How to handle a transitive dependency conflict using Git submodules and CMake?
We have a number of Git repositories, some containing our own code and some containing slightly modified third-party library code. A simplified dependency graph looks like this:
executable_A
| |
| v
| library_B
| |
v v
library_C
So the executable has two dependencies on library_C , one direct and one transitive. I am hoping to tie this all together using Git submodules and CMake, so a simplified directory structure looks like this:
executable_A/
CMakeListst.txt
library_B/
CMakeLists.txt
library_C/
CMakeLists.txt
library_C/
CMakeLists.txt
As you can see, the library_C repository is included as a submodule twice. Let's assume that both submodules are pointing at the same commit (any ideas about how to enforce that would be welcome, but are not the topic of this question).
We're using add_subdirectory , target_link_libraries and target_include_directories to manage these interdependencies. Pretty standard.
The problem is that CMake doesn't like it if you create a target with the same name twice, so it complains:
CMake Error at library_C/CMakeLists.txt:13 (add_library):
add_library cannot create target "library_C" because another target with the same name already exists. The existing target is a static library created in source directory ".../library_B/library_C".
See documentation for policy CMP0002 for more details.
I'd rather not remove the direct dependency of executable_A on library_C , because the fact that it is pulled in via library_B is an implementation detail of library_B that should not be relied on. Moreover, this approach will break down as soon as we add another dependency like executable_A --> library_D --> library_C .
(This question is the closest I could find, but is a bit more general and remains unanswered anyway.)
There are several approaches for detect and discard inclusion of the project, which has already be included in some other parts of the main project.
Check project's target existence
The simplest pattern for single inclusion of subproject is checking existence of some subproject's target:
# When include 'C' subproject
if(NOT TARGET library_C)
add_subdirectory(C)
endif()
(Here we assume that project C defines target library_C .)
After such conditional inclusion all subproject's targets and functions will be immediately available for the caller with garantee .
It is better to use this pattern in all places (in executable_A and library_B ). Such a way changing order of library_B and library_C in executable_A doesn't break correctness.
This pattern can be reworked for use by subproject itself:
# At the beginning of 'C' project
cmake_minimum_required(...)
if(TARGET library_C)
return() # The project has already been built.
endif()
project(C)
...
Check project existence
When a project is created, CMake defines several variables for it, and <PROJECT-NAME>_BINARY_DIR is among them. Note, that this variable is cached, so when cmake is called the second time (eg if some of CMakeLists.txt has been changed), the variable exists at the very beginning.
# When include 'C' subproject
if(NOT C_BINARY_DIR # Check that the subproject has never been included
OR C_BINARY_DIR STREQUAL "${CMAKE_CURRENT_BINARY_DIR}/C" # Or has been included by us.
)
add_subdirectory(C)
endif()
This pattern can be reworked for use by subproject itself:
# At the beginning of 'C' project
cmake_minimum_required(...)
if(NOT C_BINARY_DIR # Check that the project has never been created
OR C_BINARY_DIR STREQUAL "${CMAKE_CURRENT_BINARY_DIR}" # Or has been created by us.
project(C)
else()
return() # The project has already been built
endif()