CMakeToolchain

The CMakeToolchain can be used in the toolchain() method:

from conans import ConanFile, CMake, CMakeToolchain

class App(ConanFile):
    settings = "os", "arch", "compiler", "build_type"
    requires = "hello/0.1"
    generators = "cmake_find_package_multi"
    options = {"shared": [True, False], "fPIC": [True, False]}
    default_options = {"shared": False, "fPIC": True}

    def toolchain(self):
        tc = CMakeToolchain(self)
        tc.write_toolchain_files()

The CMakeToolchain will generate 2 files, after a conan install command (or before calling the build() method when the package is being built in the cache):

  • The main conan_toolchain.cmake file, that can be used in the command line.

  • A conan_project_include.cmake file, that will automatically be called right after the project() call for cmake>=3.15, containing definitions that only take effect after such call. For older cmake versions you should explicitly call include(.../conan_project_include.cmake) in your CMakeLists.txt.

These file will automatically manage the definition of cmake values according to current Conan settings:

  • Definition of the CMake generator platform and generator toolset

  • Definition of the CMake build_type

  • Definition of the CMAKE_POSITION_INDEPENDENT_CODE, based on fPIC option.

  • Definition of the C++ standard as necessary

  • Definition of the standard library used for C++

  • Deactivation of rpaths in OSX

Most of these things will be configurable, please provide feedback at: https://github.com/conan-io/conan/issues

constructor

def __init__(self, conanfile, generator=None, generator_platform=None, build_type=None,
             cmake_system_name=True, toolset=None, parallel=True, make_program=None):

Most of the arguments are optional and will be deduced from the current settings, and not necessary to define them.

definitions

This attribute allows defining CMake variables, for multiple configurations (Debug, Release, etc).

def toolchain(self):
    tc = CMakeToolchain(self)
    tc.definitions["MYVAR"] = "MyValue"
    tc.definitions.debug["MYCONFIGVAR"] = "MyDebugValue"
    tc.definitions.release["MYCONFIGVAR"] = "MyReleaseValue"
    tc.write_toolchain_files()

This will be translated to:

  • One set() definition for MYVAR in conan_toolchain.cmake file.

  • One set() definition, using a cmake generator expression in conan_project_include.cmake file, using the different values for different configurations. It is important to recall that things that depend on the build type cannot be directly set in the toolchain.

generators

The CMakeToolchain only works with the cmake_find_package and cmake_find_package_multi generators. Using others will raise, as they can have overlapping definitions that can conflict.

Using the toolchain in developer flow

One of the advantages of using Conan toolchains is that they can help to achieve the exact same build with local development flows, than when the package is created in the cache.

With the CMakeToolchain it is possible to do, for multi-configuration systems like Visual Studio (assuming we are using the cmake_find_package_multi generator):

# Lets start in the folder containing the conanfile.py
$ mkdir build && cd build
# Install both debug and release deps and create the toolchain
$ conan install ..
$ conan install .. -s build_type=Debug
# the conan_toolchain.cmake is common for both configurations
# Need to pass the generator WITHOUT the platform, that matches your default settings
$ cmake .. -G "Visual Studio 15" -DCMAKE_TOOLCHAIN_FILE=conan_toolchain.cmake
# Now you can open the IDE, select Debug or Release config and build
# or, in the command line
$ cmake --build . --config Release
$ cmake --build . --config Debug

NOTE: The platform (Win64), is already encoded in the toolchain. The command line shouldn’t pass it, so using -G "Visual Studio 15" instead of the -G "Visual Studio 15 Win64"

For single-configuration build systems:

# Lets start in the folder containing the conanfile.py
$ mkdir build_release && cd build_release
$ conan install ..
# the build type Release is encoded in the toolchain already.
# This conan_toolchain.cmake is specific for release
$ cmake .. -G "Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=conan_toolchain.cmake
$ cmake --build .  # or just "make"

# debug build requires its own folder
$ cd .. && mkdir build_debug && cd build_debug
$ conan install .. -s build_type=Debug
# the build type Debug is encoded in the toolchain already.
# This conan_toolchain.cmake is specific for debug
$ cmake .. -G "Unix Makefiles" -DCMAKE_TOOLCHAIN_FILE=conan_toolchain.cmake
$ cmake --build .  # or just "make"

CMake build helper

The CMake() build helper that works with the CMakeToolchain is also experimental, and subject to breaking change in the future. It will evolve to adapt and complement the toolchain functionality.

The helper is intended to be used in the build() method, to call CMake commands automatically when a package is being built directly by Conan (create, install)

from conans import CMake

def build(self):
    cmake = CMake(self)
    cmake.configure(source_folder="src")
    cmake.build()

It supports the following methods:

constructor

def __init__(self, conanfile, generator=None, build_folder=None, parallel=True,
             msbuild_verbosity="minimal"):
  • conanfile: the current recipe object. Always use self.

  • generator: CMake generator. Define it only to override the default one (like Visual Studio 15). Note that as the platform (x64, Win32…) is now defined in the toolchain it is not necessary to specify it here.

  • build_folder: Relative path to a folder to contain the temporary build files

  • parallel: Set it to False to deactivate using parallel builds. If activated, it will use cpu_count configuration as the number of parallel jobs to use.

  • msbuild_verbosity: Used to define the output of MSBuild builds.

configure()

def configure(self, source_folder=None):

Calls cmake, with the given generator and passing -DCMAKE_TOOLCHAIN_FILE=conan_toolchain.cmake. It will also provide the CMake generator in the command like, like -G "Visual Studio 15". Note that it is not necessary to specify the platform, like -G "Visual Studio 15 Win64", as the platform is already defined in the toolchain file.

  • source_folder: Relative path to the folder containing the root CMakeLists.txt

build()

def build(self, build_type=None, target=None):

Calls the build system. Equivalent to cmake --build . in the build folder.

  • build_type: Use it only to override the value defined in the settings.build_type for a multi-configuration generator (e.g. Visual Studio, XCode). This value will be ignored for single-configuration generators, they will use the one defined in the toolchain file during the install step.

  • target: name of the build target to run.

install()

def install(self, build_type=None):

Equivalent to run cmake --build . --target=install

  • build_type: Use it only to override the value defined in the settings.build_type. It can fail if the build is single configuration (e.g. Unix Makefiles), as in that case the build type must be specified at configure time, not build type.

test()

def test(self, build_type=None, target=None, output_on_failure=False):

Equivalent to running cmake --build . --target=RUN_TESTS.

  • build_type: Use it only to override the value defined in the settings.build_type. It can fail if the build is single configuration (e.g. Unix Makefiles), as in that case the build type must be specified at configure time, not build type.

  • target: name of the build target to run, by default RUN_TESTS or test.