[ET Trac] [Einstein Toolkit] #797: N13CarpetRegrid211snap_coarseE does not hold after being enforced in cell centering

[Einstein Toolkit]

#797: N13CarpetRegrid211snap_coarseE does not hold after being enforced in cell
centering
---------------------+------------------------------------------------------
  Reporter:  rhaas   |       Owner:  eschnett
      Type:  defect  |      Status:  review  
  Priority:  major   |   Milestone:          
 Component:  Carpet  |     Version:          
Resolution:          |    Keywords:          
---------------------+------------------------------------------------------
Changes (by rhaas):

  * status:  new => review


Comment:

 The issue seems to be an explicit check in expand_for for an empty()
 *this. Since all the other manipulation functions also seem to contain at
 least BBBOX_ASSERTs on emptyness in at least some direction, simply
 removing the test seems like a bad idea. Right I am testing (successfully
 so far!) the following code, which is just the code pieces from the
 various bbbox member functions expanded out.
 {{{#!cpp
 #if(0) // this fails for very thin strips where cotracted_for creates a
 temporary empty box
       snapped |= bb.
         expand(reffact-1 - buffers % reffact).
         contracted_for(cbase).
         expanded_for(base).
         expand(buffers % reffact);
 #else // same math as above but avoid in particular expanded_for's set-to-
 empty
       ivect lower = bb.lower();
       ivect upper = bb.upper();
       ivect stride = bb.stride();
       cout << "aa: " << ibbox(lower,upper,stride) << endl;
       // expand(reffact-1 - buffers % reffact)
       {
         lower -= (reffact-1 - buffers[0] % reffact) * stride;
         upper += (reffact-1 - buffers[1] % reffact) * stride;
       }
       cout << "aa.expand(): " << ibbox(lower,upper,stride) << endl;
       // contracted_for(cbase)
       {
         stride = cbase.stride();
         ivect loff = ((lower - cbase.lower()) % stride + stride) % stride;
         ivect uoff = ((upper - cbase.lower()) % stride + stride) % stride;
         lower += (stride - loff) % stride; // go inwards
         upper -= uoff;
       }
       cout << "aa.expand().contracted_for(): " <<
 ibbox(lower,upper,stride) << endl;
       // expanded_for(base)
       {
         stride = base.stride();
         ivect loff = ((lower - base.lower()) % stride + stride) % stride;
         ivect uoff = ((upper - base.lower()) % stride + stride) % stride;
         lower -= loff; // go outwards
         upper += (stride - uoff) % stride;
       }
       cout << "aa.expand().contracted_for().expanded_for(): " <<
 ibbox(lower,upper,stride) << endl;
       // expand(buffers % reffact);
       {
         lower -= (buffers[0] % reffact) * stride;
         upper += (buffers[1] % reffact) * stride;
       }
       cout << "aa.expand().contracted_for().expanded_for().expand(): " <<
 ibbox(lower,upper,stride) << endl;
       snapped |= ibbox(lower,upper,stride);
 #endif
 }}}

 Erik: I don't quite trust my understanding of the "reffact-1 - buffers %
 reffact" of the algorithm. Do you have an idea if it would be possible to
 rewrite the order of operations so that it never generates an empty box?

 Otherwise the algorithm seems fine, ie. even for a very tiny strip it
 produces eg. a left edge that is a the same location as that of a thick
 strip with the same left edge initially.

-- 
Ticket URL: <https://trac.einsteintoolkit.org/ticket/797#comment:2>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit