ref: remove b-field from detector

[niermann999]

This PR removes the bfield from the detector and decouples detray core from covfie. The management of the bfield (building/moving to device) is done in the client code (e.g. the detray unittests etc).

Note: this does not remove the templating/need for explicit template instantiation on the propagation kernels in detray, it simply allows to find a different solution in traccc

[beomki-yeo]

I tried to solve the warning of calling a __host__ function from a __host__ __device__ function is not allowed but I really cannot locate what is causing the problem. I guess it is on vecmem::vector being used in device code - but I could not find such case. @niermann999 Any thoughts?

[niermann999]

I tried to solve the warning of calling a __host__ function from a __host__ __device__ function is not allowed but I really cannot locate what is causing the problem. I guess it is on vecmem::vector being used in device code - but I could not find such case. @niermann999 Any thoughts?

No, I also tried to find it and failed. Additionally, it is complaining about some variables that are set but not referenced in the toy geometry creation, which is clearly incorrect. I am thinking that I have redo the whole thing, but starting with replacing the bfield before removing the template and trying to go step by step (as much as that is possible with this change)

[beomki-yeo]

Additionally, it is complaining about some variables that are set but not referenced in the toy geometry creation

That's possibly a nvcc bug or at least a warning that can be ignored.
The problem was that create_toy_geometry.hpp is included in header file which is also compiled by nvcc.
Header file usually defines detector_host_type so lots of functions should be defined for vecmem::vector which is not allowed by Cuda compiler

[niermann999]

Ok, thanks for having a look. The warning about the unreferenced variables should vanish, too, since it was in create_toy_geometry as well. Otherwise, we need to disable it, since every warning will block the CI

[niermann999]

Odd, that these warnings did not come up before...

[beomki-yeo]

That's odd indeed. I also don't know what triggered this error.

[beomki-yeo]

I could not fix the narrowing conversion warning and ended up with disabling detray propagation tutorial for the moment.

Because I need to update detray version and move on to traccc work - @niermann999 Could you kindly accept this?

[niermann999]

I could not fix the narrowing conversion warning and ended up with disabling detray propagation tutorial for the moment.

Reverted back to using a const field in the tutorials, that fixed the issue for me (at least locally). Then let's go with this

[beomki-yeo]

Yeah but let's not spend too much time for something we don't fully understand. We don't know if the covfie::cuda warning is from detray or covfie itself

[niermann999]

It seems to come from covfie, it happens during the compilation of the detray bfield header, even if the inhomogeneous type and the bfield reader are not used. But OK, if you need this asap, I will fix it in the next PR

[beomki-yeo]

I cannot do anything without this. so yes I need this ASAP.

[beomki-yeo]

BTW you will also need to disable detector construction.

I forgot to mention that the compilation fails even without covife::cuda

[beomki-yeo]

@niermann999 I removed the covfie library from detector construction

[niermann999]

Ok, I am trying to finish the rest in #568 . Let's merge this

[niermann999]

I think you need to remove the bfield header in the detector construction, but let's just comment the tutorial out for now

[beomki-yeo]

I commented them now.