concept-synthesizer

automatically synthesizing C++20 template constraints for function templates

build

After cloning the repository, you can either build the project directly on your machine, or build it inside Docker.

In either case, you may want to change N_THREADS in setup.sh before running it.

normal build

• Platform dependencies: MacOS/Linux.
• Software dependencies: python3 (>= 3.5), wget, unzip, git, cmake, make, Clang (>= 17) / AppleClang (>= 15) / GCC (>= 14).

./setup.sh

Docker build

docker build -t concept-synthesizer-image .
docker run --rm -ti concept-synthesizer-image /bin/bash
cd workspace/
CC=gcc-14 CXX=g++-14 ./setup.sh

run

If you are on MacOS, you may need to run export SDKROOT=$(xcrun --sdk macosx --show-sdk-path) before running the synthesizer. You certainly don't need this inside the above Docker container.

run on a single file

./llvm-project-llvmorg-19.1.1/build_frontend/bin/clang-19 \
  -nostdinc++ \
  -isystem ./llvm-project-llvmorg-19.1.1/build_headers/include/c++/v1 \
  -std=c++20 \
  -E \
  -I<include-path-1> -I<include-path-2> ... \
  -o <file-path>.ii \
  <file-path>.cpp

./llvm-project-llvmorg-19.1.1/build_frontend/bin/concept-synthesizer \
  <file-path>.ii \
  -- \
  -x c++-cpp-output \
  -std=c++20

The synthesizer prints 5 sections to the standard output.

• Individual results: source location, synthesized constraints, etc. for each type template parameter
• Invalid calls: generated invalid calls to these function templates (needed for testing)
• Statistics: numbers of function templates, type template parameters, etc.
• Constrained code: the C++ code with constraints added by the synthesizer
• Resource consumption: the execution time of the synthesizer

run the automatic testing script

Run python3 run.py check to do a quick check of everything. This is also what the CI currently does.

Run python3 run.py measure to run the synthesizer on test/stl_algorithm.cpp and test/boost_special_functions.cpp, and report error message reduction statistics.