Pathed RefineNestedAccess transformation.

The main addition is a patch to detect some cases that the RefineNestedAccess Transformation can not handle.
HOWEVER IT IS IMPORTANT TO NOTE, THAT THIS DOES NOT SOLVES ALL OF ITS PROBLEMS, SEE PREVOUS COMMITS.

dace/transformation/interstate/sdfg_nesting.py

@@ -929,11 +929,23 @@ def _candidates(
                 read_set, write_set = nstate.read_and_write_sets()
                 for e in nstate.in_edges(dnode):
                     if e.data.data not in write_set:
-                        # Skip data which is not in the read and write set of the state -> there also won't be a
-                        # connector
+                        # Skip data which is not in the write set of the state -> there also won't be a connector
                         continue
-                    # If more than one unique element detected, remove from
-                    # candidates
+
+                    # NOTE: This test is needed to exclude "non canonical Memlets", that can not be handled by
+                    #  this transformation. The main issue is that the transformation unconditionally calls
+                    #  `.subset` without checking if this set really refers to the correct subset.
+                    #  This fix is essentially needed to make
+                    # `tests/transformations/refine_nested_access_test.py::test_rna_read_and_write_sets_doule_use_non_canon_memlet`
+                    #  pass. The same scanning is applied to the `in_candidate`. It is however, not enough to make
+                    # `tests/transformations/refine_nested_access_test.py::test_rna_read_and_write_sets_different_storage_non_canon_memlet` pass.
+                    # TODO: Fix this correctly.
+                    if e.data.dst_subset is not e.data.subset:
+                        return ({}, {})
+                    if e.data.data != dnode.data:
+                        return ({}, {})
+
+                    # If more than one unique element detected, remove from candidates
                     if e.data.data in out_candidates:
                         memlet, ns, indices = out_candidates[e.data.data]
                         # Try to find dimensions in which there is a mismatch
@@ -944,15 +956,23 @@ def _candidates(
                         if len(indices) == 0:
                             ignore.add(e.data.data)
                         out_candidates[e.data.data] = (memlet, ns, indices)
-                        continue
-                    out_candidates[e.data.data] = (e.data, nstate, set(range(len(e.data.subset))))
+                    else:
+                        out_candidates[e.data.data] = (e.data, nstate, set(range(len(e.data.subset))))
+
                 for e in nstate.out_edges(dnode):
                     if e.data.data not in read_set:
-                        # Skip data which is not in the read and write set of the state -> there also won't be a
-                        # connector
+                        # Skip data which is not in the read set of the state -> there also won't be a connector
+                        # TODO: This is technically not enough, because the read set might be filtered.
+                        #        See `SDFGState.read_and_write_sets()` for more.
                         continue
-                    # If more than one unique element detected, remove from
-                    # candidates
+
+                    # See note above.
+                    if e.data.src_subset is not e.data.subset:
+                        return ({}, {})
+                    if e.data.data != dnode.data:
+                        return ({}, {})
+
+                    # If more than one unique element detected, remove from candidates
                     if e.data.data in in_candidates:
                         memlet, ns, indices = in_candidates[e.data.data]
                         # Try to find dimensions in which there is a mismatch
@@ -963,8 +983,8 @@ def _candidates(
                         if len(indices) == 0:
                             ignore.add(e.data.data)
                         in_candidates[e.data.data] = (memlet, ns, indices)
-                        continue
-                    in_candidates[e.data.data] = (e.data, nstate, set(range(len(e.data.subset))))
+                    else:
+                        in_candidates[e.data.data] = (e.data, nstate, set(range(len(e.data.subset))))
 
         # Check read memlets in interstate edges for candidates
         for e in nsdfg.sdfg.edges():
@@ -979,8 +999,8 @@ def _candidates(
                     if len(indices) == 0:
                         ignore.add(m.data)
                     in_candidates[m.data] = (memlet, ns, indices)
-                    continue
-                in_candidates[m.data] = (m, None, set(range(len(m.subset))))
+                else:
+                    in_candidates[m.data] = (m, None, set(range(len(m.subset))))
 
         # Check in/out candidates
         for cand in in_candidates.keys() & out_candidates.keys():