#61 Bug in shearing boundary conditions?

[Anonymous]

Originally created by: anders.j...@gmail.com
Originally owned by: anders.j...@gmail.com

The way I see it there is a bug in the shearing boundary conditions. I discovered this because the code crashes when shock is initialized to
impossible (i.e. 1e37 and a bit) when using shear-periodic boundary condition.

The main problem is that when we set shear-periodic boundary conditions we interpolate the data from only two processors at the other side of the box. In the example below we want to set the crosses and realise that we need to match them with data from the dots:

   -----------------
   |              .|
   |              .|
   |              .|
   -----------------
   |              .|
   |              .|
   |              .|
   -----------------
  x|               |
  x|               |
  x|               |
   -----------------
   |               |
   |               |
   |               |
   -----------------

Unfortunately in general the crosses will need data from three CPUs rather than two, because of the interpolation. Therefore the current implementation lets the dots take the y and z boundary conditions along.

However, this is where the problem lies: we set x boundary conditions *before* y boundary conditions. Therefore the y-boundaries are one time-step behind when we use them for the shearing x-boundaries. And if we are at time-step zero and initialize the data to impossible, then the code crashes.

[Anonymous]

Originally posted by: anders.j...@gmail.com

Anders should add a sample which crashes because of this bug!

[Anonymous]

Originally posted by: mcnal...@gmail.com

This should be fixed by change [r22634]