Wailea is a set of command-line tools and a library of basic graph operations and graph algorithms mainly for drawing both undirected and directed graphs. It is written in C++14.
It takes a simple graph as its input, and generates combinatorial and geometric information required to draw the graph on 2-plane.
For undirected graphs, the inforamtion is given in xy-coordinates for the node labels, edges, and edge labels in a visibility representation where the nodes are represented by horizontal line segments and the edges are by vertical ones.
For directed graphs, the information is given in integer xy-coordinates of nodes.
It comes with a main library and 5 command-line tools as follows. Each command-line tool takes an input file in a designated format, and generates the result in text in another format.
- libs/libwailea.so
Main library.
- bin/decomposer: main source file
This takes a simple undirected graph and decomposes it into BC-tree.
- bin/planarizer: main source file
This takes a biconnected undirected graph and planarize it by inserting virtual nodes and splitting edges at crossings. The resultant graph is planar, but this command does not find an embedding.
- bin/biconnected_embedding_finder: main source file
This takes a planar biconnected graph and finds a planar embedding of it and its dual graph with faces and dual edges.
- bin/vis_rep_finder: main source file
This takes a BC-tree, planar embeddings for its blocks, information about how incident faces around each cut vertex are unified, and rectangular label dimensions for nodes and edges as its input. It produces a visibility representation of the underlying connected graph in the real xy-coordinates for the nodes and edges.
- bin/digraph_arranger main source file
This takes a single connected directed graph and arranges the nodes in the integer x/y-coordinates (rank/position).
$ make or $ make all and then $ make install.
This will get you the library and 5 command-line tools under /usr/local{lib,bin}.
NOTE: For Mac OS X of El Capitan and later, it seems the best way to handle private shared library is to put it under /user/local. Placing a shared library to an arbitrary location and let LD_LIBRARY_PATH point to it does not seem to work due to 'System integrity protection'. See: https://support.apple.com/en-us/HT204899
It comes with two sample drawers. Those are quick-and-dirty python scripts originally written to test libwailea. Sample drawings with their corresponding input files are shown below in this file.
This internally calls decomposer, planarizer,
biconnected_embedding_finder, and vis_rep_finder.
This internally calls digraph_arranger.
$ make unit_tests undirected
$ make unit_tests directed
These commands will run unit tests using Google Test.
Update GOOGLE_TEST_INC_DIR and GOOGLE_TEST_LIB_DIR in Makefile
according to your environment.
Wailea should work on any platform that supports C++14. However, I have tested it only on the following environment.
- Macbook Pro (Intel Corei5) macOS Sierra 10.12.6
If you want to run the sample drawers, you will need Python, numpy, and matplotlib.
Copyright (c) 2017 Shoichiro Yamanishi
Wailea is released under MIT license. See LICENSE for details.
For any inquiries, please contact: Shoichiro Yamanishi
Some drawings by the sample drawers and their corresponding input files are shown below.
-
Portability : No dependencies except for standard libraries. Based on C++ 14. Implementing graph algorithms will involve complex data structures with lots of interlinks. Most of the modern C++ textbooks strongly discourage the use of raw pointers. However, my experiment showed significant overhead when
unique_ptr<>,shared_ptr<>1, andweak_ptr<>are used. As a compromise, the linkage are implemented in iterators of standard containers, withdynamic_cast<>` as a runtime safeguard mechanism. -
Reliability : Unit tested with Google Test. 100,000 lines of test codes. 670 Test cases. 26700 EXPECT_* checks. No recursive calls at runtime.
-
Usability : Released under MIT licence.
-
May 2016 Development starts
-
Apr 2017 Undirected algorithms converge
-
Jun 2017 Directed algorithms converge
-
Oct 2017 Alpha release
-
TODO Documentation
Wailea employes the edge-insersion paradigm. The input graph G is assumed to be simple. a DFS algorithm decomposes G into connected components. For each connected component Gc, a DFS algorithm by Hopcroft and Tarjan (1973) decomposes it into a BC-tree. For each biconnected component Gb in the BC-tree, Wailea finds a planar spanning subgraph Gp and a set of complementary removed edges Er. A DFS algorithm by Tarjan [T85] first finds an st-ordering of the nodes in Gp, or N(Gp). Then the 1st phase of PQ-tree planarization algorithm [JTS89] finds the first version of Gp and Er. Gp at this stage may not be maximally planar. If Gp is still biconnected, then Wailea recalculates an st-ordering and then it tries to insert each edge e in Er to Gp to make Gp maximally planar. [BL76] is used to test the planarity of Gp + {e}.
Then, for each {u,v} in Er, the edge-insersion algorithm [GMW05] inserts {u,v} into Gp. This is a very complex process involving the following algorithms.
- BC-tree decomposition
- SPQR-tree decomposition
- Path finder on a tree
- Planar embedding finder
- Dual graph finder
- Shortest path finder
The basic idea is explained in [GMW05] and the implementation details are in the comments in gmw_edge_inserter.hpp.
At this stage, each Gb has been planarized. [BL76] finds a planar embedding of Gb and its dual. Then Wailea creates something called embedded BC-tree. It contains the planar embedding (and its dual) of each block as well as how the incident faces around each cut vertex will be unified. The details are found in embedded_bctree.hpp.
Then Wailea finds a visibility representation of Gc based on the information in the embedded BC-tree. The process is basically a recursive application of traditional visibility representaion for blocks. It also considers the node and edge labels. The node labels are centered at each node, and there are three types of edge labels: near node 1, in the middle of edge, and near node 2. And there are three types of edge label placement: center of the edge curve, touching the curve on CCW side, and touching the curve on CW side. The details are found in vis_rep_finder.hpp.
Finally for each Gc, we have a visibility representation in the xy-coordinates of the nodes, edges, node labels, and edge labels. The nodes are drawn as horizontal line segments with sufficient thickness to place the node label at their center. The area for each face is calculated such that the incident labels are places with sufficient margin. The edge curves are drawn as vertical line segments along those faces.
Wailea basically follows the subset of the steps as GraphViz's dot takes [GKNV] with its original twist. It's usually called heiearchical or Sugiyama drawing.
The first step is to make the digraph acyclic by flipping some of the edges. The next step is to assign ranks to each node. The third step is to reduce crossings between two adjacent ranks. GraphViz then assigns coordinates to vertex labels using network simplex, and then draws polynomial curves along the series of adjacent rectangular areas, but Wailea does not do them (yet). It generates the rank and position of each node as kind of integer coordinates and stops.
To make the graph acyclic Wailea takes the following steps.
-
Assign cyclic ranks to each node by network simplex such that the sum of (edge length * edge cost) is minimized. This can be considered a PO set on the nodes.
-
Serialize the cyclic ranks by breaking the ties at the same rank based on their in- and out-degrees of the nodes on the rank. This can be considered a cyclic total order.
-
Break the cycle at the gap between two ranks that has the minimum total edge cost. This makes the desired acyclic order and the edges against the order will be flipped to make the graph acyclic.
This problem of making a graph acyclic is called FAS (feedback arc set) problemand known to be NP-hard, and usually some heuristics are taken to make the graph acyclic. I think this approach presented above is amenable to declarative drawings, as it considers the edge cost. The details are given in acyclic_ordering_finder.hpp.
The next step is the rank assignment. It is done by network simplex. GraphViz solves this problem in the dual space, but Wailea does it in the primary space for both phase I and II, where phase I finds the initial feasible soluion and then phase II optimizes it. The details are given in network_simplex.hpp.
The final step is crossing reduction. This is again known to be NP-hard. Wailea follows the same approach as GraphViz, called barycenter heuristics with transposes. The details are given in [GKNV].
The following is the list of main algorithms implemented in libwailea.
-
ConnectedDecomposer connected_decomposer.hpp : Decomposes a given graph into connected components in DFS.
-
BiconnectedDecomposer bctree.hpp : Decomposes a given connected graph into BC-tree in DFS [RND77].
-
SPQRDecomposer spqr_decomposer.hpp : Decomposes a given biconnected graph into SPQR-tree in DFS. The original algorithm is proposed by [HT73], and then later it is corrected by [GM01]. It is still hard to understand and I have arranged a supplementary document here
-
STNumbering (undirected/st_numbering.hpp) : st_numbering.hpp : Generates an ST-ordering for a biconnected graph in DFS [T85].
-
JTSPlanarizer jts_planarizer.hpp : Classifies the edges of a given biconnected graph into a good planar subgraph and a set of removed edges. The first phase of [JTS89], which finds a planar spanning subgraph. The claimed maximality was refuted by manu, but I think the 1st phase is still useful to find a base planar graph and in many cases the subgraph is biconnected.
-
BLPlanarityTester bl_planarity_tester.hpp : Tests if a given biconnected graph is planar, and finds a combinatorial planar embedding of a given planar biconnected graph. [BL76] The original [BL76] algorithm requires s and t to be adjacent as in {s,t}. Wailea's implementation works for any node pair as long as the graph is biconnected. A relevant technical report is found here
-
PlanarDualGraphMaker planar_dual_graph_maker.hpp : Makes a dual graph structure of the given planar biconnected graph in a palanr embedding. My own linear time algorhitm as I could not find any in public.
-
GMWEdgeInserter gmw_edge_inserter.hpp : Inserts an edge into a given connected graph such that the number of crossings is minimized [GMW05].
-
EmbeddedBCTree (undirected/embedded_bctree.hpp) : embedded_bctree.hpp : Represents an embedding of a connected graph decomposed into a BC-tree. My own algorithm.
-
VisRepFinder vis_rep_finder.hpp : Generates a visibility representation of a connected graph. My own algorithm based on [TT86]. It handles not only biconnected graphs, but also connected graphs.
-
AcyclicOrderingFinder acyclic_ordering_finder.hpp : Finds an acyclic ordering of a given graph that may contain cycles. My own algorithm using network simplex as described above.
-
NetworkSimplex network_simplex.hpp : Solves the network simplex problem, which is a variant of integer linear programming. Both phase I and II are solved in the primary space. The phase I solves an aux problem to find an initial feasible solution, and the phase II takes it to an optimum. Used by AcyclicOrderingFinder and GKNV rank assignment in SugiyamaDiGraph.
-
GKNVcrossingsReducer gknv_crossings_reducer.hpp Tries to reduce the crossings in a connected bipartite digraph in a embedding with a barycenter and adjacent transpose heuristics. Employed in dot of GrpahViz. A bit crude but it gives pretty good results.
-
SugiyamaDiGraph sugiyama_digraph.hpp Finds integer xy-coordinates of the nodes of the given digraph in Sugiyama heiearchical framework.
-
[BL76] "Testing for the consecutive ones property, interval graphs, and graph planarity using PQ-tree algorithms", Kellogg S. Booth & George S. Lueker Journal of Computer and System Sciences archive Volume 13 Issue 3, December, 1976 Pages 335-379 Academic Press, Inc. Orlando, FL, USA
-
[GKNV] "A technique for drawing directed graphs." , E. R. Gansner, E. Koutsofios, S. C. North, and K. P Vo. IEEE Transactions on Software Engineering, 19(3):214–230, March 1993.
-
[GMW05] "Inserting an Edge into a Planar Graph", Carsten Gutwenger, Petra Mutzel, Rene Weiskircher Algorithmica 41(4):289-308, April 2005
-
[GM01] "A linear time implementation of SPQR trees", Gutwenger, Carsten & Mutzel Petra, Proc. 8th International Symposium on Graph Drawing (GD 2000), Lecture Nodes in Computer Science 1984, Springer-Verlag, pp. 77-90, doi:10.1007/3-540-44541-2_8
-
[HT73] "Dividing a graph into triconnected components", Hopcroft, John & Tarjan, Robert (1973), SIAM Journal on Computing 2 (3): 135-158, doi:10.1137/0202012
-
[JTS89] "O(n2) Algorithms for Graph Planarization", R. Jayakumar, K. Thulasiraman, M.N.S. Swamy, IEEE Transactions on Computer-aided Design, Vol 8. No. 3, March 1989
-
[RND77] "Combinatorial Algorithms Theory and Practice", E.M. Reingold, J Nievergelt, & N Deo, Prentice Hall (Dec. 1977) 978-0131524477 (Biconnected decomposition algorithm is found at https://www.cs.cmu.edu/~avrim/451f12/lectures/biconnected.pdf )
-
[TT86] "A Unified Approach to Visibility Representation of Planar Graphs", R. Tamassia and I.G. Tollis, Discrete Comput Geom 1:321-341 (1986)
-
[T85] "Two Streamlined Depth-First Search Algorthms", Robert Endre Tarjan Computer Science Department Princeton University Princeton, NJ 08544 and AT&T Bell Laboratories Murray Hill, NJ 07974 July, 1985 CS-TR-013-86 ftp://ftp.cs.princeton.edu/techreports/1985/013.pdf