Let's break down your questions clearly and simply:

1. Does your CMAKE_PREFIX_PATH line take effect? #

Yes, absolutely! Here's the breakdown:

• The find_package(Torch REQUIRED) command needs CMake to locate PyTorch's configuration files (like TorchConfig.cmake). By default, CMake checks system-wide paths, but since your libtorch is in a custom directory (/User/root/libtorch), setting CMAKE_PREFIX_PATH tells CMake to prioritize this location when searching for the package.
• Since your project compiles and runs without issues, this line is definitely doing its job—without it, CMake would fail to find your custom libtorch installation and throw an error during the find_package step.

2. Why do headers work without target_include_directories? #

This is all thanks to modern CMake's imported targets—a feature PyTorch's CMake setup uses really well:

• When you run find_package(Torch REQUIRED), CMake creates an imported target (under the hood, ${TORCH_LIBRARIES} points to this target, which is officially named torch::torch). This target isn't just a library reference—it wraps up all the build details PyTorch needs, including:
  • Paths to its header files
  • Required compiler flags and C++ standard rules
  • Any dependent libraries that need to be linked
• When you use target_link_libraries(dcgan "${TORCH_LIBRARIES}"), CMake automatically passes all these pre-configured properties from the PyTorch target to your dcgan target. That means it adds PyTorch's header paths to your project's include search list automatically—no need for you to write an explicit target_include_directories line.

A quick best practice tip: PyTorch's official docs now recommend using the explicit imported target torch::torch instead of ${TORCH_LIBRARIES} for clarity. Your link line could be updated to:

target_link_libraries(dcgan torch::torch)

It works exactly the same way, but it's more aligned with modern CMake conventions.

内容的提问来源于stack exchange，提问作者Kidsunbo