Computing a metric in accordance with the lengths of required edges passing through a required point is a way to:

• avoid mesh degeneracy near required entities
• speed up the adaptation as mesh degeneracy and stuck edges are slowing down the process

For now, we compute a required size at required points:

• this computation is ok for isotropic meshes
• this computation don't work for anisotropic meshes (it is computing as an isotropic metric)
• metric computation at required points can be disabled with the noinsert option.

Then:

• the metrics are propagated from required points toward neighbours (advancing-front like algo)
• this propagation (called required gradation in Mmg) can be disabled or tuned using the hgradreq keyword

Issues:

• representation of entities on which the propagation applies seems weird for both iso and aniso metrics
• in aniso mode, we use the simultaneous reduction to evaluate the new metric: often we obtain tensors that seems to have random directions / lengths. It may be linked to collisions in the advancing-front or we may have an implementation issue.
• how to deal with collisions in advancing front?
• how to ensure that required gradation / sizes doesn't break the Mmg convergency?