from Source

macOS

# Install dependencies
brew install llvm z3

# Clone the repository
git clone https://github.com/shuaimu/rusty-cpp
cd rusty-cpp

# Build the project
cargo build --release

# Run tests
./run_tests.sh

# Add to PATH (optional)
export PATH="$PATH:$(pwd)/target/release"

Note: The project includes a .cargo/config.toml file that automatically sets the required environment variables for Z3. If you encounter build issues, you may need to adjust the paths in this file based on your system configuration.

Linux (Ubuntu/Debian)

# Install dependencies
sudo apt-get update
sudo apt-get install llvm-14-dev libclang-14-dev libz3-dev

# Clone and build
git clone https://github.com/shuaimu/rusty-cpp
cd rusty-cpp
cargo build --release

Windows

# Install LLVM from https://releases.llvm.org/
# Install Z3 from https://github.com/Z3Prover/z3/releases
# Set environment variables:
set LIBCLANG_PATH=C:\Program Files\LLVM\lib
set Z3_SYS_Z3_HEADER=C:\z3\include\z3.h

# Build
cargo build --release

🚀 Usage

Basic Usage

# Analyze a single file
rusty-cpp-checker path/to/file.cpp

# Analyze with verbose output
rusty-cpp-checker -vv path/to/file.cpp

# Output in JSON format (for IDE integration)
rusty-cpp-checker --format json path/to/file.cpp

Standalone Binary (No Environment Variables Required)

For release distributions, we provide a standalone binary that doesn't require setting environment variables:

# Build standalone release
./build_release.sh

# Install from distribution
cd dist/rusty-cpp-checker-*/
./install.sh

# Or use directly
./rusty-cpp-checker-standalone file.cpp

See RELEASE.md for details on building and distributing standalone binaries.

Environment Setup (macOS)

For convenience, add these to your shell profile:

# ~/.zshrc or ~/.bashrc
export Z3_SYS_Z3_HEADER=/opt/homebrew/opt/z3/include/z3.h
export DYLD_LIBRARY_PATH=/opt/homebrew/opt/llvm/lib:$DYLD_LIBRARY_PATH

🛡️ Safety Annotations

The borrow checker uses a unified annotation system for gradual adoption in existing codebases:

Unified Rule

@safe and @unsafe annotations attach to the next code element (namespace, function, or first statement).

// Example 1: Namespace-level safety
// @safe
namespace myapp {
    void func() { /* checked */ }
}

// Example 2: Function-level safety
// @safe
void checked_function() { /* checked */ }

void unchecked_function() { /* not checked - default is unsafe */ }

// Example 3: First-element rule
// @safe
int global = 42;  // Makes entire file safe

// Example 4: Unsafe blocks within safe functions
// @safe
void mixed_safety() {
    int value = 42;
    int& ref1 = value;
    
    // @unsafe
    {
        int& ref2 = value;  // Not checked in unsafe block
    }
    // @endunsafe
}

Default Behavior

• Files are unsafe by default (no checking) for backward compatibility
• Use @safe to opt into borrow checking
• Use @unsafe to explicitly disable checking

📝 Examples

Example 1: Use After Move

#include <memory>

void bad_code() {
    std::unique_ptr<int> ptr1 = std::make_unique<int>(42);
    std::unique_ptr<int> ptr2 = std::move(ptr1);
    
    *ptr1 = 10;  // ERROR: Use after move!
}

Output:

error: use of moved value: `ptr1`
  --> example.cpp:6:5
   |
6  |     *ptr1 = 10;
   |     ^^^^^ value used here after move
   |
note: value moved here
  --> example.cpp:5:34
   |
5  |     std::unique_ptr<int> ptr2 = std::move(ptr1);
   |                                  ^^^^^^^^^^^^^^

Example 2: Multiple Mutable Borrows

void bad_borrow() {
    int value = 42;
    int& ref1 = value;
    int& ref2 = value;  // ERROR: Cannot borrow as mutable twice
}

Example 3: Lifetime Violation

int& dangling_reference() {
    int local = 42;
    return local;  // ERROR: Returning reference to local variable
}

🏗️ Architecture

┌─────────────┐     ┌──────────┐     ┌────────┐
│   C++ Code  │────▶│  Parser  │────▶│   IR   │
└─────────────┘     └──────────┘     └────────┘
                          │                │
                    (libclang)              ▼
                                    ┌──────────────┐
┌─────────────┐     ┌──────────┐   │   Analysis   │
│ Diagnostics │◀────│  Solver  │◀──│   Engine     │
└─────────────┘     └──────────┘   └──────────────┘
                         │                │
                       (Z3)        (Ownership/Lifetime)

Components

• Parser (src/parser/): Uses libclang to build C++ AST
• IR (src/ir/): Ownership-aware intermediate representation
• Analysis (src/analysis/): Core borrow checking algorithms
• Solver (src/solver/): Z3-based constraint solving for lifetimes
• Diagnostics (src/diagnostics/): User-friendly error reporting

---

Tips in writing rusty c++

Writing C++ that is easier to debug by adopting principles from Rust.

Being Explicit

Explicitness is one of Rust's core philosophies. It helps prevent errors that arise from overlooking hidden or implicit code behaviors.

No computation in constructors/destructors

Constructors should be limited to initializing member variables and establishing the object's memory layout—nothing more. For additional initialization steps, create a separate Init() function. When member variables require initialization, handle this in the Init() function rather than in the constructor.

Similarly, if you need computation in a destructor (such as setting flags or stopping threads), implement a Destroy() function that must be explicitly called before destruction.

Composition over inheritance

Avoid inheritance whenever possible.

When polymorphism is necessary, limit inheritance to a single layer: an abstract base class and its implementation class. The abstract class should contain no member variables, and all its member functions should be pure virtual (declared with = 0). The implementation class should be marked as final to prevent further inheritance.

Use move and disallow copy assignment/constructor

Except for primitive types, prefer using move instead of copy operations. There are multiple ways to disallow copy constructors; our convention is to inherit from the boost::noncopyable class:

class X: private boost::noncopyable {};

If copy from an object is necessary, implement move constructor and a Clone function:

Object obj1 = move(obj2.Clone()); // move can be omitted because it is already a right value. 

Memory Safety, Pointers, and References

No raw pointers

Avoid using raw pointers except when required by system calls, in which case wrap them in a dedicated class.

Use POD types

Try to use POD types if possible. POD means "plain old data". A class is POD if:

• No user-defined copy assignment
• No virtual functions
• No destructor

Incrementally Migrate to Rust (C++/Rust Interop)

Some languages (like D, Zig, and Swift) offer seamless integration with C++. This makes it easier to adopt these languages in existing C++ projects, as you can simply write new code in the chosen language and interact with existing C++ code without friction.

Unfortunately, Rust does not support this level of integration (perhaps intentionally to avoid becoming a secondary option in the C++ ecosystem), as discussed here. Currently, the best approach for C++/Rust interoperability is through the cxx/autocxx crates. This interoperability is implemented as a semi-automated process based on C FFIs (Foreign Function Interfaces) that both C++ and Rust support. However, if your C++ code follows the guidelines in this document, particularly if all types are POD, the interoperability experience can approach the seamless integration offered by other languages (though this remains to be verified).

---

⚠️ Note: This is an experimental tool. Use it at your own discretion.