Conda problems

[johnwason]

This thread is for discussion problems with conda instead of creating issues in the wrong place.

[johnwason]

vtk pinning is incorrect in pcl feedstock. The pcl feedstock uses vtk==9.1.0, but the main pinning is using vtk==9.0.1. This is causing a version conflict.

https://github.com/conda-forge/pcl-feedstock/blob/4f599766cf9278362bb7e4726bc5b65de78c61b4/.ci_support/win_64_.yaml#L25

https://github.com/conda-forge/conda-forge-pinning-feedstock/blob/a8ff94eb129f599b79863135343c0bab52b54fb4/recipe/conda_build_config.yaml#L762

[Tobias-Fischer]

vtk is currently being migrated to 9.1.0 - see https://conda-forge.org/status/

Nearly all packages, including PCL, have been migrated already, so I suggest to update your packages to 9.1.0, too.

[johnwason]

I was able to remove the vtk dependency from my meta.yaml and avoid this problem. I still hope to create a variant of PCL that doesn't have dependencies on gui frameworks.

[johnwason]

Taskflow is installing its cmake file to lib64/cmake instead of lib/cmake on Linux. This is preventing cmake from finding the config file. I had to manually specify the location in by build.sh file for it to be found correctly.

[traversaro]

Interesting, in theory it should not even be necessary to fix it in the build.sh, since CMake 3.22 this should be solved by CMake upstream, see conda/conda-build#244 (comment) and https://gitlab.kitware.com/cmake/cmake/-/merge_requests/6512 .

[johnwason]

JsonCPP has a bug that if find_package(jsoncpp) is called more than once, it fails. I have had to fence the search with the following:

if(NOT TARGET JsonCpp::JsonCpp)
find_package(jsoncpp REQUIRED)
endif()

I left a suggesting on the jsoncpp repo, of how to fix the problem:

open-source-parsers/jsoncpp#1356 (comment)

[johnwason]

Selecting vs2019 compiler in conda_build_config.yaml is using the windows-latest github action instance type in the conda_build.yml workflow when generated with conda-smithy. This instance type does not have vs2019 installed. The windows-2019 instance must be used. I manually modified the file to use the correct instance.

https://github.com/tesseract-robotics-packaging/ros-industrial-cmake-boilerplate-feedstock/blob/f972385901fc04b48b7463934562412077b30ca6/recipe/conda_build_config.yaml#L1

https://github.com/tesseract-robotics-packaging/ros-industrial-cmake-boilerplate-feedstock/blob/master/.github/workflows/conda-build.yml

[johnwason]

I completed workflows to build each individual tesseract package as its own conda package instead of a single "superpack", but the resulting packages are failing tests with segfaults and other assorted errors. The updated superpack however still works. I have run out of time to work on this, so I will have to keep using the superpack.

Individual package feedstocks:

https://github.com/tesseract-robotics-packaging/ros-industrial-cmake-boilerplate-feedstock

https://github.com/tesseract-robotics-packaging/opw-kinematics-feedstock

https://github.com/tesseract-robotics-packaging/ifopt-feedstock

https://github.com/tesseract-robotics-packaging/descartes-light-feedstock

https://github.com/tesseract-robotics-packaging/tesseract-robotics-feedstock

https://github.com/tesseract-robotics-packaging/trajopt-feedstock

https://github.com/tesseract-robotics-packaging/tesseract-robotics-planning-feedstock

https://github.com/tesseract-robotics-packaging/tesseract-robotics-python-feedstock

Failing tests:

https://github.com/tesseract-robotics-packaging/tesseract-robotics-python-feedstock/actions/runs/2032458601

Superpack feedstock:

https://github.com/johnwason/tesseract-robotics-superpack-feedstock