Do we need both Release and Debug builds with this setting?

No, running it for Debug only would be ideal for the current setting. I just could not find a elegant way to exclude the Release build

And I think it is going to be too much if we double the amount of tests. I am here assuming that successful tests with the FP32 also guarantee the success with the FP64, hopefully

I just could not find a elegant way to exclude the Release build

The easy way to do this would be using the matrix.include method, as seen in https://github.com/acts-project/traccc/blob/main/.github/workflows/builds.yml#L60.

In short, you'd add something like this:

        include:
          - platform:
            - name: CUDA
              container: ghcr.io/acts-project/ubuntu2404_cuda:56
              cxx_standard: "20"
              options: -DTRACCC_CUSTOM_SCALARTYPE=double -DDETRAY_CUSTOM_SCALARTYPE=double -DTRACCC_BUILD_CUDA=TRUE -DCMAKE_CUDA_FLAGS="-std=c++20"
              run_tests: false
            build: Release

@krasznaa Is this on the github or the gitlab side?

Yepp, Pandora's box... 😦

Because of the issues, it could still make sense to set up double versions of pretty much all of the device backends. Otherwise there will always be some code that our tests don't cover. 🤔

It doesn't need to be this PR though. You'll have plenty of work with just making the core (and CUDA) code work correctly like this. We can take care of fixing up everything else in follow-up PRs.

While making this work, I also removed the awkward scalar type dependency in the seeding_config and finding_config