CHANGELOG.asciidoc

TinkerPop3 CHANGEGLOG

TinkerPop 3.0.0 (A Gremlin Rāga in 7/16 Time)

TinkerPop 3.0.0.M7 (Release Date: NOT OFFICIALLY RELEASED YET)

• Added unique identifies to Step that are not the user provided labels. Step.getLabel() now returns an Optional<String>.

• Removed UnionLinearStrategy, ChooseLinearStrategy, and RepeatLinearStrategy as nested traversals are now natively supported in OLAP.

• Fixed Neo4jGraph around manual transaction behavior on commit and rollback such that they would throw exceptions if a transaction was not open.

• Redesigned the hidden step labeling mechanism so its consistent across a cluster, easier for rewrite strategies, and will enable nested OLAP traversals.

• Traverser.incrLoops() now takes a string step label to enable nested looping constructs (i.e. loop stacks).

• Added Traversal.tryNext() which returns an Optional, where the provided default method should be sufficient for all vendors.

• Removed PathConsumer in favor of TraverserRequirement.PATH-model via Step.getRequirements().

• Step.getRequirements() returns a Set<TraverserRequirement> which is what is required of the Traverser by the Step.

• Traverser now extends Cloneable and Traverser.clone() is used to good effect in Traverser.split().

• Added AbstractTraverser for which all traversers extend.

• Added TraverserSerializer for a generic way to detach and serialize a traverser.

• Moved Traversal.SideEffects to TraversalSideEffects as sideEffects are not necessarily tied to the traversal.

• Removed Graph.of() for generating anonymous graph traversals — replaced by __-model.

• Removed Graph being stored in Traversal.SideEffects. Too dangerous when moving between OLTP and OLAP and its limited uses were worked around easily.

• No need for DefaultXXXGraphTraversal unless the vendor is extending with new methods (e.g. DefaultNeo4jGraphTraversal).

• Reworked TraversalStrategies such that the are "emanating object class"-dependant, not Traversal dependent.

• Moved Traverser.sideEffects() to Traverser.asAdmin().getSideEffects(). Users should use Traverser.sideEffects(key) and Traverser.sideEffects(key,value).

• Added SerializationTest to the StructureStandardSuite in gremlin-test which validates serialization at a lower level than IoTest.

• Removed IntervalStep and renamed interval() to between() which is simply an alias to a has().has() chain.

• Added AnonymousGraphTraversal.Tokens.__ which allows for __.out().out()-style construction of anonymous traversals (instead of g.of()).

• The only GraphTraversal steps that operate on Traverser are the base lambdas and repeat() (i.e. emit() and until()).

• Removed dependency on the reflections library in gremlin-test which removed the default implementation of GraphProvider.getImplementations() - vendors now need to implement this method themselves.

• Relaxed the <S> typing requirement for anonymous traversals when applied to choose(), repeat(), union(), etc.

• Removed LoopStep and UntilStep in favor of the new RepeatStep model of looping in Gremlin3.

• BranchStep is now exposed in GraphTraversal via branch(function).

• UnionStep now implements TraversalHolder.

• Added RepeatStep as the new looping construct supporting do/while, while/do, and emit semantics.

• Moved Traversal.sideEffects() to Traversal.Admin.getSideEffects() as cap() should be used to access the sideEffect data of a traversal.

• Renamed vendor XXXTraversal to XXXGraphTraversal (interface) and XXXGraphTraversal to DefaultXXXGraphTraversal (implementation class).

• Modified packaging for console plugins to be more consistent by moving them to the com.tinkerpop.gremlin.console.groovy.plugin namespace.

• Removed all TinkerPop specific dependencies to Guava to avoid user version conflicts.

• Added support for -e (script file execution) and -v (version display) options on gremlin.sh.

• GraphSON supports the assignment of multiple custom serialization modules.

• Traverser.get(stepLabel/sideEffectKey) no longer exists. There now exists: Traverser.path(stepLabel) and Traverser.sideEffects(sideEffectKey).

• SimpleTraverser now supports "path" but in a very loose, global cache way. Added SparsePath as a Map-backed Path implementation.

• Provided Neo4j multi-label support in Neo4j-Gremlin. Added three Neo4jVertex-specific methods: addLabel(), removeLabel(), labels().

• Removed TraverserSourceStrategy as it is no longer required.

• TraverserSource is now only a marker interface with "on first" within the TraverserSource step are traversers generated.

• Bumped to Groovy 2.3.9.

• Added Graph.Io interface which allows for simplified helper methods for end users and a way for vendors to override GraphReader and GraphWriter initial construction when custom serializers are needed.

• Removed methods from GraphProvider related to customizing serializers in IoTest from the test suite as the new Graph.Io interface now serves that purpose.

• Added Neo4jGraph.checkElementsInTransaction(boolean) which will (or not) verify whether elements retrieved via Neo4j global graph operations are transactionally consistent.

• Added ScriptInputFormat and ScriptOutputFormat to Hadoop-Gremlin for reading and writing a file according to an arbitrary parsing script.

• Added TimeLimitStep.getTimedOut() to determine if the step timed out or there were no more objects to process.

• Graph.System is now Graph.Hidden with "hidden" being the vendor namespace and the key prefix being ~.

• Much better toString() handling in Step and Traversal.

• ComparatorHolder<V> interface returns a List<Comparator<V>> instead of a Comparator<V>[].

• T now implements Function<Element,Object>.

• Added ElementValueComparator and ElementFunctionComparator in support of vendor introspection on ComparatorHolder-steps.

• Renamed Comparing marker interface to ComparatorHolder.

• FunctionHolder interface provides vendor introspection via ElementValueFunction.

• Removed OrderByStep as it is now just order() with a by()-based comparator.

• Added SampleStep (sample()) to allow for sampling the set of previous objects. Useful for doing random walks with local().

• Renamed random() to coin() to better express that the filter is a random coin toss.

• Added by()-projection to modulate the meaning of post-processing steps like aggregate(), groupCount(), path(), order(), etc.

• Removed the Strategy interface and gave StrategyGraph direct access to the GraphStrategy.

• Added Graph.strategy() to help instantiate StrategyGraph instances.

• Modified the signature of all GraphStrategy methods to include an parameter that contains a reference to the "composing strategy".

• PartitionStrategy hides the specified partition key from view when iterating properties, keys, etc.

• Change construction of GraphStrategy implementations to be consistent with singleton instances and builder pattern.

• Added Graph.Helper annotation to "protected" certain default interface methods from implementation by vendors.

• Transaction retry functions now work with "manual" transactions.

• Improved error messaging when importing "legacy" GraphSON that was not generated with "extended" properties.

• Renamed "iterator" related methods in the GraphStrategy interface to be consistent with the method names they represent.

• PropertyMapStep (valueMap()) now takes a boolean to state if the tokens of the element are desired along with its properties.

• HadoopGraph now connected to the StructureProcessSuite.

• HadoopGraph no longer supports Graph.Variables as they were in-memory. A persistence mechanism can be introduced in the future.

• Hidden properties removed in favor of using GraphStrategy for such features.

• Edge.iterators().vertexIterator(BOTH) now guarantees OUT then IN vertex iterator order.

• Graph.v(Object) and Graph.e(Object) no longer exist. Instead, use Graph.V(Object... ids) and Graph.E(Object... ids).

• Added Graph.Iterators to allow access to vertex and edge iterators based on element ids and bypassing GraphTraversal.

• Renamed GraphStrategy implementations to be less verbose - removed the word "Graph" from their names (e.g. IdGraphStrategy simply changed to IdStrategy).

• Removed Step.NO_OBJECT as the problem is solves can be solved with proper use of flatMap and EmptyTraverser.

• Path is now part of GraphSerializer and thus, not specific to a particular implementation of Path.

• Added messaging to show files being downloaded when using the Gremlin Server "install" command.

• Added test name and class arguments to the GraphProvider.loadGraphWith method.

• Merged ReferencedXXX and DetachedXXX so that all migration of graph element data is via DetachedXXX.

• Added StaticVertexProgram and StaticMapReduce which simply return this on clone().

• VertexProgram and MapReduce now implement Cloneable and is used for fast copying across workers within the same machine.

• Added TraversalHolder interface which extends PathConsumer to determine recursively if nested traversals require path calculations turned on.

• Reworked how a TraverserGenerator is retrieved and utilized.

• Added Traversal.toBulkSet() to make getting resultant data more efficiently for traversals with repeated data.

• Provided a helper LocalStep.isLocalStarGraph() so GraphComputer implementers know the requisite data boundaries.

• Created Traversal.Admin to hide administrative methods. Added Traversal.asAdmin() to get at Traversal.Admin.

• Fixed up all Step cloning operations realizing that Java8 lambdas are always bound to the calling class (no delegates).

• Usage of :remote close without configured remotes shows a reasonable message rather than a stack trace.

• Provided LocalStep to signify that the internal traversal is locally bound to the incoming object.

• Failed script evaluation in Gremlin Server now triggers the cancel of the process attempting to timeout the script if it were to run too long.

• Greatly increased the speed of ScriptEngineLambda by making use of a static ScriptEngine cache.

• Fixed a general bug in all sideEffect using steps where the sideEffect should be accessed via the Traverser not Traversal.

• GremlinPlugin interface no longer has the additionalDependencies method - those dependencies are now defined by an entry in the manifest file for the jar called Gremlin-Plugin-Dependencies.

• Added TinkerWorkerPool which is used for resource efficient threading in TinkerGraphComputer.

• MapReduce.createMapReduce(Configuration) now exists and serves the same purpose as VertexProgram.createVertexProgram(Configuration).

• Enabled SessionOps to be extended. Added eval handler hook.

• Setting a property with an unsupported data type throw IllegalArgumentException instead of UnsupportedOperationException as the operation is supported, but the argument is not.

TinkerPop 3.0.0.M6 (Release Date: December 2, 2014)

• javatuples.Pair avoided on MapReduce API in favor of a new KeyValue class.

• Renamed Gremlin-Plugin manifest entry for plugins to Gremlin-Plugin-Paths.

• Added Gremlin-Plugin-Dependencies manifest entry to list other dependencies that should be retrieved with a plugin jar.

• Memory.Admin.asImmutable() yields an immutable representation of the GraphComputer Memory.

• Fixed host selection in gremlin-driver by properly accounting for all hosts being marked unavailable at the instantiation of a Client.

• Removed Giraph-Gremlin in favor of new Hadoop-Gremlin with GiraphGraphComputer support. Future support for MapReduceGraphComputer.

• Greatly simplified the InputFormat and OutputFormat model for working with Giraph (and Hadoop).

• Added a serializer for Property for GraphSON correcting format of serialization of a single Property on an Edge.

• Fixed bug in Gremlin Console that prevented assignments to empty List objects.

• Added VertexProgram.getMessageScopes() to allow vendors to know which MessageScopes at a particular Memory state.

• Reduced the number of methods in MessageScope.Local as its up to vendors to inspect provided incident Traversal accordingly.

• Renamed MessagesType to MessageScope to make it less ambiguous regarding the class of the messages being sent.

• Changed the message type of TraversalVertexProgram to TraverserSet to support message combining.

• Added VertexProgram.getMessageCombiner() to support the combining of messages in route to a vertex.

• Reduced object creation in TraversalVertexProgram around vertex-local traversal sideEffects.

• Renamed Traverser.Admin.makeChild() and Traverser.Admin.makeSibling() to Traverser.Admin.split() to correspond with merge().

• Added Traverser.Admin.merge(Traverser) method so that the merging algorithm is with the Traverser.

• Added Operator enum that contains sack-helpful BinaryOperators: sum, minus, mult, div, max, min, etc.

• Added GraphTraversal.withSack() and renamed trackPaths() and with() to withPath() and withSideEffect(), respectively.

• Added the "Gremlin Sacks" feature to allow a Traverser to carry local information along its walk.

• GraphSON format no longer makes use of hiddens JSON key. Its all just properties.

• Added DoubleIterator to make vendor implementations of Edge.iterators().vertexIterator() efficient.

• PropertiesStep is smart about hiddens vs. properties.

• Element.iterators().hiddenProperties() no longer exists. For vendors, simply provide an iterator of properties.

• GIRAPH_GREMLIN_LIBS supports colon separated directories for loading jars from multiple paths.

• Introduced method to control the location of dependencies dynamically loaded to the Gremlin Console as part of the :install command.

• Fixed problem with the Neo4j Gremlin Plugin not loading properly after Gremlin Console restart.

• Removed the "use" configuration from Gremlin Server.

• Moved SugarGremlinPlugin from gremlin-console to gremlin-groovy so that it could be shared with Gremlin Server.

• Fixed bug in serialization of null results returned to the Gremlin Console when serializing to strings.

• Moved the GremlinPlugin for TinkerGraph to tinkergraph-gremlin module (it is no longer in gremlin-console).

• Added a plugin-info.txt file to Gremlin Console /ext/{module} subdirectories to identify the module that was originally requested.

• Gremlin Server now allows for the explicit configuration of plugin activation.

• Refactored GremlinPlugin and AbstractGremlinPlugin to better account for plugins that run on the server and those that run in the console.

• Added a plugins configuration to Gremlin Server to control the plugins that are enabled on initialization.

• Added a builder option to GremlinExecutor to control the plugins that are enabled on initialization.

• Added RemoteException for usage with RemoteAcceptor implementations for the Gremlin Console so as to better standardize their development.

• Standardized all text being written to the Gremlin Console using starting upper case letter.

• Prevented error in the Console when :submit is called but no remotes were configured.

• Provided a way to clean the grapes directory as part of a standard build with mvn clean install.

TinkerPop 3.0.0.M5 (Release Date: November 7, 2014)

• Removed PropertyFilterIterator as using Java8 streams was just as efficient for the use case.

• Renamed KryoWritable to GremlinWritable as it is not necessarily Kryo that is the serialization mechanism.

• Fixed an input split bug in Giraph that was making it so that splits were not always at vertex boundaries.

• Fixed a combiner bug in GirapGraphComputer. Combiners were always calling MapReduce.reduce(), not MapReduce.combine().

• Greatly simplified SubgraphStrategy by removing requirements for Traversal introspection.

• StrategyWrappedGraph mimics vendor use of GraphStep and GraphTraversal and no longer requires dynamic strategy application.

• TraversalStrategies make use of a dependency tree sorting algorithm to ensure proper sorts prior to application.

• TraversalStrategies are now immutable and are bound to the Traversal class.

• Fixed bug in Gephi Plugin that prevented it from communicating with the Gephi Streaming Server.

• Renamed MessageType.XXX.to() to MessageType.XXX.of() so it makes sense in both the sending and receiving context.

• Improved messaging with respect to tests that are ignored due to features to make it clear that those tests are not in error.

• Relaxed exception consistency checks in the test suite to only check that a thrown exception from an implementation extends the expected exception class (but no longer validates that it is the exact class or that the message text).

• VertexProgram now has workerIterationStart() and workerIterationEnd() to allow developers to control vertex split static data structures.

• TraversalVertexProgram startup time greatly reduced due to being smart about loadState() behavior.

• Gremlin Server sessions now allow serialization of results that were part of an open transaction.

• Refactor OpProcessors implementations in Gremlin Server for better reusability.

• Vertex.iterators() no longer have a branchFactor. This is now at the query language level with localLimit().

• Added limit(long) and localLimit(int,int) which simply call the range equivalents with 0 as the low.

• Added LocalRangeStep which supports ranging the edges and properties of an element — localRange(int,int).

• GraphTraversal.value(String) no longer exists. Instead, use GraphTraversal.values(String).

• HiddenXXXStep and ValueXXXStep no longer exist. PropertyXXXStep takes a PropertyType to denote value and hidden access.

• Added PropertyType to the structure-package which provide markers for denoting property types (vs. property classes).

• Renamed setWorkingDirectory to workingDirectory in the KryoReader builder.

• Path.get(String) returns the object if only one object is referenced by label, else it returns a List of referenced objects.

• Added overload to GremlinKryo to allow a serializer to be configured as a Function<Kryo,Serializer> to allow better flexibility in serializer creation.

• Added method to GraphProvider to allow implementers to provide a mechanism to convert GraphSON serialized identifiers back to custom identifiers as needed.

• Added methods to GraphProvider so that implementers could specify a custom built GremlinKryo class and/or SimpleModule class in case their implementation had custom classes to be serialized.

• Added Traversal.forEachRemaining(class,consumer) for those traversals whose end type is different from declared due to strategy rewriting.

• Removed Traversal.forEach() as traversal implements Iterator and users should use forEachRemaining().

• RangeStep now has an inclusive low and an exclusive high — a change from Gremlin2.

• DriverGremlinPlugin returns raw results with driver results available via the result variable.

• Removed test enforcement of private constructor for a Graph instance.

• RemoteAcceptor now supports @ prefixed lines that will grab the script string from the Gremlin Console shell.

• Modified the signature of Property.element() to simply return Element

• Added Reducing marker and ReducingStrategy which supports reduction-functions as a final step in Gremlin OLAP (e.g. fold()).

• Once strategies are complete(), no more steps can be added to a Traversal.

• Renamed Traversal.strategies() to Traversal.getStrategies() as it is not a "query language"-method.

• Added test to enforce that a label on a VertexProperty is always set to the key of the owning property.

• Fixed bug with multi-property removal in Neo4jGraph.

• Bumped to Neo4j 2.1.5.

• Used standard UUIDSerializer from the kryo-serializers library for serialization of UUID objects.

• Changed GraphSON serialization to only use iterators() - there were still remnants of Traversal usage from previous refactoring.

• Added overload for detach method to allow for the Element to be detached as a "reference" only (i.e. without properties).

• Renamed Item in gremlin-driver to Result.

• Renamed strategy to getStrategy in StrategyWrappedGraph.

• Renamed baseGraph to getBaseGraph in Neo4jGraph.

• Neo4jGraph now returns an empty property Vertex.property(k) when the key is non-existent (a problem only visible when meta/multi property configuration was turned off).

• Traversal.Strategies.apply() now takes a TraversalEngine. Greatly simplifies strategy application for STANDARD or COMPUTER.

• Renamed IdentityReductionStrategy to IdentityRemovalStrategy for reasons of clarity.

• Added ComparingRemovalStrategy that removes Comparing-marked steps unless they are the end step of the traversal.

• OrderStep now works in OLAP, but only makes sense as a traversal end step.

• MapReduce API extended to include getMapKeySort() and getReduceKeySort() to sort outputs accordingly.

• Renamed TraversalResultMapReduce to TraverserMapReduce. Shorter and makes more sense.

• Improved build automation to package javadocs and asciidoc documentation in the distribution files.

• Improved build automation with a script to automatically bump release versions in the various files that needed it such as the pom.xml files.

• The identifier on VertexProperty is now read properly to those graphs that can support identifier assignment.

• GraphSONReader.readGraph() now properly reads vertex properties.

• Removed Neo4jGraph.getCypher() as users should use Neo4jGraph.cypher() and get back TinkerPop3 graph objects.

• GiraphGraph.variables().getConfiguration() is now replaced by GiraphGraph.configuration().

• Added Graph.configuration() which returns the Configuration object of Graph.open().

• Removed TraverserTracker as now there is only a TraverserSet for all halted traversers. A nice simplification of TraversalVertexProgram.

• Renamed Traverser.isDone() to Traverser.isHalted() and DONE to HALT. Consistent with automata terminology.

• Removed PathTraverserExecutor and SimpleTraverserExecutor as a single TraverserExecutor correctly executes both types of traversers.

• TraversalVertexProgram does "reflexive message passing" to reduce the total number of iterations required to execute a traversal.

• MapReduce no-argument constructors are private and thus, only for reflection and loadState() usage.

• MapReducers for TraversalVertexProgram are now smart about with() declared data structures.

• Updated Traversal.SideEffects to use "registered suppliers" and it now works accordingly in both OLTP and OLAP environments.

• Increased the speed of FlatMapStep by approximately 1.5x.

TinkerPop 3.0.0.M4 (Release Date: October 21, 2014)

• Added features for VertexProperty user supplied ids and related data types.

• Removed SideEffectCap marker interface as there is only one SideEffectCapStep and thus, instanceof is sufficient.

• Path.getObjects()/Path.getLabels() renamed to Path.objects()/Path.labels() to be in line with "query language" naming convention.

• Greatly simplified GiraphInternalVertex due to Element.graph() — 1/2 the memory footprint and reduced construction time.

• Renamed Property.getElement() to Property.element() given the "query language" naming convention.

• Element.graph() added which returns the Graph that the element is contained within.

• Added tests for greater consistency around iterating hidden properties.

• Simplified TraversalVertexProgram where only a single TraverserTracker exists for both path- and simple-traversers.

• Fixed a major bug where Arrays.binarySearch was being used on an unsorted array in TinkerGraph and Neo4jGraph.

• Changed ComputerResult.getXXX() to graph() and memory() to be consistent with "query language" naming convention.

• Traverser.getXXX() changed to loops(), bulk(), path(), sideEffects() to be consistent with "query language" naming convention.

• Optimization to reduce the number of empty lists created due to no step class existing for respective TraversalStrategy.apply().

• Added CapTraversal as a marker interface for the cap() method.

• Added union() with GraphComputer UnionLinearStrategy.

• TimeLimitStep was moved to filter/ package. It was a mistake that it was in sideEffect/.

• Provided the configuration for generating both a "full" and "core" set of javadocs, where "full" represents all classes in all projects and "core" is the "user" subset.

• Validated bindings passed to Gremlin Server to ensure that they do not match the most common statically imported values.

• If no script engine name is provided to a LambdaHolder it is assumed to be Gremlin-Groovy.

• MapEmitter and ReduceEmitter have an emit(value) default method where the key is the MapReduce.NullObject singleton.

• Traverser.Admin now implements Attachable as the Traversal.SideEffects can be generated from the Vertex.

• Made a solid effort to ensure that all TinkerPop keys are Graph.System to leave Graph.Key for users.

• The Graph.System prefix is now ^ instead of %&%. Simpler and easier on the toString()-eyes.

• Added Traversal.SideEffects.ifPresent(Consumer) as a default helper method.

• Added profile()-step which provides detailed information about the performance of each step in a traversal.

• No more CountCapStep and CountStep, there is only CountStep and it is elegant.

• Created a AbstractTraversalStrategy with good toString(), hasCode(), and equals() implementations.

• Added CountTraversal as a marker-interface stating that the Traversal has a count() -> Long method.

• Traversal no longer has any step methods as its not required for DSL implementers to have "core steps."

• Added "linearization" strategy for ChooseStep so it is executed correctly on GraphComputer.

• Added ``GraphTraversalStrategyRegistry` which has respective global strategies to make turning on/off strategies easier.

• Added a generic BranchStep to be used for re-writing "meta-steps" for execution on GraphComputer.

• Moved JumpStep, UntilStep, and ChooseStep to a new branch/ package.

• Added test cases to the Structure Suite to enforce consistent operations of reading properties after removal of their owning Element.

• GraphSON format change for full Graph serialization - Graph variables are now serialized with the key "variables" as opposed to "properties".

• Relaxed Graph.toString() test requirements for implementers.

• Made the toString operations in GraphStrategy consistent.

• Added VertexFeatures.supportsRemoveProperty.

• Added VertexPropertyFeatures.supportsRemoveProperty.

• Added EdgeFeatures.supportsRemoveProperty.

• Added VertexFeatures.supportsRemoveVertices.

• Added EdgeFeatures.supportsRemoveEdges.

• Vendors should now get a clear error when mis-spelling something in an @OptOut (or more likely if a test name changes) and it now works all the test suites.

• All plugins now have a default prefix of "tinkerpop." as a namespace.

• GraphComputer now executes a Set<MapReduce> and hashCode()/equals() were implemented for existing MapReduce implementations.

• Changed Contains.in/notin to Contains.within/without as in is a reserved term in most languages (including Java and Groovy).

• Added helper methods for loading data into collections in TraversalHelper.

• Core Traversal methods are smart about bulking — e.g. iterate(), fill(), remove(), etc.

• GroupByStep and GroupByMapReduce leverage BulkSet as the default group data structure.

• Element.Iterator has renamed methods so implementers can do MyElement implements Element, Element.Iterators.

• Renamed MessageType.Global and MessageType.Local creators from of() to to() as it makes more sense to send messages to().

• With Traverser.get/setBulk() there is no need for a TraverserMessage. The Traverser is now the message in TraversalVertexProgram.

• Provided static make() methods for constructing Path implementations.

• Provided a more space/time efficient algorithm for Path.isSimple().

• The JumpStep GraphComputer algorithm Queue is now a TraverserSet.

• AggregateStep and StoreStep now use BulkSet as their default backing Collection (much more space/time efficient).

• Added BulkSet which is like TraverserSet but for arbitrary objects (i.e. a weighted set).

• UnrollJumpStrategy is no longer a default strategy as it is less efficient with the inclusion of TraverserSet.

• Introduced TraverserSet with bulk updating capabilities. Like OLAP, OLTP looping is now linear space/time complexity.

• TinkerGraph’s MapReduce framework is now thread safe with a parallel execution implementation.

• Added a default Traverser.asAdmin() method as a typecast convenience to Traverser.Admin.

• Renamed Traverser.System to Traverser.Admin as to not cause java.lang.System reference issues.

• Renamed Memory.Administrative to Memory.Admin to make it shorter and consistent with Traverser.Admin.

• Fixed a TinkerGraph bug around user supplied vertex property ids.

• Most Step classes are now defined as public final class to prevent inheritance.

• ShuffleStep now extends BarrierStep which enables semantically correct step-sideEffects.

• Leveraged Traverser.getBulk() consistently throughout all steps.

TinkerPop 3.0.0.M3 (Release Date: October 6, 2014)

• All Step fields are now private/protected with respective getters as currently needed and will be added to as needed.

• Gremlin Server no longer has the traverse operation as lambdas aren’t really serialized.

• Path is now an interface with ImmutablePath and MutablePath as implementations (2x speedup on path calculations).

• Traverser now implements Comparable. If the underlying object doesn’t implement Comparable, then a runtime exception.

• Added abstract BarrierStep which greatly simplifies implementing barriers like AggregateStep, OrderStep, etc.

• SelectStep is now intelligent about when to trigger path computations based on label selectors and barriers.

• T no longer has eq, neq, lt, in, etc. Renamed all respective enums and with static import, good in console (e.g. Compare.eq).

• Added Order enum which provides Order.decr and Order.incr.

• Traverser.loops and Jump.loops are now shorts (32767 max-loops is probably sufficient for 99.9999% of use cases).

• Traverser.bulk exists which is how many instances does the traverser represent. For use in grouping with bulk computations.

• Greatly simplified sideEffect steps where there is no distinction between OLAP vs. OLTP (from the Step perspective).

• Removed the need for Bulkable and VertexCentric marker interfaces in process API.

• Renamed configuration parameters in Giraph-Gremlin to be consistent with a giraph.gremlin-prefix.

• Made it possible to pass a ScriptEngine name and string script in TraversalVertexProgram and LambdaVertexProgram.

• Made TinkerGraph a plugin for the Console as it is no longer a direct dependency in gremlin-groovy.

• Added features for supporting the addition of properties via Element.property(String,Object).

• GiraphGraph OLTP tested against Gremlin-Java8 and Gremlin-Groovy — OLAP tested against Gremlin-Groovy.

• Neo4jGraph is now tested against both Gremlin-Java8 and Gremlin-Groovy.

• Renamed the test cases in ProcessTestSuite to be consistent with other Gremlin language variants.

• Add a gremlin-groovy-test suite that can be used to validate implementations against the Groovy variant of Gremlin.

• TinkerGraph is no longer serializable, use a GraphReader/GraphWriter to serialize the graph data.

• Removed implements Serializable on numerous classes to ensure safety and proper usage of utilities for cloning.

• Traversal now implements Cloneable as this is the means that inter-JVM threads are able to get sibling Traversals.

• Created "integration" test for Neo4jGraph that runs the test suite with multi/meta property features turned off.

• Added GraphStrategy methods for VertexProperty.

• Converted the id data type from string to integer in the Grateful Dead sample data.

• Removed all notions of serializable lambdas as this is a misconception and should not be part of TinkerPop.

• Greatly simplified TraversalVertexProgram with three arguments: a Traversal<Supplier>, Class<Traversal<Supplier>>, or a script string with ScriptEngine name.

• Added TraversalScript interface with GroovyTraversalScript as an instance. To be used by OLAP engines and any language variant (e.g. gremlin-scala, gremlin-js, etc.).

• UntilStep now leverages UnrollJumpStrategy accordingly.

• Fixed a bug where the toString() of Traversal was being hijacked by SugarGremlinPlugin.

• Fixed compilation bug in UntilStep that is realized when used in multi-machine OLAP.

• Simplified Enumerator and implementations for MatchStep.

TinkerPop 3.0.0.M2 (Release Date: September 23, 2014)

• Added an exhaust InnerJoinEnumerator fix in MatchStep to get all solutions correctly.

• Neo4jGraph can be configured to allow or disallow meta- and multi-properties.

• Added until()-step as a simpler way to express while-do looping which compiles down to a jump()-step equivalent.

• Added "The Crew" (CREW) toy graph which contains multi-properties, meta-properties, graph variables, hiddens, etc.

• If the Giraph job fails, then the subsequent MapReduce jobs will not execute.

• Added Graph.System class which generates keys prefixed with %&% which is considered the vendor namespace and not allowed by users.

• Added ReferencedVertex (etc. for all graph object types) for lightweight message passing of graph object ids.

• T.* now has label, id, key, value and no longer are these String representations reserved in TinkerPop.

• Traverser now has a transient reference to Traversal.SideEffects.

• "Detached" classes are now tested by the standard test suite.

• Compartmentalized Traverser interface so there is now a Traverser.System sub-interface with methods that users shouldn’t call.

• Added OrderByStep which orders Elements according to the value of a provided key.

• 2x speed increase on steps that rely heavily on ExpandableStepIterator with massive memory footprint reduction as well.

• Added VertexProperty<V> as the property type for vertices — provides multi-properties and properties on properties for vertices.

• Changed VertexProgram such that getElementComputeKeys() is simply a Set<String>.

• Significant changes to the format of the ResponseMessage for Gremlin Server - these changes break existing clients.

• Close any open transactions on any configured Graph when a session in Gremlin Server is killed.

• Grateful Dead Graph now uses vertex labels instead of "type" properties.

• There is now a GraphComputerStrategy and EngineDependent marker interface to allow steps to decide their algorithm depending if they are OLAP or OLTP.

• A labeled step now stores its current traverser value in Traversal.SideEffects (no longer can sideEffectKeys and step labels be the same).

• GraphFactory support for opening a Graph with multiple GraphStrategy instances - if there are multiple strategies they are wrapped in order via SequenceGraphStrategy.

• The result type for result termination messages returned from Gremlin Server is now set to "no content".

• The maxContentLength setting for Gremlin Driver now blocks incoming frames that are too large.

• After initialization scripts are executed in Gremlin Server, the Graph instances are re-bound back to their global references, thus allowing GraphStrategy initialization or even dynamic Graph creation through scripts.

• Added "Modern" graph back which is basically the "Classic" graph with double values for the "weight" property on edges and non-default vertex labels.

• Traversal.addStep() is now hard typed so type casting isn’t required and traversal APIs look clean.

• Changed the hidden key prefix from %$% to ~ in Graph.Key.hide().

• Added has(label,key,predicate,value) to allow for has('person','name','marko'). Various overloaded methods provided.

• Update to traversal API where if a SFunction<S,?> was required, but can process a Traverser<S>, then the function is SFunction<Traverser<S>,?>.

• Added WhereStep as a way to further constrain select() and match().

• Extensive work on GiraphMemory and its interaction with Giraph aggregators.

• If the input path of a GiraphGraphComputer does not exist, failure happens prior to job submission.

• SugarPlugin now has all inefficient methods and Gremlin-Groovy proper is only efficient Groovy techniques.

• Prevent concurrency problems by only modifying bindings within the same thread of execution in the GremlinExecutor.

• Calls to use on the DependencyManager now return the list of GremlinPlugin instances to initialize instead of just initializing them automatically because it causes problems with ScriptEngine setup if a plugin requires a script to be evaluated and a required dependency is not yet loaded.

• Traversal.SideEffects has getGraph(), setGraph(), and removeGraph() default helpers.

• Traversal.Memory → Traversal.SideEffects and GraphComputer.SideEffects → GraphComputer.Memory.

• StrategyWrappedVertex and StrategyWrappedEdge properly wrap Element objects returned from non-traversal based methods.

• Gremlin-Server now sends a single write with status 200 for Object and empty response messages.

• GremlinGroovyScriptEngine allows imports to re-import dependencies added via "use".

• Changed order in which the GremlinExecutor is initialized such that dependency loading via "use" are handled first which fixes problems with starting Gremlin Server with gremlin-server-neo4j.yaml.

• Corrected issues with the "branch factor" related traversals under SubgraphStrategy. This change also altered the semantics of the SubgraphStrategy a bit as it became more restrictive around Edge inclusion (requires both vertices to be in the subgraph).

• The Gephi Plugin now visualizes traversals and has numerous configuration options.

• Add more specific features around the types of "identifiers" a graph can support.

• Add a new test graph called MODERN that is copy of the CLASSIC graph, but represents floats as doubles. This graph will be the default graph for testing going forward.

• Fix bug in Neo4jGraph that was not processing multiple vertex labels properly when doing a has() step with IN.

• Changed semantics of @LoadGraphWith in gremlin-test to only refer to the ability of a test implementation to process the data types of the test graph (not to actually load it).

• StartStep is a SideEffect as it is a process to get data into the stream (like a keyboard) and more efficient as such.

• Greatly simplified the implementations of Map, FlatMap, Filter, and SideEffect.

• Path data structure changed to an ordered list of objects with each associated to a Set<String> of as-labels.

• All sideEffect-based steps no longer extend FilterStep with predicate equal true, but a more efficient SideEffectStep.

• TreeStep now has TreeMapReduce for executing on GraphComputer.

• Neo4jTraversal.cypher() is fluent throughout.

• Reverted back to TP2 model of as() referring to step names, not variable names of sideEffects.

• Updated AddEdge-step to support property key/value pairs for appending to newly created edges.

• Renamed Graph.getFeatures() to Graph.features() to be consistent with other API methods.

• Vertex and Edge now implement all GraphTraversal methods to ensure consistency throughout stack.

• Neo4jTraversal is auto-generated from Neo4jTraversalStub with technique generalizable to other vendors.

• Added test suite to ensure that all traversals are of the same type: g.V, g.E, g.of(), v.identity(), e.identity(), v-, e-methods.

• Giraph HDFS helpers now support hdfs.mkdir(string) and local.mkdir(string)

• Added @OptIn and @OptOut for implementers to specify on their Graph implementations for test compliance information.

• GraphComputer Memory now immutable after computation is complete.

• Dependency grabbing for plugins filter out slf4j logging dependencies so as to avoid multiple bindings with the standard TinkerPop distributions.

• Fixed GiraphMemory to be fully consistent with GraphComputer specification.

• Removed fatJar assembly from Giraph-Graph as it is no longed needed with distributed cache model.

• Reworked GiraphRemoteAcceptor to provide a result variable back to the console with ComputerResult.

• VertexProgram is no longer Serializable (use loadState and storeState for wire-propagation).

• Moved GiraphGraph.getOutputGraph() to GiraphHelper.

• Changed GIRAPH_GREMLIN_HOME to GIRAPH_GREMLIN_LIB to reference directory where jars are to be loaded.

• Updated README with release instructions.

TinkerPop 3.0.0.M1 (Release Date: August 12, 2014)

• First official release of TinkerPop3 and thus, no changes.