Change in behaviour of robotology-superbuild update-all target since August 2024 [Update: the change of behaviour was reverted in mid August 2024]

[traversaro]

Hello @robotology/everyone,

Before August 2024, the robotology-superbuild raised an error if you called make update-all or ninja update-all and there was repo with local modification with YCM_EP_DEVEL_MODE_<package> set to OFF. After August 2024, instead the robotology-superbuild will silently discard the local modifications.

To avoid losing data, never call the update-all target if you have local modifications in a package and you did not set YCM_EP_DEVEL_MODE_<package> to ON for that package.

For more context, see robotology/robotology-superbuild#1674 .

UPDATE 2024/08/14: After some issue emerged, the behaviour of the superbuild has been reverted to the one before August 2024, see https://github.com/orgs/robotology/discussions/712#discussioncomment-10338765 for more info.

[giotherobot]

This might be a problem if using ninja as generator, see: robotology/robotology-superbuild#83 (comment)

[giotherobot]

The discarded modifications are only the local ones not committed to any branch? Or also the branch is reset?

[giotherobot]

I just tested (without YCM_EP_DEVEL_MODE) and it also removes custom remotes (eg. personal forks) to the sub repos. This might have been always the case though.

[traversaro]

Thanks, I will look in the Ninja problem. In the meanwhile, you can revert to the old behaviour by manually setting YCM_USE_CMAKE_NEXT to ON in the superbuild.

[traversaro]

The Ninja problem was fixed by robotology/ycm-cmake-modules#456, can you try if you have more problems?

[giotherobot]

Seems to work! Thanks!

[traversaro]

Hello @robotology/everyone , thanks to some bug report by @giotherobot, we noticed that the change was causing repos to be removed even if one user created a new build of the superbuild, even if the YCM_EP_DEVEL_MODE_<package> option for the repo was enabled. This was highly unexpected and dangerous, so the decision was to revert the superbuild behaviour to the one existing before August 2024.

However, this was not done as a revert, but via a refactor (done in https://github.com/robotology/ycm-cmake-modules/releases/tag/v0.16.7) that permitted to have the classic superbuild behaviour even using the latest upstream CMake's ExternalProject module. For this reason, there could be other tricky aspects or regressions, if you found any feel free to open issus in https://github.com/robotology/robotology-superbuild/issues to report them, thanks a lot in advance!

[traversaro]

Hello @robotology/everyone ,
thanks to @xela-95, we found another tricky issue: robotology/robotology-superbuild#1700 . In a nutshell, after a successful build, subsequent build of the superbuild were not actually re-building, even if there were change in the underlying projects. This was fixed in robotology/ycm-cmake-modules#462 . The fixed version is going to be released as part of the upcoming distro 2024.08.0, but if you encounter any unexpected problem or behaviour, please report them at https://github.com/robotology/robotology-superbuild/issues, thanks!