Cpp2 → C++ transpiler with MLIR-based Sea-of-Nodes IR pipeline

Cppfort is an experimental compiler for the Cpp2 language (from cppfront) that implements a Sea-of-Nodes (SoN) intermediate representation using MLIR. It combines:

• Traditional AST pipeline: Direct Cpp2 → C++ transpilation
• SoN/MLIR pipeline: Graph-based optimization with MLIR dialect
• Clang AST mapping: Inverse inference from C++ AST to MLIR regions

TableGen-defined dialect with ops for:

• Control flow: if, for, while, loop, return
• Data flow: constant, add, sub, mul, div, phi, binop
• Functions: func, call, ufcs_call
• Memory: new, load, store with alias classes
• Cpp2 features: contract, metafunction, var

Python toolchain for extracting Clang AST → MLIR op mappings:

• emit_mappings.py: Extract mapping candidates from C++ source
• batch_emit_mappings.py: Process multiple files, aggregate results
• validate_against_dialect.py: Validate mappings against dialect
• run_inference.sh: Wrapper with libclang configuration

• docs/ARCHITECTURE.md: Overall design
• docs/MAPPING_SPEC.md: Mapping schema specification
• docs/MAPPING_TASK.md: Task definition
• docs/MAPPING_PROGRESS.md: Implementation status
• docs/sea-of-nodes/: 24 chapters from Cliff Click's book

# Single file
./tools/inference/run_inference.sh tools/inference/emit_mappings.py \
  -i tools/inference/samples/sample_son.cpp \
  -o mappings.json -- -std=c++20

# Batch process
python3 tools/inference/batch_emit_mappings.py \
  -i corpus/inputs \
  -o output_dir \
  --limit 10 \
  --aggregate

python3 tools/inference/validate_against_dialect.py \
  -m mappings.json \
  -d include/Cpp2Dialect.td

# macOS (Homebrew LLVM)
brew install llvm ninja cmake

# Set LLVM in PATH (add to ~/.zshrc or ~/.bashrc)
export PATH="/opt/homebrew/opt/llvm/bin:$PATH"

# Configure with CMake (Homebrew LLVM, Ninja)
cmake -B build -G Ninja .

# Build main targets
ninja -C build cppfort              # Main transpiler executable
ninja -C build Cpp2Transpiler       # Transpiler library
ninja -C build cppfront             # Cpp2 transpiler (from cppfront)

# Build and run tests
ninja -C build test                 # Run all CTest suites
./build/tests/cpp26_contracts_test  # Run specific test

# Corpus processing (requires cppfront)
ninja -C build corpus_transpile     # Transpile .cpp2 → .cpp
ninja -C build corpus_ast           # Generate AST dumps
ninja -C build corpus_reference     # Combined transpile + AST

# Clean build
rm -rf build && cmake -B build -G Ninja . && ninja -C build

# Transpile a Cpp2 file
./build/src/cppfort input.cpp2 -o output.cpp

# Generate MLIR
./build/src/cppfort input.cpp2 --emit-mlir -o output.mlir

• ✅ MLIR dialect with 24 ops (control flow, data flow, memory, cpp2-specific)
• ✅ Mapping extraction toolchain
• ✅ Schema validation (6,301 mappings, 100% pass rate)
• ✅ Sample corpus and test infrastructure
• ✅ Documentation and usage guides

• 🔄 MLIR emitter (template → MLIR IR code generation)
• 🔄 Cpp2 file transpilation integration
• 🔄 Full corpus validation

• 📋 CI/CD pipeline
• 📋 Sea-of-Nodes chapter pattern extraction
• 📋 Roundtrip validation (C++ → AST → MLIR → C++)

Latest validation (sample_son.cpp, 6,301 mappings):

• call: 2,170 mappings (CallExpr)
• func: 1,497 mappings (FunctionDecl)
• return: 1,021 mappings (ReturnStmt)
• var: 677 mappings (VarDecl)
• binop: 599 mappings (BinaryOperator)
• if: 249 mappings (IfStmt)
• while: 46 mappings (WhileStmt)
• for: 42 mappings (ForStmt)

All mappings validate successfully against the dialect definition.

C++ Source (.cpp2) 
    ↓
[cppfront transpiler] → C++ (.cpp)
    ↓
[Clang AST parser]
    ↓
[Mapping Extractor] → Mapping Artifacts (JSON)
    ↓
[MLIR Emitter] → cpp2 Dialect Ops
    ↓
[SoN Optimizer] → Optimized MLIR
    ↓
[Code Generator] → Target Code

• LLVM/MLIR: 21.1+ (Homebrew LLVM on macOS)
• Clang: 21.1+ (from Homebrew LLVM)
• CMake: 3.28+ (for build system)
• Ninja: (recommended build tool)
• Python: 3.10+ with libclang bindings (for mapping tools)

• cppfront - Herb Sutter's Cpp2 transpiler
• Sea of Nodes IR - Cliff Click's SoN book/reference
• MLIR - Multi-Level IR framework

[License details to be determined]

See .github/copilot-instructions.md for development guidelines.