yade-dev team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

> in the case of engines using getStress function, there can be used a
> hack with useDem3dof parameter, but in general getStress function
> depends on if you use SCGeom or Dem3DofGeom (for the same particles
> displacement it differs in sign).
Yes, unfortunately.

> So ideal way would be the runtime determination of interaction type in
> getStress, so from performance point of view I recommend to have one
> more parameter (const static int), which would be (for example) 0 for
> GenericSphereContact, 1 for SCGeom and 2 for Dem3DofGeom, and in
> getStress it would look like:
>
> int type = I->phys.get().type;
> if (isPeriodic) branch += (type==1? 1 :
> -1)*scene->cell->hSize*I->cellDist.cast<Real>();
>
> would such solution be ok?
> Jan
>
I thought about that once before, It could work but it is complicated
and a bit weird.
It's a short term workaround to a long standing problem, and it will
still be painful in the long run.
Everyone coding a function using contact forces would have to remember
that i->phys->force needs to be inverted depending on
I->phys.get().type. A nightmare for developers.
Coding a new law would give 50% chance to pick the wrong type.

I wanted to remove those "forceSign" flags when possible, adding new
flags in interactions would not help.
There is only one law using the inverted convention currently:
ConcretePM (not counting orphan laws that will be removed soon). The
only real fix I can imagine is to revert the force convention in this
ConcretePM  (maybe 4 lines to change), then all laws would have same
convention and we can get rid of the remaining flags, else I'd prefer to
keep current situation.

Bruno