Sassafras

Today, we released the new version v2010.10 (Sassafras) of OGDF! This release brings various new algorithms and features.

Highlights

  • Completely revised memory management; pool-memory allocator now supports thread-safe allocation.
  • Basic multithreading support for OGDF algorithms.
  • New class System provides methods for accessing system-dependent information and functionality (processor features, memory usage, file system).
  • New hierarchical graph layout method using upward planarization (class UpwardPlanarization); outperforms traditional Sugiyama-based methods with respect to crossings by far.
  • New multilevel layout algorithm (FastMultipoleMultilevelEmbedder), based on the multipole method, well-separated pair decomposition, and a new quadtree space partitioning (Martin Gronemann’s Diploma thesis); makes use of SSE and multicore processors and is significantly faster than FMMMLayout.
  • New modular framework for multilevel graph layout (class ModularMultilevelMixer) with various options for coarsening, placement, and single-level layout.
  • New force-directed layout algorithms: Kamada-Kawai (class SpringEmbedderKK) and stress majorization (class StressMajorization).
  • New layout algorithms for directed graphs: DominanceLayout and VisibilityLayout.
  • Implementation of Prims’ algorithm for minimum spanning tree computation.
  • Added support for OGML graph file format.
  • Exact algorithms for computing maximum planar and maximum c-planar subgraphs (classes MaximumPlanarSubgraph and MaximumCPlanarSubgraph).
  • OGDF can now be compiled as DLL under Windows.
  • Support for Visual Studio 2010 and latest g++ compilers.

Detailed Changelog

New Features

  • Completely revised memory management, with a new thread-safe pool-memory allocator. OGDF provides now three allocators (which can be selected with compiler predefines):
    • thread-safe pool allocator (OGDF_MEMORY_POOL_TS; usually the deafult),
    • non-thread-safe pool allocator (OGDF_MEMORY_POOL_NTS), and
    • ordinary malloc/free (OGDF_MEMORY_MALLOC_TS).
  • Basic, platform-independent multithreading support for OGDF algorithms; new classes ThreadCriticalSectionMutexBarrier.
  • New class System provides methods for accessing system-dependent information and functionality (processor features, memory usage, file system).
  • New layout algorithms for directed (hierarchical) graphs:
    • DominanceLayout, based on dominance layout of st-digraphs.
    • VisibilityLayout, based on computation of a visibility representation.
  • New hierarchical graph layout method using upward planarization (class UpwardPlanarization); implements the algorithm by Chimani, Gutwenger, Mutzel and Wong. Outperforms traditional Sugiyama-based methods with respect to number of crossings by far. Uses module options for upward planarization and layout:
    • UpwardPlanarizerModule: its implementation SubgraphUpwardPlanarizer applies a 2-phase approach: first compute a feasible upward planar subgraph (FUPSModule with implementation FUPSSimple), then insert remaining edges (UpwardEdgeInserterModule with implementation FixedEmbeddingUpwardEdgeInserter).
    • UPRLayoutModule: implementation LayerBasedUPRLayout which makes use of hierarchy layout modules from the Sugiyama framework.
  • New force-directed layout algorithms:
    • Kamada-Kawai (class SpringEmbedderKK)
    • stress majorization (class StressMajorization).
  • New multilevel layout algorithm (FastMultipoleMultilevelEmbedder), based on the multipole method, well-separated pair decomposition, and a new quadtree space partitioning (Martin Gronemann’s Diploma thesis); makes use of SSE and multicore processors and is significantly faster than FMMMLayout.
  • New modular framework for multilevel graph layout (class ModularMultilevelMixer) with various options for coarsening, placement, and single-level layout.
    • MultilevelBuilder module (coarsening): EdgeCoverMergerIndependentSetMergerLocalBiconnectedMergerMatchingMergerRandomMergerSolarMerger.
    • InitialPlacer module (placement): BarycenterPlacerCirclePlacerMedianPlacerRandomPlacerSolarPlacerZeroPlacer.
  • Implementation of Prims’ algorithm for minimum spanning tree computation (function computeMinST in ogdf/basic/extended_graph_alg.h.
  • Added support for OGML graph file format:
    • class OgmlParser implements a parser for OGML-files;
    • methods writeOGML and readClusterGraphOGML in ClusterGraphAttributes provide reading and writing of OGML files.
  • Exact algorithm for computing a maximum planar subgraph (class MaximumPlanarSubgraph); requires COIN and ABACUS.
  • Exact algorithm for computing a maximum c-planar subgraph (class MaximumCPlanarSubgraph); requires COIN and ABACUS.

Minor Modifications

  • GraphAttributes provides now a flag indicating if the graph is directed or not (default is true): methods directed for getting / setting, support in reading and writing GML files (readGML and writeGML).
  • Hash values are now of type size_t for providing better compatibility with 64-bit systems.
  • Default parameters of GEMLayout revised.
  • ClusterPlanarizationLayout now allows to pass edge weights in its call, which are used for computing a c-planar subgraph (edges with low weight are preferred).
  • Clarified Comparer interfaces:
    • Class Comparer<E> renamed to VComparer<E> (since it relies on virtual functions).
    • Added the macros OGDF_AUGMENT_COMPARER and OGDF_AUGMENT_STATICCOMPARER to allow easy generation of comparers with full interfaces.
    • Added StdComparer (standard comparers) and TargetComparer (for comparing targets of pointers).
  • Array::grow() allows now to enlarge an array with empty index set.
  • PlanarizationLayout: changed default planar subgraph to FastPlanarSubgraph.
  • Changed return type of BCTree::findPathBCTree() to pointer (instead of reference) to reflect the fact that the returned object has been allocated with new.
  • isConnected()makeConnected(): improved performance by replacing recursive with iterative implementation.
  • New, more efficient implementation of list-based stacks.

Bug Fixes

  • Fixed bug in SugiyamaLayout: possibly incorrect setting of number of crossings if there are several connected components and arrangeCCs = true.
  • Fixed bug in BoyerMyrvold: incorrect embedding of self-loops.
  • Fixed memory leaks in PlanarAugmentationFixEmbedderMaxFaceEmbedderMaxFaceLayersEmbedderMinDepthEmbedderMinDepthMaxFaceEmbedderMinDepthMaxFaceEmbedderMinDepthPiTaPlanarizationLayout.
  • Fixed bug with copy constructor of graph arrays (NodeArrayEdgeArray etc.): crashed if copied array was not initialized for a graph.
  • Fixed bug with Hierarchy: memory leaks occurred when using initByNodes().
  • Revised compiler condition for systems providing only 15 random bits with rand() function; fixes a non-termination bug with FMMMLayout on Solaris/SPARC.
  • Fixed a bug in makeConnected().
  • Fixed a memory leak in LPSolver_coin.

Build System

  • Added support for Visual Studio 2010, which introduces an new project format (.vcxproj). For creating such a project, use makeVCXProj.py (instead of makeVCProj.py). At the moment, there is just one possible template file for VS 2010 (ogdf.vcxproj.vs2010.template; which is the default).
  • Added support for compiling OGDF as DLL under Windows (currently VS 2005 only); simply select project template ogdf.dll.vcproj.vs2005.template in makeVCProj.config.
  • Added support for latest g++ compilers (4.1.x – 4.4.x) on Linux and Mac OS X.
  • Windows only: OGDF now requires to also link against Psapi.lib.
  • g++ on Linux/Mac only: OGDF now requires to link against pthreads; add option -pthread when compiling and linking with g++.