[CMake] How to build and link Externa Project with exported target

[Luigi Calori]

[...]
> He's not downloading with CMake, he just told git that in this directory is a submodule. The user has then to fetch it himself with one command:
>
> http://git.wiki.kernel.org/index.php/GitSubmoduleTutorial
>   
Sorry ... no git expert at all.... my point was just to hack  
ExternalProject_add to let specify different download ways, apart from 
URL, SVN_REPOSITORY and mine BZR_REPOSITORY .... would like to have also 
GIT_REPOSITORY and HG_REPOSITORY ....
> [...]
>   
>>> 2) also build your main project wit a ExternalProject_Add and drive the whole thing from a "master-CMakeLists.txt"
>>>
>>> The first one is probably easier to set up, but requires you to guess the installation names and paths correctly. The second option requires you to restructure your whole build system and adds considerable complexity due to the communication between your master-project and the external projects. For this communication I'd try the following:
>>>
>>> - In the master project do all the feature-detection and setting of cache variables (such as options etc)
>>> - Write a cache-initializer script to the binary tree
>>> - Do all the ExternalProject_Add calls and specify the cache-initializer script with the -C option in CMAKE_ARGS
>>>  
>>>       
>> This suggestion is really interesting:    the purpouse is to let any config  options in the "master"  projects to be passed to the "slaves"?
>> Have you any examples?
>>     
>
> You could pass every single cache variable using -D in CMAKE_ARGS, but that becomes pretty tedious very soon... Currently I don't have any example, sorry.
>   
from http://www.cmake.org/cmake/help/cmake-2-8-docs.html#opt:-Cinitial-cache
seems that what  you are suggesting is not  passing the master cache to 
the slaves, but to build a selection of variables and build a script in 
a different format....
Do you suggesto to pass the same script to ALL the subprojects? I' ll 
try to find some examples of that usage.....seems useful
>   
>> I came up with a schema like 2:
>> any project is built as external, dependencies are resolved by  ExternalProject_Add and I have used CMAKE_ARGS to communicate settings:
>> As most of cmake projects were based on FindXXX stuff for finding deps, I have overridden the necessary modules in order to make the projects find the good components at configure time.
>>     
>
> That's what you could do with the cache-initialization script, no need to override the FindXXX modules. If more than one of the external projects use the same FindXXX module, do it in the master and then communicate the results using the initializer script. This way the user can easily override the results during the configure-step and get consistent results in all the sub-projects. Only thing that has me worried is when the user aborts the build, changes the settings in the master cache and then restarts the build. You'd need to somehow set up correct dependencies...
>   
If the  cache-initialization script is single, then would be good that 
all the slave projects file_depends on that.... but do not know how to 
set a file dependency on the configure step of ExternalProject_Add
> If sub-project B does a FIND_PACKAGE(A REQUIRED) and A is built as another sub-project, then you can specify A_DIR to the B project containing the path to the directory containing its A-config.cmake.
>   
Yes, that would be good.... unfortunately not so many cmake projects 
builds xxx-config.cmake in their build-install process.... It would be 
nice to have good examples of usages:
this could be a  "standard" way of building dependencies throug cmake .

Is this ExternalProject_Add feature really used/developed?   I find it 
really nice but a little scared of weather it will be really supported 
and improved.
I have done some patching on it but not know if there is a group of 
user/developer eventually interested to submit mods to

Thanks
              Luigi