MastringCmakeVersion31:Chapter 11

From 흡혈양파의 번역工房
Jump to navigation Jump to search

CHAPTER ELEVEN::PORTING CMAKE TO NEW PLATFORMS AND LANGUAGES

In order to generate build files for a particular system, CMake needs to determine what system it is running on and what compiler tool s to use for enabled languages. To do this CMake loads a series of files containing CMake code from the Modules directory. This all has to happen before the first try-compile or try-run is executed. To avoid having to re-compute al l of this information for each try-compile and for subsequent runs of CMake, the discovered values are stored in several configured fi les that are read each time CMake i s run . These fi les are also copied into the try-compile and try-run directories. Thi s chapter w i l l describe how this process of system and tool discovery works. An understanding of the process i s necessary to extend CMake to run on new platforms, and to add support for new languages.


The Determine System Process

The first thing CMake needs to do is to determine what platform it is running on and what the target platform is. Except for when you are cross compiling, the host platform and the target platform are identical. The host platform is determined by loading the CMakeDetermineSystem.cmake file. On POSIX systems, "un­ame" is used to get the name of the system. CMAKE_HOST_SYSTEM_NAME (page 651) is set to the result of uname -s, and CMAKE_HOST_SYSTEM_VERSION (page 652) is set to the result of uname -r. On Windows systems, CMAKE_HOST_SYSTEM_NAME is set to Windows and CMAKE_HOST_SYSTEM_VERSION is set to the value returned by the system function GetVersionEx. The variable CMAKE_HOST_SYSTEM (page 651) is set to a combination of CMAKE_HOST_SYSTEM_NAME and CMAKE_HOST_SYSTEM_VERSION as fol­lows:

${CMAKE_HOST_SYSTEM_NAME}-${CMAKE_HOST_SYSTEM_VERSION}

Additionally, CMake tries to figure out the processor of the host. On POSIX systems it uses un­ame -m or uname -p to retrieve this information, while on Windows it uses the environment variable PROCESSOR_ARCHITECTURE. CMAKE_HOST_SYSTEM_PROCESSOR (page 651) holds the value of the result.

Now that CMake has the information about the host that it is running on, it needs to find this information for the target platform. The results will be stored in the CMAKE_SYSTEM_NAME (page 653), CMAKE_SYSTEM_VERSION (page 653), CMAKE_SYSTEM (page 653), and CMAKE_SYSTEM_PROCESSOR (page 653) variables, corresponding to the CMAKE_HOST_SYSTEM_* variables described above. See the "Cross compiling with CMake" chapter on how this is done when cross compiling. In all other cases the CMAKE_SYSTEM_* variables will be set to the value of their corresponding CMAKE_HOST_SYSTEM_* variable.

Once the CMAKE_SYSTEM information has been determined, CMakeSystem.cmake.in is configured into ${CMAKE_BINARY_DIR}/CMakeFiles/CMakeSyste.cmake. CMake versions prior to 2.6.0 did not support cross compiling, and so only the CMAKE_SYSTEM_* set of variables was available.


The Enable Language Process

After the platform has been determined, the next step is to enable all languages speci­fied in the project() (page 327) command. For each language specified, CMake loads CMakeDetermine(LANG)Compiler.cmake where LANG is the name of the language specified in the project() (page 327) command. For example with project(f Fortran), the file is called CMakeDetermineFortranCompiler.cmake. This file discovers the compiler and tools that will be used to compile files for the particular language. Starting with version 2.6.0 CMake tries to identify the compiler for C, C++ and Fortran not only by its filename, but by compiling some source code, which is named CMake(LANG)CompilerId.(LANG_SUFFIX). If this succeeds, it will return a unique id for ev­ery compiler supported by CMake. Once the compiler has been determined for a language, CMake configures the file CMake(LANG)Compiler.cmake.in into CMake(LANG)Compiler.cmake.

After the platform and compiler tools have been determined, CMake loads CMakeSystemSpecificationInformation.cmake which in turn will load ${CMAKE_SYSTEM_NAME}.cmake from the platform subdirectory of modules if it exists for the platform. An example would be SunOS.cmake. This file contains OS specific information about compiler flags, creation of executables, libraries, and object files.

Next, CMake loads CMake(LANG)Information.cmake for each LANG that was enabled, which loads ${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG-${CMAKE_SYSTEM_PROCESSOR}.cmake if it exists, and after that ${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG.cmake. In these file names COMPILER_ID references the compiler identification determined as described above. The CMake(LANG)Information.cmake file contains default rules for creating executables, libraries, and object files on most UNIX systems. The de­faults can be overridden by setting values in either ${CMAKE_SYSTEM_NAME}.cmake or ${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG.cmake.

${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG-${CMAKE_SYSTEM_PROCESSOR}.cmake is intended to be used only for cross compiling, and is loaded before ${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG.cmake, so variables can be set up which can then be used in the rule variables.

In addition to the files with the COMPILER_ID in their name, CMake also supports these files using the COMPILER_BASE_NAME. COMPILER_BASE_NAME is the name of the compiler with no path infor­mation. For example, cl would be the COMPILER_BASE_NAME for the Microsoft Windows compiler, and Windows-cl.cmake would be loaded. If a COMPILER_ID exists, it will be preferred over the COMPILER_BASE_NAME, since on one side the same compiler can have different names, but there can be also different compilers all with the same name. This means, if

${CMAKE_SYSTEM_NAME}-${COMPILER_ID}-LANG-${CMAKE_SYSTEM_PROCESSOR}.cmake

was not found, CMake tries

${CMAKE_SYSTEM_NAME}-${COMPILER_BASE_NAME}.cmake

and if

${CMAKE_SYSTEM_NAME}-${COMPILER_BASE_ID}-LANG.cmake

was not found, CMake tries

${CMAKE_SYSTEM_NAME}-${COMPILER_BASE_NAME}.cmake

CMake(LANG)Information. cmake and associated Platform files define special CMake variables, called rule variables. A rule variable consists of a list of commands separated by spaces and each enclosed by quotes. In addition to the normal variable expansion performed by CMake, some special tag variables are expanded by the Makefile generator. Tag variables have the syntax of <NAME>, where NAME is the name of the variable. An example rule variable is CMAKE_CXX_CREATE_SHARED_LIBRARY, and the default setting is

set (CMAKE_CXX_CREATE_SHARED_LIBRARY
      "<CMAKE_CXX_COMPILER> <CMAKE SHARED_LIBRARY_CXX_FLAGS>
      <LINK_FLAGS> <CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS>
      <CMAKE_SHARED_LIBRARY_SONAME_CXX_FLAG><TARGET_SONAME> -o
      <TARGET> <OBJECTS> <LINK_LIBRARIES>")

At this point, CMake has determined the system it is running on, the tools it will be using to compile the enabled languages, and the rules to use the tools. This means there is enough information for CMake to perform a try-compile. CMake will now test the detected compilers for each enabled language by loading CMakeTest(LANG)Compiler.cmake. This file will usually run a try-compile on a simple source file for the given language to make sure the chosen compiler actually works.

Once the platform has been determined, and the compilers have been tested, CMake loads a few more files that can be used to change some of the computed values. The first file that is loaded is CMake(PROJECTNAME)Compatibility.cmake, where PROJECTNAME is the name given to the top level PROJECT command in the project. The project compatibility file is used to add backwards compatibil­ity fixes into CMake. For example, if a new version of CMake fails to build a project that the previous version of CMake could build, then fixes can be added on a per project basis to CMake. The last file that is loaded is ${CMAKE_USER_MAKE_RULES_OVERRIDE}. This file is an optionally user supplied variable, that can allow a project to make very specific platform-based changes to the build rules.


Porting to a New Platform

Many common platforms are already supported by CMake. However, you may come across a compiler or platform that has not yet been used. If the compiler uses an Integrated Development Environment (IDE), then you will have to extend CMake from the C++ level. However, if the compiler supports a standard make program, then you can specify in CMake the rules to use to compile object code and build libraries by creating CMake configuration files. These files are written using the CMake language with a few special tags that are expanded when the Makefiles are created by CMake. If you run CMake on your system and get a message like the following, you will want to read how to create platform specific settings.

System is unknown to CMake, create:
Modules/Platform/MySystem.cmake
to use this system, please send your config file to
cmake@www.cmake.org so it can be added to CMake

At a minimum you will need to create the Platform/${CMAKE_SYSTEM_NAME}.cmake file for the new platform . Depending on the tools for the platform, you may also want to create Platform/${CMAKE_SYSTEM_NAME}-${COMPILER_BASE_NAME}cmake. On most systems, there is a vendor compiler and the GNU compiler. The rules for both of these compilers can be put in Platform/${CMAKE_SYSTEM_NAME} cmake instead of creating separate files for each of the compilers. For most new systems or compilers, if they follow the basic UNIX compiler fags you will only need to specify the system specific flags for shared library and module creation.

The following example is from Platform/IRIX.cmake. This file specifies several fags, and also one CMake rule variable. The rule variable tells CMake how to use the IRIX CC compiler to create a static library, which is required for template instantiation to work with IRIX CC.

# there is no -ldl required on this system
set (CMAKE_DL_LIBS "")

# Specify the flag to create a shared c library
set (CMAKE_SHARED_LIBRARY_CREATE_C_FLAGS
        "-shared -rdata_shared")

# Specify the flag to create a shared c++ library
set (CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS
        "-shared -rdata_shared")

# specify the flag to specify run time paths for shared
# libraries -rpath
set (CMAKE_SHARED_LIBRARY_RUNTIME_C_FLAG "-Wl,-rpath,")

# specify a separator for paths on the -rpath, if empty
# then -rpath will be repeated.
set (CMAKE_SHARED_LIBRARY_RUNTIME_C_FLAG_SEP "")

# if the compiler is not GNU, then specify the initial flags
if (NOT CMAKE_COMPILER_IS_GNUCXX)

    # use the CC compiler to create static library
    set (CMAKE_CXX_CREATE_STATIC_LIBRARY
        "<CMAKE_CXX_COMPILER> -ar -o <TARGET> <OBJECTS>")

    # initializes flags for the native compiler
    set (CMAKE_CXX_FLAGS_INIT "")
    set (CMAKE_CXX_FLAGS_DEBUG_INIT "-g")
    set (CMAKE_CXX_FLAGS_MINSIZEREL_INIT "-O3 -DNDEBUG")
    set (CMAKE_CXX_FLAGS_RELEASE_INIT "-02 -DNDEBUG")
    set (CMAKE_CXX_FLAGS_RELWITHDEBINFO_INIT "-O2")
endif (NOT CMAKE_COMPILER_IS_GNUCXX)


Adding a New Language

In addition to porting CMake to new platforms, a user may want to add a new language. This can be done either through the use of custom commands, or by defining a new language for CMake. Once a new language is defined, the standard add_library() (page 274) and add_executable() (page 273) commands can be used to create libraries and executables for the new language. To add a new lan­guage, you need to create four files. The name LANG has to match-in exact case- the name used in the PROJECT() (page 327) command to enable the language. For example, Fortran has the file CMakeDeterminFortranCompiler.cmake, and it is enabled with a call like this project(f Fortran). The four files are as follows:

CMakeDetermine(LANG)Compiler.cmake This file will find the path to the compiler for LANG and then configure CMake(LANG)Compiler.cmake.in.

CMake(LANG)Compiler.cmake.in This file should be used as input to a configure file call in the CMakeDetermine(LANG)Compiler.cmake file. It is used to store compiler information and is copied down into try-compile directories so that try compiles do not need to re-detennine and test the LANG.

CMakeTest(LANG)Compiler.cmake This should make use of a try compile command to make sure the compiler and tools are working. If the tools are working, the following variable should be set in this way:

set (CMAKE_(LANG)_COMPILER_WORKS 1 CACHE INTERNAL "")

CMake(LANG)lnformation.cmake Set values for the following rule variables for LAN G :


Rule Variable Listing

For each language that CMake supports, the following rule variables are expanded into build Makefiles at generation time. LANG is the name used in the PROJECT (name LANG) command. CMake currently supports CXX, C, Fortran, and Java as values for LANG.


General Tag Va riables

The following set of variables will be expanded by CMake.

<TARGET> The name of the target being built (this may be a full path).

<TARGET_QUOTED> The name of the target being built (this may be a full path) double quoted.

<TARGET_BASE> This is replaced by the name of the target without a suffix.

<TARGET_SONAME> This is replaced by CMAKE_SHARED_LIBRARY_SONAME_(LANG)_FLAG

<OBJECTS> This is the list of object files to be linked into the target.

<OBJECTS_QUOTED> This is the list of object files to be linked into the target double quoted.

<OBJECT> This is the name of the object file to be built.

<LINK_LIBRARIES> This is the list of libraries that are linked into an executable or shared object.

<FLAGS> This contains the command line fags for the linker or compiler.

<LINK_FLAGS> These are the fags used at link time.

<SOURCE> The source file name.


Language Specific Information

The following set of variables related to the compiler tools will also be expanded.

<CMAKE_(LANG)_COMPILER> This is the (LANG) compiler command.

<CMAKE_SHARED_LIBRARY_CREATE_(LANG)_FLAGS> These are the fags used to create a shared library for (LANG) code.

<CMAKE_SHARED_MODULE_CREATE_(LANG)_FLAGS> These are the fags used to create a shared module for (LANG) code.

<CMAKE_(LANG)_LINK_FLAGS> These are the fags used to link a (LANG) program.

<CMAKE_AR> This is the command to create a .a archive file.

<CMAKE_RANLIB> This is the command to ranlib a.a archive file.


Compiler and Platform Examples

Como Compiler

A good example to look at is the como compiler on Linux, found in Modules/Platforms/Linux-como.cmake. This compiler requires several non-standard com­mands when creating libraries and executables in order to instantiate C++ templates.

# create a shared C++ library
set (CMAKE CXX_CREATE_SHARED_LIBRARY
    "<CMAKE CXX_COMPILER> --prelink_objects <OBJECTS>"
    "<CMAKE_CXX_COMPILER>
    <CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS> <LINK_FLAGS> -o <TARGET>
    <OBJECTS> <LINK_LIBRARIES>")

# create a C++ static library
set (CMAKE_CXX_CREATE_STATIC_LIBRARY
    "<CMAKE_CXX_COMPILER> --prelink_objects <OBJECTS>"
    "<CMAKE_AR> cr <TARGET> <LINK_FLAGS> <OBJECTS>"
    "<CMAKE_RANLIB> <TARGET> ")

set (CMAKE_CXX_LINK_EXECUTABLE
    "<CMAKE_CXX_COMPILER> --prelink_objects <OBJECTS>"
    "<CMAKE_CXX_COMPILER> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS>
    <FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
set (CMAKE_SHARED_LIBRARY_RUNTIME_FLAG "")
set (CMAKE_SHARED_LIBRARY_C_FLAGS "")
set (CMAKE_SHARED_LIBRARY_LINK_FLAGS "")

This overrides the creation of libraries (shared and static), and the linking of executable C++ programs. You can see that the linking process of executables and shared libraries requires an extra command that calls the compiler with the flag --prelink_objects, and gets all of the object files passed to it.


Borland Compiler

The full Borland compiler rules can be found in Platforms/Windows-bcc32.cmake. The following code is an excerpt from that file, showing some of the features used to define rules for the Borland compiler set.

set (CMAKE_CXX_CREATE_SHARED_LIBRARY
    "<CMAKE_CXX_COMPILER> ${CMAKE_START_TEMP_FILE}-e<TARGET>
    -tWD <LINK_FLAGS> -tWR <LINK_LIBRARIES> <OBJECTS>S$ {CMAKE_END_TEMP_FILE}"
    "implib <c -w <TARGET_BASE>.lib <TARGET_BASE>.dll"
   )

set (CMAKE_CXX_CREATE_SHARED_MODULE
    ${CMAKE_CXX_CREATE_SHARED_LIBRARY})

# create a C shared library
set (CMAKE_C_CREATE_SHARED_LIBRARY
    "<CMAKE_C_COMPILER> ${CMAKE_START_TEMP_FILE}-e<TARGET> -tWD
    <LINK_FLAGS> -tWR <LINK_LIBRARIES>
    <OBJECTS>${CMAKE_END_TEMP_FILE}"
    "implib -c -w <TARGET_BASE>.lib <TARGET_BASE>.dll"
    )

# create a C++ static library
set (CMAKE_CXX_CREATE_STATIC_LIBRARY "tlib
    ${CMAKE_START_TEMP_FILE}/p512 <LINK_FLAGS> /a <TARGET_QUOTED>
    <OBJECTS_QUOTED>${CMAKE_END_TEMP_FILE}")

# compile a C++ file into an object file
set (CMAKE_CXX_COMPILE_OBJECT
    "<CMAKE_CXX_COMPILER> ${CMAKE_START_TEMP_FILE}-DWIN32 -P
    <FLAGS> -o<OBJECT> -c <SOURCE>${CMAKE_END_TEMP_FILE}")


Extending CMake

Occasionally you will come across a situation where you want to do something during your build process that CMake cannot seem to handle. Examples of this include creating wrappers for C++ classes to make them available to other languages, or creating bindings for C++ classes to support runtime introspection. In these cases you may want to extend CMake by adding your own commands. CMake supports this capability through its C plugin API Using this API, a project can extend CMake to add specialized commands to handle project-specific tasks.

A loaded command in CMake is essentially a C code plugin that is compiled into a shared library (a.k.a. DLL). This shared library can then be loaded into the running CMake to provide the functionality of the loaded command. Creating a loaded command is a two step process. You must first write the C code and CMakeLists file for the command, and then place it in your source tree. Next you must modify your project's CMakeLists file to compile the loaded command and load it. We will start by looking at writing the plugin. Before resorting to creating a loaded command, you should first see if you can accomplish what you want with a macro. With the commands in CMake a macro/function has almost the same level of flexibility as a loaded command, but does not require compilation or as much complexity. You can almost always, and should, use a macro/function instead of a loaded command.


Creating a Loaded Command

While CMake itself is written in C++, we suggest that you write your plugins using only C code. This avoids a number of portability and compiler issues that can plague C++ plugins being loaded into CMake executables. The API for a plugin is defined in the header file cmCPluginAPI.h. This file defines all of the CMake functions that you can invoke from your plugin. It also defines the cmLoadedCommandInfo structure that is passed to a plugin. Before going into detail about these functions, consider the following simple plugin:

#include "cmCPluginAPI.h"
Static int InitialPass (void *inf, void *mf,
                       int argc, char *argv[})
{
  cmLoadedCommandinfo *info = (cmLoadedCommandinfo *) inf;
  info->CAPI->AddDefinition(mf, "FOO", "BAR");

  return 1;
}

void CM_PLUGIN_EXPORT
HELLO WORLDInit (cmLoadedCommandinfo *info)
{
  info->InitialPass = InitialPass;
  info->Name = "HELLO WORLD";
}

First this plugin includes the cmCPluginAPI.h file to get the definitions and structures required for a plugin. Next it defines a static function called InitialPass that will be called whenever this loaded com­mand is invoked. This function is always passed four parameters: the cmLoadedCommandInfo structure, the Makefile, the number of arguments, and the list of arguments. Inside this function, we typecast the inf argument to its actual type and then use it to invoke the C API(CAPI) AddDefinition function. This function will set the variable FOO to the value of BAR in the current cmMakefile instance.

The second function is called HELLO_WORLD init, and it will be called when the plugin is loaded. The name of this function must exactly match the name of the loaded command with Init appended. In this example the name of the command is HELLO_WORLD, so the function is named HELLO_WORLD init. This function will be called as soon as your command is loaded. It is responsible for initializing the elements of the cmLoadedCommandInfo structure. In this example it sets the InitialPass member to the address of the InitialPass function defined above. It will then set the name of the command by setting the Name member to HELLO_WORLD.


Using a Loaded Command

Now let us consider how to use this new HELLO_WORLD command in a project. The basic process is that CMake will have to compile the plugin into a shared library and then dynamically load it. To do this you first create a subdirectory in your project's source tree called CMake or CMakeCommands (by convention, any name can be used). Place the source code to your plugin in that directory. We recommend naming the file with the prefix cm and then the name of the command. For example, cmHELLO_WORLD.c. Then you must create a simple CMakeLists.txt file for this directory that includes instructions to build the shared library. Typically this will be the following

project (HELLO_WORLD)
set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}"
  "${(CMAKE_ANSI_CXXFLAGS}"
  )

set (CMAKE_C_FLAGS "${CMAKE_C_FLAGS}"
  "${CMAKE_ANSI_CFLAGS}"
  )

include_directories (${CMAKE_ROOT}/include
  ${CMAKE_ROOT}/Source
  )

add_library (cmHELLO_WORLD MODULE cmHELLO_WORLD.c)

It is critical that you name the library cm, followed by the name of the command as shown in the add_library call in the above example (e.g. cmHELLO_WORLD). When CMake loads a command it assumes that the command is in a library named using that pattern. The next step is to modify your project's main CMakeLists file to compile and load the plugin. This can be accomplished with the following code:

# if the command has not been loaded, compile and load it
if (NOT COMMAND HELLO WORLD)

  # try compiling it first
  try_compile (COMPILE_OK
    ${PROJECT_BINARY_DIR}/CMake
    ${PROJECT_SOURCE_DIR}/CMake
    HELLO WORLD
    )

  # if it compiled OK then load it
  if (COMPILE_OK)
    load_command (HELLO WORLD
      ${PROJECT_BINARY_DIR}/CMake
      ${PROJECT_BINARY_DIR}/CMake/Debug
    )

  # if it did not compile OK, then display an error
  else (COMPILE_OK)
    message ("error compiling HELLO_WORLD extension")
  endif (COMPILE_OK)

endif (NOT COMMAND HELLO WORLD)


In the above example you would simply replace HELLO_WORLD with the name of your command and replace ${PROJECT_SOURCE_DIR}/CMake with the actual name of the subdirectory where you placed your loaded command. Now, let us look at creating loaded commands in more detail. We will start by looking at the cmLoadedCommandInfo structure.

typedef const char* (*CM_DOC_FUNCTION) ();

typedef int (*CM_INITIAL_PASS FUNCTION) (
    void *info, void *mf, int argc, char *[]);

typedef void (*CM_FINAL_PASS_FUNCTION) (
    void *info, void *mf);
typedef void (*CM_DESTRUCTOR_FUNCTION) (void *info);

typedef struct {
    unsigned long reserved1;
    unsigned long reserved2;
    cmCAPI *CAPI;
    int m_Inherited;
    CM_INITIAL_PASS_FUNCTION InitialPass;
    CM_FINAL_PASS_FUNCTION FinalPass;
    CM_DESTRUCTOR_FUNCTION Destructor;
    CM_DOC_FUNCTION GetTerseDocumentation;
    CM_DOC_FUNCTION GetFullDocumentation;
    const char *Name;
    char *Error;
    void *ClientData;
    } cmLoadedCommandinfo;


The first two entries of the structure are reserved for future use. The next entry, CAPI, is a pointer to a structure containing pointers to all the CMake functions you can invoke from a plugin. The m_Inherited member only applies to CMake versions 2.0 and earlier. It can be set to indicate if this command should be inherited by subdirectories or not. If you are creating a command that will work with versions of CMake prior to 2.2 then you probably want to set this to zero. The next five members are pointers to functions that your plugin may provide. The InitialPass function must be provided, and it is invoked whenever your loaded command is invoked from a CMakeLists file. The FinalPass function is optional, and is invoked after configuration but before generation of the output. The Destructor function is optional, and will be invoked when your command is destroyed by CMake (typically on exit). It can be used to clean up any memory that you have allocated in the InitialPass or FinalPass. The next two functions are optional, and are used to provide documentation for your command. The Name member is used to store the name of your command. This is what will be compared against when parsing a CMakeLists file. It should be in all caps in keeping with CMake's naming conventions. The Error and ClientData members are used internally by CMake; you should not directly access them. Instead you can use the CAPI functions to manipulate them.

Let us consider some of the common CAPI functions you will use from within a loaded command. First, we will consider some utility functions that are provided specifically for loaded commands. Since loaded com­ mands use a C interface they will receive arguments as (int argc, char *argv[]). For convenience, you can call GetTotalArgumentSize(argc, argv), which will return the total length of all the ar­guments. Likewise, some CAPI methods will return an (argc, argv) pair that you will be responsible for freeing. The FreeArguments(argc, argv) function can be used to free such return values. If your loaded command has a FinalPass(), then you might want to pass data from the InitialPass() to the FinalPass() invocation. This can be accomplished using the SetClientData(void *info, void *data) and void *GetClientData(void *info) functions. Since the client data is passed as a void *argument, any client data larger than a pointer must be allocated and then finally freed in your Destructor() function. Be aware that CMake will create multiple instances of your loaded command so using global variables or static variables is not recommended. If you should encounter an error in executing your loaded command, you can call SetError(void *info, const char *errorSteing) to pass an error message on to the user.

Another group of CAPI functions worth noting are the cmSourceFile functions. cmSourceFile is a C++ object that represents information about a single file including its full path, file extension, spe­cial compiler flags, etc. Some loaded commands will need to either create or access cmSourceFile instances. This can be done using the void *CreateSourceFile() and void *GetSource (void *mf, constchar *sourceName) functions. Both of these functions return a pointer to a cmSourceFile as a void *return value. This pointer can then be passed into other functions that manip­ulate cmSourceFiles such as SourceFileGetProperty() or SourceFileSetProperty().


Notes