Release 3.2 Dockerfile.base numpy version mismatch

[raytsou]

On version 3.2 :

docker/Dockerfile.base:174 numpy is installed with version <2
docker/Dockerfile.base:211 cupy-cuda12x is installed with no version limit.

cupy-cuda12x v14 was released Feb 20 and uses numpy >= 2, which installs a newer version and is causing dependency mismatches downstream.

Particulary, when I try to colcon build --packages-up-to isaac_ros_managed_nitros isaac_ros_nitros_image_type isaac_ros_gxf, I get:

...
CMake Error in CMakeLists.txt:
  Imported target
  "isaac_ros_tensor_list_interfaces::isaac_ros_tensor_list_interfaces__rosidl_generator_py"
  includes non-existent path
    "/usr/local/lib/python3.10/dist-packages/numpy/core/include"
  in its INTERFACE_INCLUDE_DIRECTORIES.  Possible reasons include:
  * The path was deleted, renamed, or moved to another location.
  * An install or uninstall procedure did not complete successfully.
  * The installation package was faulty and references files it does not
  provide.
CMake Generate step failed.  Build files cannot be regenerated correctly.
...

Solution here should be to hold cupy-cuda12x<14

Workaround for now is to downgrade numpy back to a version <2 before colcon building isaac ros nitros related packages.

[Exokern]

Thanks for the clear repro + line pointers.

This looks like an implicit upgrade path: the Dockerfile pins NumPy <2, but leaves cupy-cuda12x unpinned. If CuPy v14+ gets pulled, it can upgrade NumPy to >=2, and downstream builds can still end up referencing the old NumPy include path (/usr/local/lib/python3.10/dist-packages/numpy/core/include) and fail as you saw.

A minimal fix would be to pin cupy-cuda12x<14 (or otherwise co-pin NumPy/CuPy to a compatible pair) in docker/Dockerfile.base around the cupy-cuda12x install step.

If useful, I can send a tiny PR that pins cupy-cuda12x<14 (or equivalent) and adds a quick sanity check such as python -c "import numpy; print(numpy.__version__, numpy.get_include())" to prevent silent breakage.