An inductive representation of manipulating graph data structures.

Original website can be found at


* Hopefully clearer documentation for `&`, `Context` and the
`ufold`-based functions.

* Thanks to David Feuer, the existing benchmark suite is now runnable
with `cabal bench`.

* Some performance improvements for `PatriciaTree`, thanks to David

* Additional closure functions by Matthew Danish.

* `Bifunctor` instances for base >=

* An `ST`-based `GraphM` instance.

* Addition of `order` and `size` functions for finding the number of
nodes and edges respectively in a graph (the former is an alias for
the existing `noNodes` function).

* The rules for faster implementations of `insNode` and `insEdge` for
`PatriciaTree` should fire more often now.

* Earlier fix for `NFData` wasn't quite complete/correct.

* Ensure firing of specialised rules for `PatriciaTree`.

* Better way of only creating `NFData` instances when possible.

* Only create `NFData` instances for GHC >= 7.4.1.

* Documentation, clean-up and refactoring of various parts of the

- As part of this, various types now have instances for classes
like `Show`, `Eq`, `Ord`, `NFData`, etc. where applicable.

- In particular, the various options for use with depth-first
search and shortest path queries was documented by David

* Addition of a proper test-suite. So far it covers the
`Data.Graph.Inductive.Graph` module and all
`Data.Graph.Inductive.Query.*` modules except for `Monad`.

- The tests are also automatically run for every (set of) commits
thanks to Travis-CI.

* Arbitrary instances for the two graph types are now available in the
new `fgl-arbitrary` sub-package.

* Now depends solely on the `transformers` library rather than `mtl`.

* Potentially breaking changes:

These changes are those where the behaviour was un-specified or
didn't match the documentation.

- `nodeRange` and `nodeRangeM` for the various graph data
structures erroneously returned `(0,0)` for empty graphs (making
them indistinguishable from graphs containing the single node
`0`). They now match the default implementation of throwing an

- The behaviour of `delLEdge` when dealing with multiple edges was
unspecified; it now deletes only a single edge and the new
function `delAllLEdge` deletes all edges matching the one

* Additional functions thanks to Sergiu Ivanov:

- Creating sub-graphs by `Node`- and `Context`-filtering as well
as induced by a set of `Node`s.

- Graph condensation (i.e. graph of

- Various edge- and neighbor-based helper functions.

* The graph types now have `Generic` instances thanks to Piotr

* The `OrdGr` wrapper by Trevor Cook allows performing `Ord`-based
comparisons on graphs.

* Support added for GHC 7.10 by Herbert Valerio Riedel.

* Additional DFS query functions added by Conrad Parker.

* Repository location changed to GitHub.

* Code cleanup:

- Replaced usage of internal FiniteMap copy with Data.Map and
Data.Set from the containers library.

- Remove usage of data type contexts.

- Use newtypes where applicable.

* Fix up Eq instances for Tree and PatriciaTree so that they work with
multiple edges.

* Add proper Show, Read and Eq instances to Data.Graph.Inductive.Tree
and Data.Graph.Inductive.PatriciaTree.

* Add pretty-printing functions to Data.Graph.Inductive.Graph. These
are based upon the old Show implementation for

* Now use PatriciaTree by default rather than Tree (and recommend as
such). IntMap has been receiving a lot of optimisation work on it,
whereas the internal FiniteMap implementation hasn't received any

* The `version :: IO ()` action now uses the actual Cabal version.

* Remove Data.Graph.Inductive.Graphviz; use the graphviz package

* Update to work with GHC-7.2 and Cabal-1.6.

* Maintainership taken over by Ivan Miljenovic.

* Allow Data.Graph.Inductive.PatriciaTree to deal with multiple edges
between nodes. (November 2008)

* Bugfix in Graphviz.sq (June 2008)

* bug fix in bcc by Reid Barton

* added new dynamic graph implementation:
Data.Graph.Inductive.PatriciaTree (thanks to Pho)

* added test/benchmark.hs: a benchmark to compare Tree and PatriciaTree
implementations (thanks to Pho)

5.4.2 (May 2008)

* added Setup.hs to tar file

* reimplementation of Data.Graph.Inductive.Query.Dominators
by Bertram Felgenhauer:

It was buggy and very slow for large graphs. See

This patch also adds a new function, iDom, that returns the
immediate dominators of the graph nodes.

* Exported xdf*With functions from DFS.hs

* many little cleanups thanks to many people
(use 'darcs changes' to see the details)

5.4 (April 2007)

* changed the implementation for inspection functions (suc, pred, ...)
to correct the behavior in the presence of loops (thanks to Ralf
Juengling for pointing out the inconsistency)

5.3 (June 2006)

* fixed a bug in findP (thanks to

* added function delLEdge in Graph.hs (thanks to Jose Labra)

* changed implementation of updFM and mkGraph (thanks to Don Stewart)

February 2005

* fixed an import error in Basic.hs

* removed Eq instance of gr because it caused overlapping instance
problems. Instead the function equal defined in Graph.hs can be

* added some more functions to the export list of DFS.hs

* changed the definition of LPath into a newtype to avoid overlapping
instances with lists

* fixed the Makefile (for GHC and GHCi)

January 2004

* bug fix for nearestNode (src/Data/Graph/Inductive/Query/GVD.hs)
Update contributed by Aetion Technologies LLC (

* Refactor into hierarchical namespace

* Build changes:
- build a standard haskell library (libHSfgl.a, HSfgl.o)
- install as ghc package (fgl), uses Auto so no -package is needed

* Automatic Node generation for labels: Data.Graph.Inductive.NodeMap

* Graphviz output: Data.Graph.Inductive.Graphviz

September 2002

* Introduction of graph classes

* Monadic graphs and graph computation monad

* Graph implementation based on balanced (AVL) trees

* Fast graph implementation based on IO arrays

* New algorithms:
- Maximum flow
- Articulation points
- biconnected components
- dominators
- transitive closure

* minor changes in utility functions
- changed signatures (swapped order of arguments) of
functions context and lab to be consistent with other graph functions
- changed function first in RootPath: not existing path is now reported
as an empty list and will not produce an error
- esp version that returns a list of labeled edges
(to find minimum label in maxflow algorithm)
- BFS uses amortized O(1) queue
- Heap stores key and value separately
- ...

March 2001
* Changes to User Guide

* a couple of new functions

* some internal changes

April 2000

* User Guide

* Systematic structure for all depth-first search functions

* Graph Voronoi diagram

* Several small changes and additions in utility functions

February 2000

* Representation for inward-directed trees

* Breadth-first search

* Dijkstra's algorithm

* Minimum-spanning-tree algorithm

August 1999

* First Haskell version
comments powered byDisqus