diff options
Diffstat (limited to 'infra/data-impl/src')
-rw-r--r-- | infra/data-impl/src/main/java/io/fd/honeycomb/data/impl/ModificationDiff.java | 114 |
1 files changed, 91 insertions, 23 deletions
diff --git a/infra/data-impl/src/main/java/io/fd/honeycomb/data/impl/ModificationDiff.java b/infra/data-impl/src/main/java/io/fd/honeycomb/data/impl/ModificationDiff.java index 86666ba70..9903d99fc 100644 --- a/infra/data-impl/src/main/java/io/fd/honeycomb/data/impl/ModificationDiff.java +++ b/infra/data-impl/src/main/java/io/fd/honeycomb/data/impl/ModificationDiff.java @@ -24,6 +24,7 @@ import static org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationT import static org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType.WRITE; import com.google.common.collect.ImmutableMap; +import java.util.Collection; import java.util.Collections; import java.util.EnumSet; import java.util.HashMap; @@ -296,19 +297,23 @@ final class ModificationDiff { private final DataTreeCandidateNode dataCandidate; // Using Object as type for schema node since it's the only type that's a parent to all schema node types from // yangtools. The hierarchy does not use e.g. SchemaNode class for all types + private final Object parentNode; private final Object schemaNode; Modification(final YangInstanceIdentifier id, final DataTreeCandidateNode dataCandidate, + final Object parentNode, final Object schemaNode) { this.id = id; this.dataCandidate = dataCandidate; + this.parentNode = parentNode; this.schemaNode = schemaNode; } - Stream<Modification> streamChildren() { - return dataCandidate.getChildNodes().stream() - .map(child -> new Modification(id.node(child.getIdentifier()), child, schemaChild(schemaNode, child.getIdentifier()))); + Modification(final YangInstanceIdentifier id, + final DataTreeCandidateNode dataCandidate, + final Object schemaNode) { + this(id, dataCandidate, schemaNode, schemaNode); } List<Modification> getChildNodes() { @@ -355,35 +360,98 @@ final class ModificationDiff { return dataCandidate.getDataAfter().isPresent(); } + private AugmentationSchema findAugmentation(Object currentNode, + final YangInstanceIdentifier.AugmentationIdentifier identifier) { + if (currentNode != null) { + // check if identifier points to some augmentation of currentNode + if (currentNode instanceof AugmentationTarget) { + Optional<AugmentationSchema> augmentationSchema = + ((AugmentationTarget) currentNode).getAvailableAugmentations().stream() + .filter(aug -> identifier.equals(new YangInstanceIdentifier.AugmentationIdentifier( + aug.getChildNodes().stream() + .map(SchemaNode::getQName) + .collect(Collectors.toSet())))) + .findFirst(); + if (augmentationSchema.isPresent()) { + return augmentationSchema.get(); + } + } + + // continue search: + Collection<DataSchemaNode> childNodes = Collections.emptyList(); + if (currentNode instanceof DataNodeContainer) { + childNodes = ((DataNodeContainer) currentNode).getChildNodes(); + } else if (currentNode instanceof ChoiceSchemaNode) { + childNodes = ((ChoiceSchemaNode) currentNode).getCases().stream() + .flatMap(cas -> cas.getChildNodes().stream()).collect(Collectors.toList()); + } + return childNodes.stream().map(n -> findAugmentation(n, identifier)).filter(n -> n != null).findFirst() + .orElse(null); + } else { + return null; + } + } + + Stream<Modification> streamChildren() { + return dataCandidate.getChildNodes().stream() + .map(child -> { + final YangInstanceIdentifier childId = id.node(child.getIdentifier()); + final Object schemaChild = schemaChild(schemaNode, child.getIdentifier()); + // An augment cannot change other augment, so we do not update parent node if we are streaming + // children of AugmentationSchema (otherwise we would fail to find schema for nested augmentations): + final Object newParent = (schemaNode instanceof AugmentationSchema) + ? parentNode + : schemaNode; + return new Modification(childId, child, newParent, schemaChild); + }); + } + /** * Find next schema node in hierarchy. */ - private Object schemaChild(final Object schema, final YangInstanceIdentifier.PathArgument identifier) { + private Object schemaChild(final Object schemaNode, final YangInstanceIdentifier.PathArgument identifier) { Object found = null; if (identifier instanceof YangInstanceIdentifier.AugmentationIdentifier) { - if (schema instanceof AugmentationTarget) { + if (schemaNode instanceof AugmentationTarget) { // Find matching augmentation - found = ((AugmentationTarget) schema).getAvailableAugmentations().stream() - .filter(aug -> identifier.equals(new YangInstanceIdentifier.AugmentationIdentifier( - aug.getChildNodes().stream() - .map(SchemaNode::getQName) - .collect(Collectors.toSet())))) - .findFirst() - .orElse(null); + found = ((AugmentationTarget) schemaNode).getAvailableAugmentations().stream() + .filter(aug -> identifier.equals(new YangInstanceIdentifier.AugmentationIdentifier( + aug.getChildNodes().stream() + .map(SchemaNode::getQName) + .collect(Collectors.toSet())))) + .findFirst() + .orElse(null); + + if (found == null) { + // An augment cannot change other augment, but all augments only change their targets (data nodes). + // + // As a consequence, if nested augmentations are present, + // AugmentationSchema might reference child schema node instances that do not include changes + // from nested augments. + // + // But schemaNode, as mentioned earlier, contains all the changes introduced by augments. + // + // On the other hand, in case of augments which introduce leaves, + // we need to address AugmentationSchema node directly so we can't simply do + // found = schemaNode; + // + found = + findAugmentation(parentNode, (YangInstanceIdentifier.AugmentationIdentifier) identifier); + } } - } else if (schema instanceof DataNodeContainer) { + } else if (schemaNode instanceof DataNodeContainer) { // Special handling for list aggregator nodes. If we are at list aggregator node e.g. MapNode and // we are searching for schema for a list entry e.g. MapEntryNode just return the same schema - if (schema instanceof ListSchemaNode && - ((SchemaNode) schema).getQName().equals(identifier.getNodeType())) { - found = schema; + if (schemaNode instanceof ListSchemaNode && + ((SchemaNode) schemaNode).getQName().equals(identifier.getNodeType())) { + found = schemaNode; } else { - found = ((DataNodeContainer) schema).getDataChildByName(identifier.getNodeType()); + found = ((DataNodeContainer) schemaNode).getDataChildByName(identifier.getNodeType()); } - } else if (schema instanceof ChoiceSchemaNode) { + } else if (schemaNode instanceof ChoiceSchemaNode) { // For choices, iterate through all the cases - final Optional<DataSchemaNode> maybeChild = ((ChoiceSchemaNode) schema).getCases().stream() + final Optional<DataSchemaNode> maybeChild = ((ChoiceSchemaNode) schemaNode).getCases().stream() .flatMap(cas -> cas.getChildNodes().stream()) .filter(child -> child.getQName().equals(identifier.getNodeType())) .findFirst(); @@ -391,12 +459,12 @@ final class ModificationDiff { found = maybeChild.get(); } // Special handling for leaf-list nodes. Basically the same as is for list mixin nodes - } else if (schema instanceof LeafListSchemaNode && - ((SchemaNode) schema).getQName().equals(identifier.getNodeType())) { - found = schema; + } else if (schemaNode instanceof LeafListSchemaNode && + ((SchemaNode) schemaNode).getQName().equals(identifier.getNodeType())) { + found = schemaNode; } - return checkNotNull(found, "Unable to find child node in: %s identifiable by: %s", schema, identifier); + return checkNotNull(found, "Unable to find child node in: %s identifiable by: %s", schemaNode, identifier); } @Override |