Manage Shared Libraries with Environment Variables

Shared libraries are loaded at runtime. The application executable needs to know where to find the required shared libraries when it runs.

Depending on the operating system, we can use environment variables to help the dynamic linker to find the shared libraries:









If your package recipe (A) is generating shared libraries you can declare the needed environment variables pointing to the package directory. This way, any other package depending on (A) will automatically have the right environment variable set, so they will be able to locate the (A) shared library.

Similarly if you use the virtualenv generator and you activate it, you will get the paths needed to locate the shared libraries in your terminal.


We are packaging a tool called toolA with a library and an executable that will, for example, compress data.

The package offers two flavors, shared library or static library (embedded in the executable of the tool and available to link with). You can use the toolA package library to develop another executable or library or you can just use the executable provided by the package. In both cases, if you choose to install the shared package of toolA you will need to have the shared library available.

import os
from conans import tools, ConanFile

class ToolA(ConanFile):
    name = "tool_a"
    version = "1.0"
    options = {"shared": [True, False]}
    default_options = {"shared": False}

    def build(self):
        # build your shared library

    def package(self):
        # Copy the executable
        self.copy(pattern="tool_a*", dst="bin", keep_path=False)

        # Copy the libraries
        if self.options.shared:
            self.copy(pattern="*.dll", dst="bin", keep_path=False)
            self.copy(pattern="*.dylib", dst="lib", keep_path=False)
            self.copy(pattern="*.so*", dst="lib", keep_path=False)

Using the tool from a different package

If we are now creating a package that uses the tool_a executable to compress some data, we can execute directly tool_a using RunEnvironment build helper to set the environment variables accordingly:

import os
from conans import tools, ConanFile

class PackageB(ConanFile):
    name = "package_b"
    version = "1.0"
    requires = "tool_a/1.0@myuser/stable"

    def build(self):
        exe_name = "tool_a.exe" if self.settings.os == "Windows" else "tool_a"[exe_name, "--someparams"], run_environment=True)

Building an application using the shared library from tool_a

As we are building a final application, we will probably want to distribute it together with the shared library from the tool_a, so we can use the Imports to import the required shared libraries to our user space.




 bin, *.dll -> ./bin # Copies all dll files from packages bin folder to my "bin" folder
 lib, *.dylib* -> ./bin # Copies all dylib files from packages lib folder to my "bin" folder
 lib, *.so* -> ./bin # Copies all so files from packages lib folder to my "bin" folder

Now you can build the project:

$ mkdir build && cd build
$ conan install ..
$ cmake .. -G "Visual Studio 14 Win64"
$ cmake --build . --config Release
$ cd bin && mytool

The previous example will work only in Windows and OSX (changing the CMake generator), because the dynamic linker will look in the current directory (the binary directory) where we copied the shared libraries too.

In Linux you still need to set the LD_LIBRARY_PATH, or in OSX, the DYLD_LIBRARY_PATH:

$ cd bin && LD_LIBRARY_PATH=$(pwd) && ./mytool

Using shared libraries from dependencies

If you are executing something that depends on shared libraries belonging to your dependencies, those shared libraries have to be found at runtime. In Windows, it is enough if the package added its binary folder to the system PATH. In Linux and OSX, it is necessary that the LD_LIBRARY_PATH and DYLD_LIBRARY_PATH environment variables are used.

Security restrictions might apply in OSX (read this thread), so the DYLD_LIBRARY_PATH and DYLD_FRAMEWORK_PATH environment variables are not directly transferred to the child process. In that case, you have to use it explicitly in your

def build(self):
    env_build = RunEnvironment(self)
    with tools.environment_append(env_build.vars):
        #"./myexetool") # won't work, even if 'DYLD_LIBRARY_PATH' and 'DYLD_FRAMEWORK_PATH' are in the env"DYLD_LIBRARY_PATH=%s DYLD_FRAMEWORK_PATH=%s ./myexetool" % (os.environ['DYLD_LIBRARY_PATH'], os.environ['DYLD_FRAMEWORK_PATH']))

Or you could use RunEnvironment helper described above.

Using virtualrunenv generator

virtualrunenv generator will set the environment variables PATH, LD_LIBRARY_PATH, DYLD_LIBRARY_PATH pointing to lib and bin folders automatically.




In the terminal window:

$ conan install .
$ source activate_run
$ tool_a --someparams
# Only For Mac OS users to avoid restrictions: