如何为cmake提供package以便于find_package

作者: 生活简单些 | 来源:发表于2021-11-17 22:38 被阅读0次

1. CMake带来的改变

1.1 依赖关系的思维转变：用仓库的概念代替目录层级依赖

层级依赖:

├── TaihuApp
│       └── Qt::Quick
│       └── Qt5::Core
│       └── Qt5::Widgets
│       └── opencv ─────────────────────────────────┐ 
│       └── smt-logger ─────────────────┐           |
│       └── gtest ────────────────┐ [duplicate]     |
│       └── swc-camera            |     |       [duplicate]
│               └── opencv     ───|─────|───────────┘   
│               └── baumer        |     |           
│               └── tucsen  [duplicate] |           
│               └── protocol      |     |           
│               └── smt-logger ───|─────┘   
│               └── gtest ────────┘

扁平依赖:

├── Repository ────>──>──>──────│ 
│       └── Qt::Quick           |
│       └── Qt5::Core           |
│       └── Qt5::Widgets        |────────── Taihu
│       └── swc-camera          |
│       └── smt-logger          |
│       └── gtest               |────────── Kunlun
│       └── opencv              |
│       └── baumer              |
│       └── tucsen              |────────── other App
│       └── protocol            |
│       └── any other libs      |

图一：在每个项目里都存放一套自身需要的依赖库，类似离线式依赖包含关系；
图二：swc-camera依赖了opencv、baumer等别的库，但不存在包含关系，仓库里所有库的依赖关系都是通过配置进行关联的，本质所有的库都在项目之外的仓库里存放的。

1.2 简洁优雅的库依赖集成方式

project(swc-camera VERSION 1.0.0)

find_package(protocol REQUIRED)
find_package(smt-logger REQUIRED)
find_package(smt-timer REQUIRED)
find_package(opencv REQUIRED)
find_package(baumer REQUIRED)
find_package(tucsen REQUIRED)

aux_source_directory(. SRC_LIST)
add_library(${PROJECT_NAME} STATIC ${SRC_LIST})

target_link_libraries(${PROJECT_NAME} PRIVATE
    protocol
    smt-logger
    smt-timer
    baumer
    tucsen
    opencv)

target_include_directories(${PROJECT_NAME} PUBLIC
    $<BUILD_INTERFACE:${CMAKE_SOURCE_DIR}/include>
    $<INSTALL_INTERFACE:include>)

建议用target_include_directory()代替include_directory()，如果当前也是一个对外提供api的库

1.3 依赖库版本控制（vcpkg赋能）

{
  "name": "project",
  "version-string": "1.0.0",
  "supports": "(x64 | arm64) & (linux | osx | windows)",
  "dependencies": [
    { "name": "zlib", "version>=": "1.2.11#9" },
    { "name": "fmt", "version>=": "7.1.3#1" }
  ]
}

允许指定当前库的对外名字、版本、适用于哪些平台系统、以及依赖哪些别的库甚至那些库的指定版本

2. 自己的库如何能被find_pakcage(xxx)

cmake有两种方式让find_package(xxx) 能找到库，如果没有找到会报错，如下：

find_package(OpenCV)出现错误如下：

CMake Warning at CMakeLists.txt:37 (find_package):
  By not providing "FindOpenCV.cmake" in CMAKE_MODULE_PATH this project has
  asked CMake to find a package configuration file provided by "OpenCV", but
  CMake did not find one.

  Could not find a package configuration file provided by "OpenCV" with any of
  the following names:

    OpenCVConfig.cmake
    OpenCV-config.cmake

  Add the installation prefix of "OpenCV" to CMAKE_PREFIX_PATH or set "OpenCV_DIR"
  to a directory containing one of the above files.  If "OpenCV" provides a
  separate development package or SDK, be sure it has been installed.

简单翻译下：

cmake优先会以Moudule模式寻找，即：搜索CMAKE_MODULE_PATH指定路径下的FindXXX.cmake文件，默认路径按系统平台区分如下：

windows： C:/Program Files/CMake/share/cmake-3.xx/Modules
linux: /usr/share/cmake-3.xx/Modules

一旦找到了FindXXX.cmake, 则此库一般会提供以下变量，目的是方便调用者快速集成它：

<NAME>_FOUND
<NAME>_INCLUDE_DIRS or <NAME>_INCLUDES 
<NAME>_LIBRARIES or <NAME>_LIBS

如果没能找到FindXXX.cmake, 则尝试以Config模式：搜索指定路径下的XXXConfig.cmake或者XXX-config.cmake文件，搜索路径优先是cmake install的路径：

windows: C:/Program Files
linux: /usr/local

当然也支持在项目里通过CMAKE_PREFIX_PATH指定了寻找路径，或者直接通过设置XXX_DIR告知准确的查找路径。其实，还有一种做法是通过指定toolchain让cmake统一从toolchain里寻找。

2.1 Config方式

这是一种基于有项目源码的方式，需要为cmake组织的项目提供完整的install脚本，当执行install时候会在install目的地的lib目录下创建share目录，并在share目录里自动生成XXXConfig.cmake或者xxx-config.cmake等配置文件

cmake install的脚本相对比较通用，已经被我整理并抽取出来了，一般只要加在cmake项目的实现模块的CMakeList.txt最下面即可，如下：

# ============================== install script ==============================
set(HEADERS ${CMAKE_SOURCE_DIR}/include/swc_camera.h)
set_target_properties(${PROJECT_NAME} PROPERTIES PUBLIC_HEADER "${HEADERS}")

# Install the target and create export-set
install(TARGETS ${PROJECT_NAME}
    EXPORT ${PROJECT_NAME}Targets
    LIBRARY DESTINATION lib
    ARCHIVE DESTINATION lib
    RUNTIME DESTINATION bin
    PUBLIC_HEADER DESTINATION include)

# Generate the version file for the config file
include(CMakePackageConfigHelpers)
write_basic_package_version_file(
    ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}ConfigVersion.cmake
    VERSION ${PACKAGE_VERSION}
    COMPATIBILITY SameMajorVersion)

# Exporting Targets from the Build Tree
install(EXPORT ${PROJECT_NAME}Targets
    DESTINATION "lib/cmake/${PROJECT_NAME}")

# Create config file
configure_package_config_file(
    ${CMAKE_SOURCE_DIR}/Config.cmake.in ${PROJECT_NAME}Config.cmake
    INSTALL_DESTINATION "lib/cmake/${PROJECT_NAME}")

# Install config files
install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}Config.cmake
    ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}ConfigVersion.cmake
    DESTINATION "lib/cmake/${PROJECT_NAME}")

可以通过设置 CMAKE_INSTALL_PATH指定库安装的位置，cmake install library的命令是cmake --build ./ --target install，在linux下配合make可以简化为make install，这是makefile支持的：

cmake --build ./ --target install执行后如下：

PS E:\swc-camera\build> cmake --build ./ --target install
Microsoft (R) Build Engine version 16.11.1+3e40a09f8 for .NET Framework
Copyright (C) Microsoft Corporation. All rights reserved.

  swc-camera.vcxproj -> E:\swc-camera\build\src\Debug\swc-camera.lib
  example.vcxproj -> E:\swc-camera\build\example\Debug\example.exe
  tests.vcxproj -> E:\swc-camera\build\tests\Debug\tests.exe
  -- Install configuration: "Debug"
  -- Up-to-date: E:/LOCAL_REPOSITORY/lib/swc-camera.lib
  -- Up-to-date: E:/LOCAL_REPOSITORY/include/swc_camera.h
  -- Up-to-date: E:/LOCAL_REPOSITORY/lib/cmake/swc-camera/swc-cameraTargets.cmake      
  -- Up-to-date: E:/LOCAL_REPOSITORY/lib/cmake/swc-camera/swc-cameraTargets-debug.cmake
  -- Up-to-date: E:/LOCAL_REPOSITORY/lib/cmake/swc-camera/swc-cameraConfig.cmake       
  -- Up-to-date: E:/LOCAL_REPOSITORY/lib/cmake/swc-camera/swc-cameraConfigVersion.cmake

2.2 Module方式

这是一种当第三方库仅仅提供了编译好的binary库时候, 有时候有些库编译过程非常复杂且依赖多而且非常耗时，我们也可以用这种方式，为了让find_package(xxx)找到它的方式。我们需要写一个对应的FindXXX.cmake，在FindXXX.cmake里会指定尝试寻找库所在的路径，一般非常主流的库cmake的modules目录会提供，但以下三种情况需要自己编写FindXXX.cmake：

cmake的modules目录里提供的FindXXX.cmake描述的版本号和要用的不一致
非大众库，如baumer或者tucsen，cmake是不可能提供FindXXX.cmake的

在linux/mac系统里，大众库的FindXXX.cmake一般存在/usr/share/cmake-3.xx/Modules
在windows系统里，大众库的FindXXX.cmake存在C:\Program Files\CMake\share\cmake-3.xx\Modules

2.2.1 问：自己编写的FindXXX.cmake放哪里

答：默认find_package(xxx)会优先从cmake的Modules目录查找，意味着我们可以把自己的FindXXX.cmake放到cmake的Modules目录，但更优雅的方式是跟着项目走。在没有集成vcpkg的情况下，我们可以在项目根目录创建一个cmake目录，并将各种编写的FindXXX.cmake放于此处，随后需要在项目的CMakeList.txt里告知FindXXX.cmake所在目录，即：list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake")，当然有了vcpkg就简单多了，只要为此库创建一个独立的仓库，并将FindXXX.cmake直接放于其中，后续通过vcpkg将其install即可。

2.2.2 问：如何编写FindXXX.cmake

答：其实，FindXXX.cmake本质不一定要写，因为FindXXX.cmake的主要目的是通过find_library和find_path指定库的头文件和binary所在路径，但因为很多时候第三方库往往有很多头文件很多库文件而且还分debug/release，不能像下面这种方式简单描述，因此有必要提供一个独立的文件来描述库是怎么寻找和定义的，这样能让库寻找和库使用完全分离解耦。

find_path(TIFF_INCLUDE_DIR tiff.h
    /usr/local/include
    /usr/include)
    
find_library(TIFF_LIBRARY
    NAMES tiff tiff2
    PATHS /usr/local/lib /usr/lib)

include_directories(${TIFF_INCLUDE_DIRs})
add_executable(mytiff mytiff.c)
target_link_libraries(myprogram ${TIFF_LIBRARY})

因为实际的编写过程会很复杂，取决于不同形式的库，因此下面单独章节描述如何编写FindXXX.cmake.

3. 如何编写FindXXX.cmake

一个合格完整的FindXXX.cmake包含以下3个部分：

定义XXX_INCLUDE_DIRS和XXX_LIBRARIES： find_path()每次只能获得一个头文件所在路径，对于有很多头文件的库，需要通过多次find_path找到各自路径，并将它们合并为XXX_INCLUDE_DIRS, 如果一个库有很多库文件，那么也需要多次find_library()找到各个库对应的路径，并将其合并为XXX_LIBRARIES；
定义XXX_FOUND和 XXX_VERSION: 确认XXX_INCLUDE_DIRS和XXX_LIBRARIES都不为空，再定义XXX_FOUND和XXX_VERSION。至此，library已经可以被大幅简化集成，只是集成时候需要导入XXX_INCLUDE_DIRS作为库头文件，链接 XXX_LIBRARIES作为库文件，如果库区分Debug和Release，那么cmake还要以optimize和debug方式依赖对应的库；
创建Target: 确认XXX_FOUND不为空后再创建Target，通过add_library()定义库类型(SHARED|STATIC|INTERFACE)，通过set_target_properties()设置LIB的头文件路径、静态库地址、动态库地址、共享库的地址以及DLL路径。至此，库的集成简易程度已和源码库完全一样。

在写FindXXX.cmake前需要分析提供的第三方库的特性，根据不同的特性将会采取不同的方式编写FindXXX.cmake：

是否单个头文件或者单个库文件：相对来说，单个头文件和库文件的库写FindXXX.cmake会简洁很多，一个find_path和find_library就能描述所有的依赖关系；
库文件是否区分debug和release：只有windows库才有可能区分debug和release，如果区分意味着需要让cmake能动态找到对应版本的库文件；
windows库除了静态库是否还有动态库：在定义Target时候，需要在property里设置静态库和动态库的文件路径

3.1 单头文件&单库文件&单dll的情况

# FindOpenCV
# --------
#
# Find the opencv libraries
#
# Result Variables
# ^^^^^^^^^^^^^^^^
#
# The following variables will be defined:
#
# ``opencv_FOUND`` True if opencv found on the local system
#
# ``opencv_VERSION`` Version of opencv found
#
# ``opencv_INCLUDE_DIRS`` Location of opencv header files
#
# ``opencv_LIBRARIES`` List of the opencv libraries found
#

find_package(PkgConfig)

# ======================= define XXX_ROOT_DIR =======================
if (DEFINED ENV{LOCAL_REPOSITORY})
    set(opencv_ROOT_DIR $ENV{LOCAL_REPOSITORY})
endif()

if (DEFINED ENV{VCPKG_ROOT} AND DEFINED ENV{VCPKG_DEFAULT_TRIPLET})
    set(opencv_ROOT_DIR $ENV{VCPKG_ROOT}/installed/$ENV{VCPKG_DEFAULT_TRIPLET})
endif()

# ======================= find header files =======================
find_path(opencv_INCLUDE_DIR
    NAMES opencv2/opencv.hpp
    PATHS ${opencv_ROOT_DIR}/include /usr/local/include)

# ======================= find library files =======================
# define macro func to find libs
macro(opencv_FIND_LIBRARY libname)
    if(NOT opencv_${libname}_LIBRARY)
        find_library(opencv_${libname}_LIBRARY
            NAMES ${libname}
            PATHS ${opencv_ROOT_DIR}/lib /usr/local/lib)

        list(APPEND opencv_LIBRARY ${opencv_${libname}_LIBRARY})
    endif()
endmacro(opencv_FIND_LIBRARY)

if(WIN32)
    find_library(opencv_LIBRARY_DEBUG
        NAMES opencv_world412d.lib
        PATHS ${opencv_ROOT_DIR}/debug/lib /usr/local/lib)

    find_library(opencv_LIBRARY_RELEASE
        NAMES opencv_world412.lib
        PATHS ${opencv_ROOT_DIR}/lib /usr/local/lib)

    include(SelectLibraryConfigurations)
    select_library_configurations(opencv)
elseif(UNIX)
    # call macro func to find libs
    opencv_FIND_LIBRARY(libopencv_core.so)
    opencv_FIND_LIBRARY(libopencv_cudaarithm.so)
    opencv_FIND_LIBRARY(libopencv_cudafilters.so)
    opencv_FIND_LIBRARY(libopencv_cudaimgproc.so)
    opencv_FIND_LIBRARY(libopencv_highgui.so)
    opencv_FIND_LIBRARY(libopencv_imgcodecs.so)
    opencv_FIND_LIBRARY(libopencv_imgproc.so)
endif()

# ======================= find bin files =======================
if(WIN32)
    find_file(opencv_LIBRARY_DLL_DEBUG
        NAMES opencv_world412d.dll
        PATHS ${opencv_ROOT_DIR}/debug/bin)

    find_file(opencv_LIBRARY_DLL_RELEASE
        NAMES opencv_world412.dll
        PATHS ${opencv_ROOT_DIR}/bin)
endif()

# ======================= verify dependencies =======================
if (opencv_INCLUDE_DIR AND opencv_LIBRARY)
    set(opencv_FOUND TRUR CACHE BOOL "")
    set(opencv_VERSION "4.1.2" CACHE STRING "")

    set(opencv_INCLUDE_DIRS ${opencv_INCLUDE_DIR} CACHE STRING "")
    set(opencv_LIBRARIES ${opencv_LIBRARY} CACHE STRING "")

    find_package_handle_standard_args(opencv
        REQUIRED_VARS opencv_INCLUDE_DIRS opencv_LIBRARIES
        VERSION_VAR opencv_VERSION)
    mark_as_advanced(opencv_INCLUDE_DIRS opencv_LIBRARIES)
endif()

# ======================= create target =======================
if (opencv_FOUND)
    include(CMakePushCheckState)
    cmake_push_check_state()

    # set required properties
    set(CMAKE_REQUIRED_QUIET ${opencv_FIND_QUIETLY})
    set(CMAKE_REQUIRED_INCLUDES ${opencv_INCLUDE_DIRS})
    set(CMAKE_REQUIRED_LIBRARIES ${opencv_LIBRARIES})

    cmake_pop_check_state()

    if(NOT TARGET opencv)
        add_library(opencv SHARED IMPORTED)
        set_target_properties(opencv PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${opencv_INCLUDE_DIRS}")

        if(opencv_LIBRARY_DEBUG)
            set_property(TARGET opencv APPEND PROPERTY IMPORTED_CONFIGURATIONS DEBUG)
            set_target_properties(opencv PROPERTIES
                IMPORTED_LOCATION_DEBUG "${opencv_LIBRARY_DLL_DEBUG}"
                IMPORTED_IMPLIB_DEBUG "${opencv_LIBRARY_DEBUG}")
        endif()

        if(opencv_LIBRARY_RELEASE)
            set_property(TARGET opencv APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
            set_target_properties(opencv PROPERTIES
                IMPORTED_LOCATION_RELEASE "${opencv_LIBRARY_DLL_RELEASE}"
                IMPORTED_IMPLIB_RELEASE "${opencv_LIBRARY_RELEASE}")
        endif()

        if(NOT opencv_LIBRARY_RELEASE AND NOT opencv_LIBRARY_DEBUG)
            set_property(TARGET opencv APPEND PROPERTY IMPORTED_LOCATION "${opencv_LIBRARY}")
        endif()
    endif()
endif()

备注1：${opencv_ROOT_DIR}指向的库目录是动态的，如果定义了VCPKG_ROOT, 那么vcpkg就是库的寻找源；如果未定义VCPKG_ROOT但定义了LOCAL_REPOSITORY, 那么本地目录即为库寻找源；若都没有定义，那么头文件和库文件就只能从系统路径寻找了。不管当前是哪个平台，如：x64-windows、x86-windows、arm64-linux、x64-linux等等，${opencv_ROOT_DIR}下一般目录结构都是: include、lib以及bin。

3.1 批量寻找头文件: 模板化find_path()

因为find_path每次只能寻找一个头文件，需要多次调用将最终结果合并为XXX_INCLUDE_DIRS。其实，也可以如下通过定义宏或者函数批量寻找头文件：

# ======================= find header files =======================
# define macro func to find headers
macro(baumer_FIND_INCLUDE varname foldername headername)
    if(NOT baumer_${foldername}_INCLUDE_DIR)
        find_path(baumer_${foldername}_INCLUDE_DIR
            NAMES ${foldername}/${headername}
            PATHS ${baumer_ROOT_DIR}/include /usr/local/include)

        list(APPEND baumer_INCLUDE_DIRS ${baumer_${foldername}_INCLUDE_DIR})
        list(REMOVE_DUPLICATES baumer_INCLUDE_DIRS)
    endif()
endmacro(baumer_FIND_INCLUDE)

# call macro func to find headers
baumer_FIND_INCLUDE(bgapi2_ext            bgapi2_ext      bgapi2_ext.h)
baumer_FIND_INCLUDE(bgapi2_ext_addons     bgapi2_ext      bgapi2_ext_addons.h)
baumer_FIND_INCLUDE(bgapi2_ext_sc         bgapi2_ext_sc   bgapi2_ext_sc.h)
baumer_FIND_INCLUDE(bgapi2_def            bgapi2_genicam  bgapi2_def.h)
baumer_FIND_INCLUDE(bgapi2_featurenames   bgapi2_genicam  bgapi2_featurenames.h)
baumer_FIND_INCLUDE(bgapi2_genicam        bgapi2_genicam  bgapi2_genicam.hpp)

最终，baumer_INCLUDE_DIRS=E:\vcpkg\installed\x64-windows\include，你可能会疑惑那么多次的find_path定位到的include路径都是一样的，不是浪费么。No，其实目的是确保每个头文件都能被find_path到。

以上脚本演示了将每个public的头文件被寻找到并成为baumer_INCLUDES的一部分，库调用者include头文件则为#include <foldername/headername>，如：#include <bgapi2_ext/bgapi2_ext.h>

3.2 批量寻找库文件：模板化find_library()

和find_path一样find_library每次只能寻找一个库文件，需要多次调用将最终结果合并为XXX_LIBRARIES。同样也可以如下通过定义宏或者函数批量寻找库文件：

# ---------------- find library files----------------
# define macro func to find libs
macro(baumer_FIND_LIBRARY libname)
  if(NOT baumer_${libname}_LIBRARY)
     find_library(baumer_${libname}_LIBRARY 
      NAMES ${libname} 
      PATHS ${baumer_ROOT_DIR}/lib /usr/local/lib)

     list(APPEND baumer_LIBRARIES ${baumer_${libname}_LIBRARY})
  endif()
endmacro(baumer_FIND_LIBRARY)

# call macro func to find libs
Baumer_FIND_LIBRARY(bgapi2_ext.lib)
Baumer_FIND_LIBRARY(bgapi2_ext_sc.lib)
Baumer_FIND_LIBRARY(bgapi2_genicam.lib)

最终，baumer_LIBRARIES=E:\vcpkg\installed\x64-windows\lib\bgapi2_ext.libE:\vcpkg\installed\x64-windows\lib\bgapi2_ext_sc.libE:\vcpkg\installed\x64-windows\lib\bgapi2_genicam.lib

3.3 寻找库文件，但区分DEBUG和RELEASE

对于一些区分Debug和Release的windows库，我们不能一味用xxx_LIBRARIES描述所有库文件:

find_library(serial_LIBRARY_DEBUG
    NAMES seriald.lib
    PATHS ${serial_ROOT_DIR}/debug/lib /usr/local/lib)

find_library(serial_LIBRARY_RELEASE
    NAMES serial.lib
    PATHS ${serial_ROOT_DIR}/lib /usr/local/lib)

include(SelectLibraryConfigurations)
select_library_configurations(serial)

select_library_configurations(xxx)的引入使得cmake configure会自动选取对应版本的库赋值与xxx_LIBRARY，意味着target_link_library时候不用区分optimize和debug的库

至此，库的集成将大幅简化，因为可以做到了屏蔽不同路径的头文件不同路径，不同名字的库文件，甚至不用区分系统平台：

find_package(XXX)
target_include_directory(app ${XXX_INCLUDE_DIRS})
target_link_library(app PRIVATE
     optimize ${XXX_LIBRARIES_RELEASE}
     debug ${XXX_LIBRARIES_DEBUG})

3.3 创建Target

为了彻底、完全做到跟源码库集成一样简洁：省略头文件导入和不区分debug/release，我们可以手动创建Target。

3.3.1 只有单个static或者单个so的Target

if(NOT TARGET xxx)
    add_library(xxx SHARED|STATIC|UNKNOWN IMPORTED)
    set_target_properties(xxx PROPERTIES 
        INTERFACE_INCLUDE_DIRECTORIES "${xxx_INCLUDE_DIRS}"
        IMPORTED_LOCATION "${xxx_LIBRARY}")
endif()

add_library()时候可以指定library是SHARED库、STATIC库，如果不确定那种也可以填UNKNOWN，因为SHARED库和STATIC库的必要设置的property是一样的

3.3.2 只有单个dll和单个lib的Target

if(NOT TARGET xxx)
    add_library(xxx SHARED IMPORTED)
    set_target_properties(xxx PROPERTIES 
        INTERFACE_INCLUDE_DIRECTORIES "${xxx_INCLUDE_DIRS}"
        IMPORTED_LOCATION "${xxx_LIBRARY_DLL}"
        IMPORTED_IMPLIB "${xxx_LIBRARY}")
endif()

3.3.3 有多个static和多个dll的Target

if(NOT TARGET xxx)
    add_library(xxx INTERFACE IMPORTED)
    set_target_properties(xxx PROPERTIES 
        INTERFACE_INCLUDE_DIRECTORIES "${xxx_INCLUDE_DIRS}"
        INTERFACE_LINK_LIBRARIES "${xxx_LIBRARIES}"
        IMPORTED_LOCATION "${xxx_LIBRARY_DLLS}")
endif()

3.3.4 区分Debug/Release的Target

if(NOT TARGET xxx)
    add_library(xxx STATIC IMPORTED)
    set_target_properties(xxx PROPERTIES INTERFACE_INCLUDE_DIRECTORIES "${xxx_INCLUDE_DIRS}")

    if(xxx_LIBRARY_DEBUG)
        set_property(TARGET xxx APPEND PROPERTY IMPORTED_CONFIGURATIONS DEBUG)
        set_target_properties(xxx PROPERTIES IMPORTED_LOCATION_DEBUG "${xxx_LIBRARY_DEBUG}")
    endif()

    if(xxx_LIBRARY_RELEASE)
        set_property(TARGET xxx APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
        set_target_properties(xxx PROPERTIES IMPORTED_LOCATION_RELEASE "${xxx_LIBRARY_RELEASE}")
    endif()

    if(NOT xxx_LIBRARY_RELEASE AND NOT xxx_LIBRARY_DEBUG)
        set_property(TARGET xxx APPEND PROPERTY IMPORTED_LOCATION "${xxx_LIBRARY}")
    endif()
endif()

这是一个典型的兼容Windows和Linux的Target创建案例，因为很多时候Windows库区分Debug和Release，而Linux的SO库是不区分的。

至此，库的集成将变为如下方式：

find_package(xxx REQUIRED)
target_link_library(app xxx)

4. vcpkg

CMake提供了依赖库寻找的功能，当找不到依赖库时只会报错，并不会自动下载依赖库，因为CMake不是真正的包管理器，顶多是一个包寻找器，所以得配合vcpkg，vcpkg是宗旨是即时编译当前或者指定平台的库，vcpkg内部提供了众多下载依赖途径选项，如：http, ftp, gitlab, github, bitbucket，svn，cvs等等。

4.1 如何让vcpkg和cmake结合让库依赖自动化

vcpkg的详细功能丰富，可自行查看官网，这里只指导如何快速上手创建私有port，并让vcpkg和cmake集成达到编译即下载依赖库的作用，我们分别以2种案例来介绍：

基于源码构建的库：相对更简单
基于只有头文件和二进制库文件的第三方库: 需要区分多平台独立，复杂一些

4.2 创建私有port

所谓port就是一个以库名为名字的目录，里面放此库的获取来源配置、版本及依赖关系配置、使用说明等，创建一个私有port的步骤如下：

在你的C++项目根目录创建一个文件夹叫vcpkg-ports, 如果你的库名叫xxx，那么在vcpkg-ports里创建一个目录叫xxx，那么这个xxx即是一个port。
在port目录里创建一个名为portfile.cmake的文本文件，在此配置文件里定义库代码或者二进制获取方式，以及相关配置。

portfile.cmake (以源码方式将vcpkg和cmake集成):

set(SOURCE_PATH ${CURRENT_BUILDTREES_DIR}/src/${PORT})

vcpkg_from_git(
    OUT_SOURCE_PATH SOURCE_PATH
    URL ssh://git@lmsman-bitbucket01.europe.leicams.com:7999/lmswet-bitbucket01/~fei.zhang/smt-timer.git
    REF 9e950a8a52b7304e7da2ab59fd485f39095dca9b
    HEAD_REF master
)

# configure project and try to enable ninja
vcpkg_configure_cmake(
    SOURCE_PATH ${SOURCE_PATH} 
    PREFER_NINJA)

vcpkg_install_cmake()
vcpkg_copy_pdbs()

# relocate target to vcpkg
vcpkg_fixup_cmake_targets(
    CONFIG_PATH lib/cmake/${PORT} 
    TARGET_PATH /share/${PORT})

# remove headers in debug mode
file(REMOVE_RECURSE ${CURRENT_PACKAGES_DIR}/debug/include)

# install license and copyright
file(
    INSTALL ${SOURCE_PATH}/COPYING
    DESTINATION ${CURRENT_PACKAGES_DIR}/share/${PORT}
    RENAME copyright)

用已经编译好的二进制库的方式将vcpkg和cmake集成：

set(SOURCE_PATH "${CURRENT_BUILDTREES_DIR}/src/${PORT}")

vcpkg_from_git(
    OUT_SOURCE_PATH SOURCE_PATH
    URL git@10.10.231.59:fei.zhang/baumer.git
    REF 2d2d0a217da810f1ea71ca807c5337aed3e84e12
    HEAD_REF master
)

if (WIN32)
    # install headers
    file(GLOB_RECURSE HEADERS "${SOURCE_PATH}/windows/include/*")
    foreach(HEADER ${HEADERS})
         get_filename_component(ABSOLUTE_DIR ${HEADER} DIRECTORY)
         string(REGEX MATCH "include.+" NODE_DIR ${ABSOLUTE_DIR})
         file(INSTALL ${HEADER} DESTINATION "${CURRENT_PACKAGES_DIR}/${NODE_DIR}")
    endforeach()

    # install libs
    file(GLOB_RECURSE LIBRARIES "${SOURCE_PATH}/windows/lib/*")
    file(INSTALL ${LIBRARIES} DESTINATION "${CURRENT_PACKAGES_DIR}/lib")

    # install dlls
    file(GLOB_RECURSE BINS "${SOURCE_PATH}/windows/bin/*")
    file(INSTALL ${BINS} DESTINATION "${CURRENT_PACKAGES_DIR}/bin/${PORT}")
elseif (UNIX)
    # install headers
    file(GLOB_RECURSE HEADERS "${SOURCE_PATH}/linux/include/*")
    foreach(HEADER ${HEADERS})
         get_filename_component(ABSOLUTE_DIR ${HEADER} DIRECTORY )
         string(REGEX MATCH "include.+" NODE_DIR ${ABSOLUTE_DIR})
         file(INSTALL ${HEADER} DESTINATION "${CURRENT_PACKAGES_DIR}/${NODE_DIR}")
    endforeach()

    # install libs
    file(GLOB_RECURSE LIBRARIES "${SOURCE_PATH}/linux/lib/*")
    file(INSTALL ${LIBRARIES} DESTINATION "${CURRENT_PACKAGES_DIR}/lib")
endif()

# install FindXXX.cmake allow App can find pakcage via find_package(xxx)
file(INSTALL ${SOURCE_PATH}/FindBaumer.cmake DESTINATION "${CURRENT_PACKAGES_DIR}/share/${PORT}")

# relocate the directory for cmake to find FindXXX.cmake
file(INSTALL ${CMAKE_CURRENT_LIST_DIR}/vcpkg-cmake-wrapper.cmake DESTINATION "${CURRENT_PACKAGES_DIR}/share/${PORT}")

# install cmake integration usage
file(INSTALL ${CMAKE_CURRENT_LIST_DIR}/usage DESTINATION "${CURRENT_PACKAGES_DIR}/share/${PORT}")

# allow incomplete packages to pass validation
set(VCPKG_POLICY_EMPTY_PACKAGE enabled)

在此port目录里创建一个名为vcpkg.json的文本文件，在此配置里定义当前库的基础信息以及依赖关系等，如下以swc-camera的vcpkg.json为例（如果此库没有其他任何依赖且没有版本，可以不用提供）：

{
    "name": "swc-camera",
    "version-string": "v1.0.0",
    "description": "swc-camera is an integrated camera feature sdk",
    "dependencies": [
        "protocol",
        "smt-logger",
        "smt-timer",
        "opencv",
        "baumer",
        "tucsen",
        "gtest"
    ]
}

4.3 如何通过vcpkg自动下载库和依赖库

以swc-camera为例，安装命令为: vcpkg install swc-camera --port-overlay=vcpkg-ports
当看到如下过程log，意味着下载swc-camera以及其依赖库已经开始工作了

--port-overlay=vcpkg-ports 作用是指定vcpkg安装库，库的port配置来自vcpkg-ports目录，如果项目不依赖任何私有托管的仓库，则不用指定--port-overlay

PS E:\vcpkg> .\vcpkg.exe install swc-camera --overlay-ports=vcpkg-ports
Computing installation plan...
The following packages will be built and installed:
  * baumer[core]:x64-windows -> v1.0.0 -- E:\vcpkg\vcpkg-ports\baumer
  * opencv[core]:x64-windows -> v4.1.2 -- E:\vcpkg\vcpkg-ports\opencv
  * smt-logger[core]:x64-windows -> v1.0.0 -- E:\vcpkg\vcpkg-ports\smt-logger
  * smt-timer[core]:x64-windows -> v1.0.0 -- E:\vcpkg\vcpkg-ports\smt-timer
    swc-camera[core]:x64-windows -> v1.0.0 -- E:\vcpkg\vcpkg-ports\swc-camera
  * tucsen[core]:x64-windows -> v1.0.0 -- E:\vcpkg\vcpkg-ports\tucsen
Additional packages (*) will be modified to complete this operation.
Detecting compiler hash for triplet x64-windows...

当最后出现如下信息，意味着swc-camera安装成功：

Total elapsed time: 16.2 s

The package swc-camera:x64-windows provides CMake targets:

    find_package(swc-camera CONFIG REQUIRED)
    target_link_libraries(main PRIVATE swc-camera)

4.4 如何通过vcpkg卸载库

以swc-camera为例，卸载命令为: vcpkg remove swc-camera --port-overlay=vcpkg-ports

需要注意的是remove不会将依赖库一同卸载，因为cmake库依赖都是引用依赖，不是包含依赖，你不用的库可能别的项目在用。

4.5 如何将vcpkg集成到cmake项目中

# use local repository if defined
if (DEFINED ENV{LOCAL_REPOSITORY})
  set(CMAKE_PREFIX_PATH $ENV{LOCAL_REPOSITORY})
endif()

# preferred to use vcpkg if defined
if(DEFINED ENV{VCPKG_ROOT})
    set(CMAKE_TOOLCHAIN_FILE "$ENV{VCPKG_ROOT}/scripts/buildsystems/vcpkg.cmake")
endif()

---------------------------------------------------------------------

project(testApp)
find_package(smt-logger)
add_executable(${PROJECT_NAME} main.cpp)
target_link_libraries(${PROJECT_NAME} smt-logger)

如上，演示了如何集成vcpkg到cmake项目，同时也指定了本地统一库寻找目录，且有寻找优先级。

首先，find_package()默认优先会尝试从vcpkg里寻找，假设系统环境变量定义了vcpkg根目录 —— VCPKG_ROOT
随后，若VCPKG_ROOT 未定义（假设你不喜欢vcpkg，想自己折腾），则尝试从本地统一库寻找目录里寻找，假设系统环境变量定义了统一库寻找目录 —— LOCAL_REPOSITORY
如果以上环境变量都没有，那么则尝试找默认的路径，linux从/usr/local里找，windows从C:/Program File里找