How to get a non-round off corner cell in EB and the coincide EB boundary problem

[pkufourier]

I have the following test code with BoxIF:
//box_test.cpp
#include <AMReX.H>
#include <AMReX_EB2.H>
#include <AMReX_EB2_IF.H>
#include <AMReX_EBFabFactory.H>

int main(int argc, char* argv[])
{
using namespace amrex;
amrex::Initialize(argc, argv);

const int n_cell = 16;
const int max_grid_size = 32;

Geometry geom;
{
	RealBox rb({ 0.0,0.0,0.0 }, { n_cell,n_cell,n_cell }); // physical domain
	Array<int, AMREX_SPACEDIM> is_periodic{ false, false, false };
	Geometry::Setup(&rb, 0, is_periodic.data());
	Box domain(IntVect(0), IntVect(n_cell - 1));
	geom.define(domain);
}

//Testing parameters
const Real x0 = 1e-15, z0=-100;
EB2::BoxIF cube({ x0,0,z0 }, { n_cell,n_cell,100 }, true);

const int nghost = 3;
auto gshop = EB2::makeShop(cube);
EB2::Build(gshop, geom, 0, nghost);

MultiFab mf;

BoxArray ba(geom.Domain());
ba.maxSize(max_grid_size);
DistributionMapping dm{ ba };
Vector<int> ng_ebs = { nghost,nghost,nghost };
std::unique_ptr<EBFArrayBoxFactory> ebfactory = amrex::makeEBFabFactory(
	geom, ba, dm, ng_ebs, EBSupport::full);

const FabArray<EBCellFlagFab>& flags = ebfactory->getMultiEBCellFlagFab();
mf.define(ba, dm, 1, nghost, MFInfo(), *ebfactory);

const MultiFab& vfrac_mf = ebfactory->getVolFrac();
for (MFIter mfi(mf); mfi.isValid(); ++mfi)
{
	const Box& box = mfi.validbox();
	Array4<const EBCellFlag> const& flag_arr = flags.array(mfi);
	auto const& vfrac_arr = vfrac_mf[mfi].const_array();

	printf("is_singleV=%d,%d,%d\n", 
		flag_arr(-1, 0, 0).isSingleValued(), flag_arr(0, 0, 0).isSingleValued(), 
		flag_arr(1, 0, 0).isSingleValued());
	printf("vfrac(-1,0,0),(0,0,0),(1,0,0)=%6f,%6f,%6f\n", 
		vfrac_arr(-1, 0, 0),vfrac_arr(0, 0, 0), vfrac_arr(1, 0, 0));
}

amrex::Finalize();

}

Notice the testing parameters x0 and z0. For simplicity, we can set z0=-100 and this is a 2D box actually. Then with x0=0 or x0=1e-15, the final output would be
//x0=0, z0=-100
is_singleV=0,0,0
vfrac(-1,0,0),(0,0,0),(1,0,0)=1.000000,1.000000,1.000000

//x0=1e-15, z0=-100
is_singleV=0,1,1
vfrac(-1,0,0),(0,0,0),(1,0,0)=0.000000,0.500000,1.000000

Here I have two questions:
(1)When the EB box’s boundary coincides with the Geometry domain, the EB build shop seems do nothing? (see the first output) It seems to be a little buggy?
(2)When setting a little offset for the EB box(x0=1e-15), obviously the (0,0,0) cell was rounded off to be a triangle at the corner (see the second output). Could I get a non-round off cell at the box corner?

The testing code can be found here:
box_test.zip

[WeiqunZhang]

(1) By default we extrapolate the implicit function's value from the domain face to what's outside the domain. You can turn it off by setting a runtime parameter eb2.extend_domain_false.

(2) The EB in a cut cell is approximated by a single plane. So we cannot have a sharp corner in a cell.