Migrate YCMBootstrap to use CMake's FetchContent

[traversaro]

The YCMBootstrap process has been historically quite error-prone. This PR substitute this process by a more straightforward use of the CMake's FetchContent repo.

With respect to the past, the main difference is the superbuild will not use exactly the same YCM downloaded in robotology-superbuild/robotology/YCM , as a private copy of YCM used only for the robotology-superbuild configuration will be downloaded in build/_deps/ycm-src . However, as the new script still observers the YCM_TAG variable, the version downloaded in build/_deps/ycm-src will be exactly the same version of YCM that will be downloaded in robotology-superbuild/robotology/YCM .

On my Windows machine, this decreases the time spent in the YCM bootstrap to ~5 seconds from ~20 seconds of the previous procedure.

If this new module will work fine, it will be proposed in YCM to fix robotology/ycm-cmake-modules#138 .

[diegoferigo]

The main caveat I found on other projects that bootstrap similarly YCM is that if you use add_subdirectory, all the targets of YCM are added to the the parent project. I saw that you just append the folders you need to the CMAKE_MODULE_PATH variable, and it prevents what I just described.

Furthermore, this PR works fine unless YCM functionality is needed in the top-level CMake file. It does not seem to be the case though. Otherwise, you should have used PARENT_SCOPE when extending CMAKE_MODULE_PATH.

Beyond this clarification, I don't have any objections :)

[traversaro]

Furthermore, this PR works fine unless YCM functionality is needed in the top-level CMake file. It does not seem to be the case though. Otherwise, you should have used PARENT_SCOPE when extending CMAKE_MODULE_PATH.

The boostrap file is already included in the top-level CMake file, so I don't think it is need to specify the PARENT_SCOPE when extending CMAKE_MODULE_PATH .

[traversaro]

For some reason that I don't get the CI is not starting on this PR.

[traversaro]

On my Windows machine, this decreases the time spent in the YCM bootstrap to ~5 seconds from ~20 seconds of the previous procedure.

This seems to be confirmed in the CI tests, in which the configuration time on Windows passed from ~1:20 seconds to ~55 seconds.

[traversaro]

It turns out that consuming YCMEPHelper directly is quite difficult, as YCMEpHelper has a check for YCM_FOUND in its code, and if YCM_FOUND is not found, it uses bootstrap-like behaviour (downloading stuff over the internet): https://github.com/robotology/ycm/blob/master/modules/YCMEPHelper.cmake#L106 . Setting YCM_FOUND for fooling that check is not feasible, as it will prevent the superbuild to correctly setup the use of YCM as a subproject. So it is necessary first of all to modify YCMEPHelper .

[traversaro]

Closing this PR as it is not possible to achieve this without modifying YCM itself.

[traversaro]

Setting YCM_FOUND for fooling that check is not feasible, as it will prevent the superbuild to correctly setup the use of YCM as a subproject. So it is necessary first of all to modify YCMEPHelper .

I actually did that in #1078 and it is working fine.