Andreas Salzburger:
Not undefined. If a sensitive surface is outside the layer build, the well defined behaviour is that the geometry is blind to this. The approach surface is the starting point for finding the enxt surface. Everything seen is expected.
How to test:
If shifts are small enough, that we can put envelopes, you could expand them. This is probably not possible, because the shift seems significant larger.
Simple way for testing: Try to put all layers into a single layer description. The layer would be overpopulated and innermost/outermost surface can be defined. layerbinning can be put into a single bin. The navigation will be really slow because it tries to find every single surface. This shows, if the issue can be partially solved with that (except for bending effects).
Currently, there is no easy shift with the geometry context. (not implemented yet). In the future this will come hopefully. Alternatively try with submodules. Then with 3 concentrical layers inside the subvolume can be tried. Navigation again slow. In Gen3 this will be supported more easily.
Paul:
Restrictions from navigation: We will have this kind of problem always. Maybe a solution for Gen3. Classical collission detection for the boundaries of the layers. If the alignment is almost static, this might be done once per job. Not possible for fast changes in the alignment.
Andreas Stefl:
There can be subvolumes, but not possible yet to navigate into subvolumes.
Andreas Salzburger:
Maybe we can manage a quick trial implementation for sPHENIX.
It seems like only in a shifted subvolume the layer can be constructed. That would require to build the geometry with alignment context. This was time in ATLAS a long time ago. Looking at the size of the shift, this might improve it but not fully solve it. Because we cannot make concentric layers yet. Surfaces can only found between entrance and exit of layers. Is there a way, that we can easily shift the setup into a concentric one? (Paul:) If envelope can be enlarged, this step can be skipped.
Also we want to come up with a longterm implementation to properly fix that.
Andras Stefl:
passive surfaces should be able to be aligned the same as sensitive, but never tried it out. The setup is planned for micrometer/mm scale. Since we have shifts of cm order, maybe changing the detector description might be more reasonable.
Joe:
Longterm chaning the description might be a good longterm idea.
In TGeometry some layers are built with an offset.
Andreas Salzburger:
Implement in geometry and then get a layer clash because cylinders are tried to be centered around 0. To get an easy way out: Starting from nominal geomtry. Then run the geometry as it is and adopt the TGeo description, maybe with a flag. It will also be a great testbed for Gen3 and usefull for all sides.
