qiBuild documentation

CMake coding guide


  • Keep the length of the line below 80 characters when possible, and when it does not hurt readability, and below 100 characters at any case.

  • Indentation is made with two spaces

  • No trailing whitespace are allowed.

  • Wrap the line at 80-100 characters.

  • Every text file must be pushed using UNIX line endings. (On windows, you are advised to set core.autocrlf to true).

  • Please use a spell checker when you write comments. Typos in comments are annoying and distractive.

  • Never use old CMake syntax code for loop constructs:

    # NO
    #  ...
    #  ...
    # YES
    #  ...
    # ...
  • Although CMake is rather lenient with case sensitivity, please write every function lower-case, and separate words by underscores:

    #  ...
    #  ...
  • Every function in the public API of qiBuild code (i.e: that could en up in a user cmake code) must start with qi, other should not start with qi (prefer using _qi for example).

  • When writing a convenience function, not to be used outside, start the name with an underscore, if you have a whole bunch of internal functions, put them in a separated file, in the internal subdirectory.

  • The CMakeParseArguments module is very useful, please use it.

  • Please do not use C-like construct for strings spanning on several lines; rather use nice CMake feature for this:

    # NO
    message(STATUS  "This is a very long\n"
                  "message on\n"
                  "several lines\n"
    # YES
    message(STATUS "This is a very long
      message spanning on
      several lines

    See CMake Syntax

  • Use the log functions carefully. The output of CMake must stay minimal (when it gets too long, it is impossible for the user to see if something went wrong)

  • If you run into a CMake warning, never ignore it. Fix your code or file a bug report. (CMake warnings almost always mean there is a nasty bug somewhere)


In the CMake code

  • Every function in the public API must have corresponding documentation. It works a bit like Doxygen, but with the python-sphinx syntax
#! foobar : this function does foo then bar! (small description)
# this is a long description for the function, the function have two
# parameters accept two flags, two params and two groups.
# \argn: a list of optional arguments
# \arg:first_arg the first argument
# \arg:second_arg the second argument
# \param:PARAM1 PARAM1 specify the fooness of the function
# \param:PARAM2 PARAM2 should always be 42
# \group:GROUP1 GROUP1 is a list of project to foo
# \group:GROUP2 This group represent optional projects to pass to bar
function(foobar first_arg second_arg)
  cmake_parse_arguments(ARG "FLAG1;FLAG2" "PARAM1;PARAM2" "GROUP1;GROUP2" ${ARGN})

Note the bang in the first line of the documentation of the function.

The rest is straightforward

this represent a function parameter, the name is the name of the parameter you are documenting.
This represent a boolean value, the flag could be present or not. (see CMakeParseArguments)
indicates a “one-value option” : the keyword must be followed by a value (see CMakeParseArguments)
indicates a “multi-value option” : the keyword will be followed by a list of values (see CMakeParseArguments)

In sphinx

  • Note: if you add a completely new functionality, you may want to add the new functions in a new file. For instance qi_make_coffee in coffee.cmake In this case you have to:
    • add include(qibuild/coffee.cmake) somewhere in qibuild/general.cmake
    • add you file to the list of the documented files in doc/tools/gen_cmake_doc.py
    • and of course adding a tutorial on how to make coffee with qibuild :)

Conditions and Variables

  • Always quote variable that represent a string:

    set(myvar "foo")
    if ("${myvar}" STREQUAL "bar")
    # ...
  • Do not quote variable that are booleans

    set(mybvar ON)
    set(mybvar OFF)
    if (myvar)
    # ...
    # Note that this will NOT produce the
    # expected result:
    if(${myvar}) # bug!
  • When storing paths in variables, do NOT have the CMake variables end up with a slash:

    # YES:
    set(_my_path "path/to/foo")
    set(_my_other_path "${_my_path}/${_my_var}")
    # NO:
    set(my_path "path/to/foo/")
    set(_my_other_path "${_my_path}${_my_var}")   # wrong: this is ugly
    set(_my_other_path "${_my_path}/${_my_var}")  # this is a bug!, see below

If you don’t do this, you may end up with paths containing //. This does not matter much on Linux, but on Windows, this path may be re-converted into native paths (for instance in the .bat generated by CMake), so you end up with \\ in the path name on Windows, which is the notation for shared folders ...

  • Declaring a list:

    # declare an empty list:
  • Declaring and initializing a list at the same time:

    # a list with 3 items:
    set(mylist item1 "a second item" item3)
      "a second item"
  • Always use list(APPEND) to append to a list:

    list(APPEND mylist "one item")
  • Always quote string when comparing string in a if:

    set(myvar "test")
    if ("${myvar}" STREQUAL "test")
    # ...
  • Do not use “empty” vars:

    # YES:
    qi_create_bin(bar bar.cpp)
    set(_deps baz)
    if (WITH_FOO)
      list(APPENDS _deps FOO)
    qi_use_lib(bar ${_deps})
    # NO:
      set(_foo FOO)
    qi_use_lib(bar baz ${_foo})

    This is confusing and does not save that much lines. Actually the best solution is:

    qi_create_bin(bar bar.cpp)
    qi_use_lib(bar baz)
      qi_use_lib(bar FOO)
  • Always use if(DEFINED varname) to check if a variable is set:

    if (DEFINED myvar)
    #  ...
  • Do not quote variables that CMake expects to be a list:

    set(_foo_args "--foo" "--bar")
    # YES:
    execute_process(COMMAND foo ${_foo_args})
    # NO:
    execute_process(COMMAND foo "${_foo_args}")

In the second line, since you have quoted the list, you are calling foo with one argument, (“–foo –bar”).

  • When you need a function to return a result, use:

    function(compute_stuff arg res)
      # Store something in _result using ${arg}
      set(${res} ${_result} PARENT_SCOPE)
    compute_stuff(my_arg result)
    # NOT set(res ... PARENT_SCOPE)

Common mistakes

  • A very common mistake is to use something like:

    set(_my_out ${CMAKE_BINARY_DIR}/sdk)
    This will work fine most of the time, but :
    • qibuild users may have chosen a unique sdk dir
    • they also may have chose a unique build directory (useful for eclipse, for instance)

    so please use QI_SDK_DIR instead

  • Do not set CMAKE_CXX_FLAGS:

    # This will break cross-compilation
    set(CMAKE_CXX_FLAGS "-DFOO=42")
    # use:
    # or, better, set the compile flags
    # only when necessary:
    # (this will save compile time when you change the define!)
  • Do not set CMAKE_FIND_ROOT_PATH:

    # This will break finding packages in the toolchain:
    set(CMAKE_FIND_ROOT_PATH "/path/to/something")
    # Use this instead:
    # (create an empty list if CMAKE_FIND_ROOT_PATH does not exist)
    list(APPEND CMAKE_FIND_ROOT_PATH "/path/to/something")
  • Do not set CMAKE_MODULE_PATH:

    # This will break finding the qibuild framework
    #  include (qibuild/general) will no longer work
    set (CMAKE_MODULE_PATH "/path/to/something")
    # Use this instead:
    # (create an empty list if CMAKE_FIND_ROOT_PATH does not exist)
    list(APPEND CMAKE_MODULE_PATH "/path/to/something")
  • Do not use ${PROJECT_NAME}, or ${CMAKE_PROJECT_NAME}, especially when not at the top CMakeLists.txt:

    # YES:
    qi_create_lib(foo foo.cpp)
    qi_use_lib(foo BAR BAZ)
    # NO:
    qi_create_lib(${PROJECT_NAME} foo.cpp)
    qi_use_lib(${PROJECT_NAME} BAR BAZ)

The small duplication (writing the name of the target foo 3 times) is OK, because:

  • otherwise you need to scroll to the top of the file to find out the name of the library
  • PROJECT_NAME is something that ends up in the IDE, so it’s not a target name
  • calling project() causes the toolchain file to be parsed again for no good reason
  • there’s often more than one lib per project