How to build an executable while installing in normal and editable mode

Hello, I'm currently learning how to use CMake and pybind11 with this project. As part of the installation process, I need to build and install a C++ source file. To ensure that everything is clear, I've included the reproduction code below.

Description

• My project structure: It's basically the same except there is my cpp source file helloworld.cpp in src/ folder.

$ tree -L 2
.
├── CMakeLists.txt
├── pybind11
├── pyproject.toml
├── setup.py
└── src
    ├── helloworld.cpp
    └── main.cpp

• src/helloworld.cpp: A trivial helloworld program. $ cat src/helloworld.cpp

#include <iostream>

int main(int argc, char **argv) {
  std::cout << "Hello World!!!\n";
  return 0;
}

• CMakeLists.txt: It's basically the same as CMakeLists.tx except add_executable(helloworld src/helloworld.cpp) is added to build executable.

$ cat CMakeLists.txt 
cmake_minimum_required(VERSION 3.22)
project(cmake_example)

add_executable(helloworld src/helloworld.cpp)

add_subdirectory(pybind11)
pybind11_add_module(cmake_example src/main.cpp)

target_compile_definitions(cmake_example
                           PRIVATE VERSION_INFO=${EXAMPLE_VERSION_INFO})

• pyproject.toml:

$ cat pyproject.toml 
[build-system]
requires = ["setuptools>=62.0.0", "wheel", "cmake>=3.22"]
build-backend = "setuptools.build_meta"

• pybind11 is a git submodule, and setup.py is exactly the same as this setup.py.

Build in normal mode.

• Create virtual env and update pip: python3 -m venv venv && source venv/bin/activate && python3 -m pip install --upgrade pip
• Install in normal mode: python3 -m pip install .
• cmake_example library is installed correctly but my helloworld program is not. There is no helloworld program in venv.

$ ls venv/lib/python3.8/site-packages/
cmake_example-0.0.1.dist-info             
cmake_example.cpython-38-x86_64-linux-gnu.so

Build in editable mode.

• Create virtual env and update pip: python3 -m venv venv && source venv/bin/activate && python3 -m pip install --upgrade pip
• Install in editable mode: python3 -m pip install -e .
• cmake_example library is also installed correctly but my helloworld program is not. There is no helloworld program created in the current folder.

$ ls
cmake_example.cpython-38-x86_64-linux-gnu.so  cmake_example.egg-info  CMakeLists.txt  pybind11  pyproject.toml  setup.py  src  venv

My experiment:

I actually tried to make it work by modifying setup.py with the following way.

• Add code snippet in setup.py in Line:47.

        bindir = extdir / "bin"
        print(f"User's executables resides in {bindir}")
        cmake_args = [
            # My innovation. helloworld can be called in scripts.
            f"-DCMAKE_RUNTIME_OUTPUT_DIRECTORY={bindir}{os.sep}",
            f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={extdir}{os.sep}",
            f"-DPYTHON_EXECUTABLE={sys.executable}",
            f"-DCMAKE_BUILD_TYPE={cfg}",  # not used on MSVC, but no harm
        ]

This approach kind of work. Now I can do pip install . and helloworld program appears in venv/lib/python3.8/site-packages/bin/helloworld. However, this approach doesn't work in editable model. I.e., I do pip install -e ., but helloworld program is no created anywhere in the project.

My question is how to achieve the following goal: I would like the helloworld program to be created automatically when I run pip install . and pip install -e ., and I want this program to be installed in the correct Python root path in both modes. Do you have any suggestions or advice on how I can make this happen? Thank you for your help!

System info:
OS: ubuntu20.04
Python: Python3.8

Start by reading setup.py and CMakeLists.txt, then reproduce both pip install . and pip install -e . using the commands and project layout in the issue. Compare where the helloworld executable is emitted in each mode and trace the existing CMake output-directory settings. Done means both installation modes create the executable under the appropriate Python root path.