Define a "ghcomp" version of gcomp. Normal comp looks like:
comp^i A [ phi -> u ] u0 = hcomp^i A(1/i) [ phi -> forward A i u ] (forward A 0 u0)
So for "gcomp" we compute:
gcomp^i A [ phi -> u ] u0 = hcomp^i A(1/i) [ phi -> forward A i u, ~ phi -> forward A 0 u0 ] (forward A 0 u0)
The point of this is that gcomp does not produce any empty systems (if phi = 0 it will reduce to "forward A 0 u".