Tunnel two bend tests fail

[Jochem Snuverink]

ctest -R tunnel-several-bends-two-way gives (for Geant4.10.02-patch-02):

The following tests FAILED:
    290 - tunnel-several-bends-two-way-elliptical (OTHER_FAULT)
    291 - tunnel-several-bends-two-way-rectangular (OTHER_FAULT)
    293 - tunnel-several-bends-two-way-square (OTHER_FAULT)
    294 - tunnel-several-bends-two-way-rectaboveground (OTHER_FAULT)

Verbose output for rectangular case:

291: Checking overlaps for volume tunnel_2_floor_pv ... OK! 
291: 
291: -------- EEEE ------- G4Exception-START -------- EEEE -------
291: *** G4Exception : GeomSolids0002
291:       issued by : G4CutTubs::G4CutTubs()
291: Invalid Low or High Normal to Z plane; Crossing Cutted Planes.
291: Invalid Norm to Z plane ((0.0526073,0,-0.998615) and (0.0526073,0,0.998615)) in solid: tunnel_3_face_intersection_solid
291: *** Fatal Exception *** core dump ***
291: -------- EEEE -------- G4Exception-END --------- EEEE -------
291: 
291: 
291: *** G4Exception: Aborting execution ***

[Laurie Nevay]

I've tested with Geant4.10.2.p03 (as that's what I have in the 4.10.2 series) without problem on both mac osX and SLC6.8. On SLC6.8 with Geant4.10.1.p04 I can reproduce this problem. Geant4.10.1.p03 on mac (again, builds available) doesn't show a problem here.

I know the implementation of these checks in the G4CutTubs solid has been revised many times in Geant.10.1, 2, 3, 4.

Up until recently, the tunnel building algorithm that decides how long to make sections of tunnel was flawed and always built one section twice. This has recently been fixed, but the algorithm requires refinement. In this case, it's making a cylinder with angled faces that is very close to having touch faces on either side. We can generally choose a better selection of tunnel sections.

[Laurie Nevay]

This has been resolved by fixing the tunnel building algorithm (wrong iterators for end piece). The algorithm has also been improved to be more robust and treat tilted dipoles correctly.

All tests now pass with Geant4.10.1 and 4.10.4. I will see the other Geant4 version test results tomorrow from the test server, but I'm happy the problem is fixed. Attached is a screenshot of the same model reported as failing that now has different sections of tunnel.

[Jochem Snuverink]

I confirm the tests pass locally with Geant4.10.2