Manage Shared Libraries with Environment Variables
The 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:
OPERATING SYSTEM |
ENVIRONMENT VARIABLE |
---|---|
WINDOWS |
PATH |
LINUX |
LD_LIBRARY_PATH |
OSX |
DYLD_LIBRARY_PATH |
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.
Example
We are packaging a tool called toolA
with a library and an executable that, 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 = "toolA"
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="toolA*", 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)
else:
...
def package_info(self):
self.env_info.PATH.append(os.path.join(self.package_folder, "bin"))
if self.options.shared:
self.env_info.LD_LIBRARY_PATH.append(os.path.join(self.package_folder, "lib"))
self.env_info.DYLD_LIBRARY_PATH.append(os.path.join(self.package_folder, "lib"))
Using the tool from a different package
If we are creating now a package that uses the ToolA
executable to compress some data. You can
call directly to the toolA.exe
, the required environment variables to locate both the executable
and the shared libraries are automatically available:
import os
from conans import tools, ConanFile
class PackageB(ConanFile):
....
name = "packageB"
version = "1.0"
requires = "toolA/1.0@myuser/stable"
def build(self):
...
# we can call directly the ``toolA`` executable. the shared library will be located too
exe_name = "toolA.exe" if self.settings.os == "Windows" else "toolA"
self.run("%s --someparams" % exe_name)
...
Building an application using the shared library from toolA
As we are building a final application, probably we will want to distribute it together with the
shared library from the toolA
, so we can use the Imports to import the required
shared libraries to our user space.
conanfile.txt
[requires]
toolA/1.0@myuser/stable
[generators]
cmake
[options]
toolA:shared=True
[imports]
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 dylib files from packages lib folder to my "bin" folder
In the terminal window and 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, such 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
environment variable is not directly transferred to the child process. In that case, you have to use it explicitely in
your conanfile.py:
def test(self):
# self.run('./myexe") # won't work, even if 'DYLD_LIBRARY_PATH' is in the env
self.run('DYLD_LIBRARY_PATH=%s ./myexe" % os.environ['DYLD_LIBRARY_PATH'])
Using the virtualenv generator
We could also use a virtualenv generator to get the
toolA
executable available:
conanfile.txt
[requires]
toolA/1.0@myuser/stable
[options]
toolA:shared=True
[generators]
virtualenv
In the terminal window:
conan install .
source activate
toolA --someparams
Using the virtualrunenv generator
Even if toolA
doesn’t declare the variables in the package_info
method, you can use
the virtualrunenv generator. It will set automatically
the environment variables poiting to the “lib” and “bin” folders.
conanfile.txt
[requires]
toolA/1.0@myuser/stable
[options]
toolA:shared=True
[generators]
virtualenv
In the terminal window:
conan install .
source activate
toolA --someparams