在编译时为运行时设置环境变量

Question

I have a C++ Vulkan program that needs multiples libraries to be available at run-time. Also, Vulkan has a feature called "Validation Layers" which is configured with a config file.

In run-time my program needs to know where those libraries are and where that config file is. I'm guessing there's no way of doing it programmatically, but if there is let me know. To workaround this I set environment variables, namely LD_LIBRARY_PATH (for it to find the libraries) and VK_LAYER_PATH (for it to find Vulkan's Validation Layer config file).

This works, but I want a better way to do this, because this doesn't allow me to simply double-click the file and run it. I must first set the env vars, which is bad if I'm deploying the program.

My question is: is there a compiler/linker option to do this?

This is the workaround I'm using in my makefile:

run:
    LD_LIBRARY_PATH=./path/to/lib1/:./path/to/lib2 VK_LAYER_PATH=./path/to/vulkan/config ./program_name

I am using Linux, g++ and make.

Answer 1

If you know where the libraries you need to link against will be installed you could set an rpath. This will add the search path to the ELF header. When the dynamic linker runs it will search these locations in addition to the default locations.

Add to your compilation line -Wl,-rpath ./path/to/lib1/ to drop lib1 from the LD_LIBRARY_PRELOAD list. The -Wl is needed so the the compiler passes the flag onto the linker where it is actually recognized.

This blog seems to have a good description of all the different options