Paul created TINKERPOP-3304:
-------------------------------
Summary: mergeV() and mergeE() pass a null old value to
MutationListener when the property did not exist yet
Key: TINKERPOP-3304
URL: https://issues.apache.org/jira/browse/TINKERPOP-3304
Project: TinkerPop
Issue Type: Bug
Components: process
Affects Versions: 3.8.2, 3.7.3
Reporter: Paul
When EventStrategy is configured, \{{MergeVertexStep}} and \{{MergeEdgeStep}}
(onMatch) report the properties they set through
\{{EventUtil.registerVertexPropertyChange()}} and
\{{EventUtil.registerEdgePropertyChange()}}. When the property does not exist
yet, both pass \{{null}} as the old value:
{code:java}
final Property<Object> oldValue = p.isPresent() ?
eventStrategy.detach(edge.property(key)) : null;
{code}
{\{AddPropertyStep}} does not. Since 3.4.0 (TINKERPOP-1831, see the 3.4.x
upgrade documentation: "the event now raises with a \{{KeyedVertexProperty}} or
\{{KeyedProperty}} instance, which only contains a property key and no value in
them"), events raised from new mutations carry a keyed property instead of
\{{null}}, so a listener can tell which property was added. This is asserted by
\{{EventStrategyProcessTest}} (\{{shouldDetachPropertyOfEdgeWhenNew}},
\{{shouldDetachVertexPropertyWhenNew}}). \{{mergeV()}}/\{{mergeE()}}, added in
3.6.0, still raise \{{null}}: the key is lost, and a listener written for the
documented contract fails with a \{{NullPointerException}} on
\{{oldValue.key()}} or \{{oldValue.isPresent()}}.
To reproduce, with TinkerGraph:
{code:java}
MutationListener listener = ...; // records the oldValue received by
edgePropertyChanged()
GraphTraversalSource g = traversal().withEmbedded(graph)
.withStrategies(EventStrategy.build().addListener(listener).create());
Vertex v = graph.addVertex();
v.addEdge("self", v);
g.E().property("a", 1).iterate();
// oldValue is a KeyedProperty, key() == "a"
g.V(v).mergeE(Map.of(T.label, "self")).option(Merge.onMatch, Map.of("b",
2)).iterate();
// oldValue is null: nothing tells the listener that "b" is the property that
was set
{code}
The same happens with \{{mergeV(...).option(Merge.onMatch, ...)}} and
\{{vertexPropertyChanged()}}. The code is identical on 3.7.3, 3.8.2 and the
current 3.8-dev and master branches.
We hit this with a listener that records the history of property changes: the
first modification of an edge made through \{{mergeE()}} is never recorded.
Expected: the same as \{{property()}}, a \{{KeyedProperty}} /
\{{KeyedVertexProperty}} when the property is new.
A patch with tests is ready; a pull request will follow.
Notes: the Javadoc of \{{Event.EdgePropertyChangedEvent}} and
\{{Event.VertexPropertyChangedEvent}} still says that \{{oldValue}} is
\{{null}} for a new property; it predates the 3.4.0 change and is updated in
the fix. A listener that tests \{{oldValue == null}} on merge events would
receive a non-null, non-present property after the fix.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
