I agree that manually keeping track of transitive dependencies is a very bad
and error-prone idea.
The ideal solution is the separate repositories, but I don’t think that’s
really feasible.
The multi-root setup would work, but the ugly part of it is requiring an
“app*_root” directory for each application, which is needed to be able to
create a solution/makefile with only the application’s dependencies. When
compiling an app I don’t want to be compiling targets it does not depend on,
which is what would happen if I were to use the app*_root directory directly.
So in your example:
-top/
- app1 /
- CMakeLists.txt
- app1_root/
- CMakeLists.txt
- app2/
- CMakeLists.txt
- app2_root/
- CMakeLists.txt
- app3/
- CMakeLists.txt
- app3_root/
- CMakeLists.txt
- components/
- CMakeLists.txt
- libA/
- libB/
- libC/
And developers would have to CMake the app*_root directories and then open
solutions / run makefiles generated for the individual apps. Not terrible, but
definitely not intuitive. But thank you for sharing that, I haven’t thought of
that 😊
From: Avraham Shukron <avraham.shuk...@gmail.com>
Sent: Friday, September 27, 2019 5:23 PM
To: Marek Vojtko <marek.voj...@oxidegames.com>
Cc: cmake@cmake.org
Subject: Re: [CMake] Project Structure and add_subdirectory()
I think option 2 (keeping track on transitive dependencies) should be out of
the question.
It is bound to fail.
Superbuild setup makes sense where all the components are part of a bigger
whole.
But if each application may depend on a different "flavor" of a dependency
library, I think it should be reflected in your project structure, E.g each
application in a separate repository, with all the dependencies as submodules.
alternatively, you can create a multi-root setup, where you have a
project-root-directory for each application, with a top-level CMakeLists that
adds all the necessary subdirectories.
This way, you get the best of all worlds:
1. Each component's source code is kept only once, and changes are immediately
affecting all products.
2. You can compile each application independently from one another.
Consider the following structure:
-top/
- app1_root/
- CMakeLists.txt
- app2_root/
- CMakeLists.txt
- app3_root/
- CMakeLists.txt
- components/
- CMakeLists.txt
- libA/
- libB/
- libC/
The only issue I can think of is that in each application's top-level
CMakeLists.txt you have to add_subdirectory(../components) and the ".." is kind
of ugly.
Other than that I think it should work
On Fri, Sep 27, 2019 at 7:00 PM Marek Vojtko
<marek.voj...@oxidegames.com<mailto:marek.voj...@oxidegames.com>> wrote:
Using add_subdirectory() in a "superbuild" setup is straightforward: Create a
root CMakeLists.txt which calls add_subdirectory() on all directories of your
project and each target can then use target_link_libraries() without worrying
where the target came from.
Unfortunately the "superbuild" setup also means that each target is compiled
exactly once, which means that you cannot change a target's build environment
per-application (e.g. preprocessor definitions that change the sizes of stack
arrays).
If I need per-application control of its dependencies' build environment I can
have each application's CMakeLists.txt call add_subdirectory() for all of its
dependencies. However, that means that each application now has to know all of
its *transitive* dependencies and the application's CMakeLists.txt has to be
kept up-to-date with any changes in its dependency tree, no matter how deep. So
if AppA depends on LibA, which depends on LibB, which depends on LibC. Not only
does AppA's CMakeLists.txt have to call add_subdirectory() on LibA, LibB, and
LibC, but if LibC is modified to depend on LibX, now AppA's CMakeLists.txt has
to also be modified to call add_subdirectory() on LibX.
Having each target call add_subdirectory() on its own dependencies seems silly,
because that would create an insane number of duplicated targets. If LibA
depends on LibB and LibC, and both LibB and LibC depend on LibD, this approach
would result in both LibB and LibC calling add_subdirectory() on LibD, creating
the target twice, which would likely not compile.
Are these my only two options? Either:
- use a superbuild to have CMake handle my transitive dependencies, but give up
per-application build environment changes, or
- track all transitive dependencies manually in each application's
CMakeLists.txt, but retain the ability to change the build environment per
application.
--
Powered by www.kitware.com<http://www.kitware.com>
Please keep messages on-topic and check the CMake FAQ at:
http://www.cmake.org/Wiki/CMake_FAQ
Kitware offers various services to support the CMake community. For more
information on each offering, please visit:
CMake Support: http://cmake.org/cmake/help/support.html
CMake Consulting: http://cmake.org/cmake/help/consulting.html
CMake Training Courses: http://cmake.org/cmake/help/training.html
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Follow this link to subscribe/unsubscribe:
https://cmake.org/mailman/listinfo/cmake
--
Powered by www.kitware.com
Please keep messages on-topic and check the CMake FAQ at:
http://www.cmake.org/Wiki/CMake_FAQ
Kitware offers various services to support the CMake community. For more
information on each offering, please visit:
CMake Support: http://cmake.org/cmake/help/support.html
CMake Consulting: http://cmake.org/cmake/help/consulting.html
CMake Training Courses: http://cmake.org/cmake/help/training.html
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Follow this link to subscribe/unsubscribe:
https://cmake.org/mailman/listinfo/cmake
