[CMake] Makefile doesn't automatically create executable path as Ninja

[Matthew Woehlke]

On 2013-09-03 14:18, Andreas Pakulat wrote:
> On Tue, Sep 3, 2013 at 7:41 PM, Felipe Menezes Machado wrote:
>> On Tue, Sep 3, 2013 at 2:33 PM, Matthew Woehlke wrote:
>>> On 2013-09-03 13:21, Felipe Menezes Machado wrote:
>>>> I'm having trouble understanding this behavior: what happens if the
>>>> directory of the executable doesn't exist?
>>>>
>>>> I created a minimum example CMakeLists.txt (where foo.cpp is just a hello
>>>> world):
>>>>
>>>> add_executable(test/test.bin foo.cpp)
>>>>
>>>> If the "test" directory doesn't exist, should the build system create it?
>>>> Does that depend on the build system?
>>>
>>> Yes; ninja has a feature where it creates any non-existing directories
>>> that are used by any outputs. At least make does not have a similar
>>> feature. I don't know about other generators (e.g. xcode, msvc).
>>>
>>> I suppose CMake could be taught to generate make
>>> rules to emulate ninja's behavior.
>>>
>>> Whether or not such a thing should be done is an interesting question.
>>> (Especially as it probably should be done for all generators, it at all.)
>>
>> It would at least be a very interesting optional feature. Like something
>> you specify in your CMakeLists.txt file, that every output directory should
>> be created.
>
> Well, IMO you're somewhat misusing cmake here, the add_executable and
> add_library commands are intended to get the _names_ of targets. Not some
> relative path to the dedicated output file. If you want to influence where
> the binary goes you can do so with the corresponding target properties (see
> RUNTIME_OUTPUT_DIRECTORY and OUTPUT_NAME).
>
> So supporting relative paths as target name would pose a second way to
> achieve the same thing which may be considered uneeded and possibly
> confusing.

Well, sure, that too :-). However the problem still exists for more 
usual ways of determining output paths. (I've run into it with 
add_custom_command, for example.)

-- 
Matthew