generate-conda-packages: Use conda-forge-pinning conda_build_config.yaml

[traversaro]

As documented in https://docs.conda.io/projects/conda-build/en/latest/resources/variants.html#creating-conda-build-variant-config-files, the conda_build_config.yaml files can be used to specified the specific version of the libraries used, to build multiple versions of the libraries, or to specific more options.

Until now when generating conda packages we just used our local conda/conda_build_config.yml file, now as explained in #708 we will switch to consider the conda-forge conda_build_config.yaml (mantained at https://github.com/conda-forge/conda-forge-pinning-feedstock) to ensure the maximum compatibility of our binaries with the one compiled by conda-forge.

[traversaro]

Manual invocation of generate-conda-packages without upload running in: https://github.com/robotology/robotology-superbuild/actions/runs/810822307 .

[traversaro]

Manual invocation of generate-conda-packages without upload running in: https://github.com/robotology/robotology-superbuild/actions/runs/810822307 .

This shows that the boost 1.76 regression is indeed fixed (until 1.74 is actually used, see robotology/blocktest#47). However, it seems to be another OpenCV regression, tracked in #711 and probably related to conda-forge/opencv-feedstock#271 . As I don't have time to debug that now, for the time being I will just pin opencv to 4.5.1 in our local conda_build_config file (showing its use).

[traversaro]

Manual invocation of generate-conda-packages without upload running in: https://github.com/robotology/robotology-superbuild/actions/runs/810822307 .

This shows that the boost 1.76 regression is indeed fixed (until 1.74 is actually used, see robotology/blocktest#47). However, it seems to be another OpenCV regression, tracked in #711 and probably related to conda-forge/opencv-feedstock#271 . As I don't have time to debug that now, for the time being I will just pin opencv to 4.5.1 in our local conda_build_config file (showing its use).

CI run after doing those modifications: https://github.com/robotology/robotology-superbuild/actions/runs/813592724 .

[diegoferigo]

LGTM. A curiosity, what happens when the pinnings want to pin different versions of the same package?

[traversaro]

LGTM. A curiosity, what happens when the pinnings want to pin different versions of the same package?

I think the last one passed to conda build . wins as it overwrite the one specified by the earlier file, but I am not sure.

[traversaro]

Manual invocation of generate-conda-packages without upload running in: https://github.com/robotology/robotology-superbuild/actions/runs/810822307 .

This shows that the boost 1.76 regression is indeed fixed (until 1.74 is actually used, see robotology/blocktest#47). However, it seems to be another OpenCV regression, tracked in #711 and probably related to conda-forge/opencv-feedstock#271 . As I don't have time to debug that now, for the time being I will just pin opencv to 4.5.1 in our local conda_build_config file (showing its use).

CI run after doing those modifications: https://github.com/robotology/robotology-superbuild/actions/runs/813592724 .

That job is working fine, and the OpenCV regression is tracked in #711 while the conda:Tags:Default@windows-2019@Release failure is tracked in #712, so I think we can merge this PR are build some conda packages (the weekly build this week was skipped due to this regression).