Open j-rivero opened 1 year ago
Expected behavior: testing examples should ideally not require to install the library into the system.
The way that this was implemented was to not require installation, but instead use the libraries out of the build folder. I will take a look and see what is happening here.
Okay, after taking a look, it's somewhat related to this conversation: https://github.com/gazebosim/gz-cmake/pull/301#discussion_r1163354177
The issue is that we can't really have a self-contained examples/CMakeLists.txt
file without a full installation somewhere. Note that the installation doesn't have to be a system installation, but generally assumes that colcon
was the way that the package was built (there was cmake && make && make install
before make test
). This probably also explains why we didn't catch it in any of our CI before the deb packaging job.
In the past, we had worked around this by making a FAKE_INSTALL
target, but this didn't work on Windows for one reason or another.
There may be another work around here, but the crux of the issue is that our gz_find_package
logic doesn't work until the package is installed and we require our examples to make use of that flow.
CC: @azeey
Not sure if this is ready to be closed yet.
Ah, sorry, github linked issues.
While testing https://github.com/gazebo-release/gz-math7-release/pull/19 I found that gz-math7 nightlies are probably broken after https://github.com/gazebosim/gz-math/pull/482. The CI is installing the files into the system before running the test suite (something probably that we should change) so the error has skipped from our CI mechanisms.
Environment
Description
Steps to reproduce
Output