If Tensor-Newton sub-problem fails (solved via recursive call to nlls_iterate the tensor_inform%status is not propagated to inform%status, actually, tensor_inform%status is only used to break the iteration loop and is never propagated! see here
Proposed fix is to add a check and propagate on non-zero exit status after
calling nlls_finalize.
The proposed change aligns with the current behaviour of all other sub-problems in the allow_fallback scheme.
NOTE: This change induces hard exit (no_progress_x error) of (outer) calculate_step due to (inner) calculate_step non-zero exit status.
If Tensor-Newton sub-problem fails (solved via recursive call to
nlls_iterate
thetensor_inform%status
is not propagated toinform%status
, actually, tensor_inform%status is only used to break the iteration loop and is never propagated! see hereProposed fix is to add a check and propagate on non-zero exit status after calling
nlls_finalize
.The proposed change aligns with the current behaviour of all other sub-problems in the
allow_fallback
scheme.NOTE: This change induces hard exit (no_progress_x error) of (outer)
calculate_step
due to (inner)calculate_step
non-zero exit status.