As of 0.95 Graph is unsupported, sorry about that.  See README.

2013-05-24  Jarkko Hietaniemi  <>

	* Mop-up release for 0.95.  Still is and will be unsupported.

	* Release as 0.96.

2013-05-23  Jarkko Hietaniemi  <>

	* Address #85449:
	  "Graph-0.94 tests fail under perl 5.18.0"

	* Address #82324:
	  "Test failures due to hash randomisation in perl 5.17.6"

	The two above fixes were the same: the biconnectedness
	code was rewritten from scratch.  The new code behaves
	differently (but I believe more correctly) on certain
	edge cases, in general it will generate more biconnected
	components and bridges, for example for "a=b=c" it will
	now return the same two biconnected components and bridges
	(cut edges), namely "a=b" and "b=c", the "b" of course being
	the articulation point (cut vertex).

	* Address #67213:
	  "[PATCH] pod fixes"

	* Remove the t/u_bo.t and t/u_bo1.t since they die in 5.18 due
	  to some strange failure, looks unrelated to Graph as such,
	  probably some fix/change made by newer Perls. 

	* Release as 0.95.

2010-03-13  Jarkko Hietaniemi  <>

	* Address #43580:
	  "Reversed logic on overload::StrVal() in AdjacencyMap::Vertex::__set_path"
	  Had to add a new option, refvertexed_stringified.

	* Address #50210:
	  "Graph-0.91: bug in unionfind parameter"

	  One cannot delete from a unionfind graph: now enforcing that.

	* Address #48090:
	  "all_reachable on directed $g->add_edges(['a','b'],['b','a'])"

	  Now if there are loops, all_reachable() will include
	  the initial vertices themselves.  Also all_neighbors()
	  had some problems in certain cases, fixed those too.

	* Address #50210:
	  "Graph-0.91: bug in unionfind parameter"
	  One cannot delete edges or vertices from a unionfind
	  graph: now enforce that in code.

	* Address #42549: "stron"

	  Document that strongly connected components will include
	  isolated and sink and source vertices.

	* Release as 0.94.

2010-03-07  Jarkko Hietaniemi  <>

	* Revert the change made in Graph 0.92,
	  installing Heap 0.80 broke Graph.  Better be conservative.

	* Release as 0.93.

2010-03-03  Jarkko Hietaniemi  <>

	* Address #55912 "Broken links in the documentation"

	* Address #55196 "Heap 0.80 compatibility fix"

	* Add copyright and clearer license statement.

	* Release as 0.92.
2009-01-16  Jarkko Hietaniemi  <>

	* Minor documentation cleanups.

	* Add 'use strict;' to lib/Graph/

	* Modernize the META.yml.

	* Release as 0.91.

2008-12-29  Jarkko Hietaniemi  <>

	* Storable deparse of coderefs for deep_copy() does not
	  work at all with 5.6.2: if modern enough Storable
	  and B::Deparse are not available, fall back to
	  the previous version which used Data::Dumper.

	* Release as 0.90.

2008-12-27  Jarkko Hietaniemi  <>

	* Some PAUSE upload problem with 0.88, retrying.

	* Release as 0.89.

2008-12-26  Jarkko Hietaniemi  <>

	* The 0.87 forgot to specify the Storable (and Safe,
	  used in the deserialization step of deep_copy)
	  prerequirement(s) in Makefile.PL.

	* Release as 0.88.

2008-12-26  Jarkko Hietaniemi  <>

	* Addressed a performance problem in successors()
	  and predecessors(), reported by Jonathan Moore.

	* Reimplement deep_copy() by using Storable
	  freeze() and thaw() instead of Data::Dumper,
	  inspired by Jonathan Moore.  Probably now safer
	  and faster, but Storable is now a prerequirement.

	* Release as 0.87.

2008-11-27  Jarkko Hietaniemi  <>

	* Addressed a performance problem in connected_components()
	  for 1000+ vertex graphs, reported by David Grobe.
	  Should in general speed up graph traversal.

	* Release as 0.86.

2008-11-27  Jarkko Hietaniemi  <>

	* Address #31608 "Graph::Undirected, unionfind and

	* Address #34377 "recursive successors and predecessors"
	  (added all_successors/all_predecessors/all_neighbours/all_reachable)

	* Address #39444 "inconsistent return value"
	  (make add_edges and add_vertices to always return the graph)

	* Address #39614 "copy should retain more attributes"
	  (now copies also refvertexed/hypervertexed/countvertexed/

	* Address #39805 "UnionFind: Repeated adds clobbers
	  graph component information"

	* Address #41190 "add_edge_by_id on multigraph

	* Added betweenness(), clustering_coefficient(), and
	  subgraph_by_radius(), contributed by Matt Spear.

	* Release as 0.85.

2007-08-13  Jarkko Hietaniemi  <>

	* Tels found one more attributed edge problem.

	* Release as 0.84.

2007-08-12  Jarkko Hietaniemi  <>

	* One test in 73_diameter.t had too many possible answers,
	  dependent on the hash ordering, removed the test.

	* Release as 0.83.

2007-08-11  Jarkko Hietaniemi  <>

	* Since Heap 0.80 broke Graph, as a stop-gap measure
	  I will include the Heap::Elem and Heap::Fibonacci
	  of Heap 0.71, renamed as 'Heap071', addresses
	  #26943: "Heap 0.80 breaks Graph", and numerous bug reports
	  by email

	* Address #27840: "add-edge_attributes() on
	  undirected graph wrongly depends on node order", from Tels

	* Address #27959: "radius method incorrect",
	  code and test case from ROSULEK.

	* Release as 0.82.

2007-01-21  Jarkko Hietaniemi  <>

	* Address #24417: "next_successor unavailable in
	  Traversal (PATCH)", from Ted Carnahan.

	* Small pod tweaks.

	* Minor internal cleanup for the caching code.

	* Release as 0.81.

2006-09-10  Jarkko Hietaniemi  <>

	* SP_Bellman_Ford() used actually SPT_Dijkstra(), not
	  SPT_Bellman_Ford(), noticed by "aramos".  This changed
	  some regression test results a bit.

	* The NAME line of Graph::Undirected said "directed graphs",
	  noticed by "Ruslan", the first one to notice this in two
	  years, including the author yours truly...

	* Add Scalar::Util to the prereqs listed in Makefile.PL since
	  one of the tests uses refaddr, shouldn't be a problem because
	  List::Util already is a prereq.

	* Fix few broken intra-pod links.

	* Release as 0.80.

2006-08-06  Jarkko Hietaniemi  <>

	* The u_bo_ap1.t wasn't really testing the same for 20 times,
	  which meant that one bug was waiting for Koen van der Drift.
	  (If one start vertex of biconnectivity search was a self-loop,
	  an empty list of articulation points was returned.)

	* Add a new API family: $g->..._clear_cache(), which allows
	  one to forget a cached value such as biconnectivity().
	  Without clearing the cache once the result has been computed
	  it stays the same (until the graph is modified, of course).
	  If the cache is cleared, (pseudo)randomness is applied again,
	  and the algorithm results may become different.

	* Release as 0.79.

2006-07-16  Jarkko Hietaniemi  <>

	* Address #20476: "SPT_Bellman_Ford does not respect
	  refvertexed" - now fixed, and SPT_Dijkstra() had the same problem.

	* Release as 0.78.

2006-07-08  Jarkko Hietaniemi  <>

	* Address #20185: "problem with SPT_Bellman_Ford",
	  SPT_Bellman_Ford() was broken for undirected graphs
	  (they were handled as directed ones, therefore missing vertices).

	* weakly_connected_component_by_vertex() usage example
	  was wrong, was using weakly_connected_component(),
	  noted by 'yanick' in

	* Implement and document the saving of the SPT_Dijkstra()
	  and SPT_Bellman_Ford() start vertices.

	* Document add_edges() alternative API.

	* Aerate the Changes by adding empty lines between the * items.

	* Release as 0.77.

2006-06-28  Jarkko Hietaniemi  <>

	* Problem found by Xiaoli Zheng in diameter(): adding vertices
	  did not change diameter, this was due to a deeper bug where
	  the transitive closure matrix was being cached wrong:
	  sometimes saved too long, sometimes recomputed too often.
	  Enhanced 73_diameter.t to detect this.

	* Problem found by Andree Toonk and Ronald van der Pol:
	  SP_Dijkstra() tried too hard to find a path between
	  vertices even if there was none - and returned rubbish.
	  Added t/u_at3.t to detect this.

	* Address #20021: "bridges() sometimes returns
	  empty list when isolated vertices present".  biconnectivity()
	  did not work right if isolated vertices were picked as roots,
	  it either hung or returned empty.  Added t/u_bill.t to detect this.

	* Document that add_vertex() is often unnecessary.

	* and were missing "use strict".

	* Release as 0.76.

2006-06-09  Jarkko Hietaniemi  <>

	* Had accidentally removed Digest::MD5 from the Makefile.PL
	  prereq list, found by Anton Berezin (using Perl 5.6.2, which
	  doesn't include that), solved by removing the dependency
	  to Digest::MD5.

	* Speeded up repeated longest/shortest paths computations
	  by using a cached (Floyd-Warshall) transitive closure.

	* Implemented:
	  - graph radius
	  - graph center (vertices)
	  - vertex eccentricity

	* Release as 0.75.

2006-05-31  Jarkko Hietaniemi  <>

	* Bug in SP_Dijkstra() found by Andree Toonk and Ronald van der Pol.
	  The edge weights of the Dijkstra shortest paths graph were
	  not cumulative (if the whole graph is a-b = 1 and b-c = 2,
	  a-c should be 3), which caused the SP_Dijsktra() results
	  sometimes to be nonsense.  Two test cases, one of them rather
	  large (about 5000 edges).

	* Bugs when using references as vertices in bridges()
	  and (Traversal) seen() found by Brian Osborne.

	* Added another articulation_points() test by Brian Obsorne.

	* Explicitly disallow adding undef as a vertex.

	* Release as 0.74.

2006-05-27  Jarkko Hietaniemi  <>

	* Still one bug hiding in articulation points: if the
	  (randomly chosen) first vertex was a self-loop, an
	  empty list was returned for articulation points.
	  t/u_bo_ap.t now tests the test case from Brian
	  Osborne 20 times to stress test more cases,
	  and extra five tests testing self-loops and
	  articulation points.

	* Release as 0.73.

2006-05-27  Jarkko Hietaniemi  <>

	* Brian Osborne found a graph where articulation_points()
	  ended up in an infinite loop.	 Resolved and the graph
	  test case added as t/u_bo_ap.t.

	* Release as 0.72.

2006-05-22  Jarkko Hietaniemi  <>

	* Tweak the pod-coverage.t so that it looks more like
	  Test::Pod::Coverage documentation suggests in this case.

	* Fix the u_bo.t not to have a test class with a broken
	  stringification method to avoid spurious warnings and
	  failure (also do away with the use of Math::Complex to
	  avoid problems because of different Math::Complex releases),
	  and more even importantly fix the "next_root" logic in
	  connected components not to advance to the next component
	  if there is nothing to advance to.  This seems to be prone
	  to failure in 5.6.2, for some reason 5.8.8 works fine.

	* Test under Perl 5.6.2.

	* Force has_cycle() to return true/false, not the list of edges,
	  reported by Casey Bergman.

	* Release as 0.71.

2006-05-21  Jarkko Hietaniemi  <>

	* delete_vertex() from a refvertexed graph left an unnecessary
	  reference to the referenced vertex hanging around in the graph,
	  reported by Christoph Lamprecht.

	* Implement new 'super_component' option for connected_graph(),
	  biconnected_graph(), and strongly_connected_graph(), to allow
	  more complex ways of forming 'supercomponents' (and more
	  customized ways of naming them).

	* Address #17159: "Nodes appear to unblessed after
	  using articulation_points() - 2" (elaboration of
	  #17108: "Nodes appear to unblessed after using

	* Address #17160: "Nodes appear to unblessed after
	  using connected_components()"

	* Address #17161: "Nodes appear to unblessed after
	  using bridges()"

	* Address #17162: "Nodes appear to unblessed after
	  using connected_graph()"

	* Address #17163, "SP_Dijkstra() is complaining"

	* Address #17164, "SP_Bellman_Ford() is complaining"

	* Address #17165, documentation error in

	* Address #17405: "has_cycle with empty args
	  should return FALSE"

	* Address #17592: "articulation_points doesn't
	  find all vertices" (didn't find all the vertices of non-connected
	  graphs, only the vertices of the first (randomly chosen) connected

	The cases 17159-17592 reported by Brian Obsorne.

	* Add Test::Pod and Test::Pod::Coverage tests.

	* Release as 0.70.

2005-12-06  Jarkko Hietaniemi  <>

	* Add SP_Dijkstra() and SP_Bellman_Ford() to find the shortest
	  path between any two vertices, the result is returned as
	  the list of the vertices in the path.

	* In addition to the SPT per vertex result weight, also add
	  a predecessor ('p') vertex attribute (the SP_Dijkstra() and
	  SP_Bellman_Ford() unsurprisingly use this.)

	* Cache the SPT results for better speed.

	* Document that the SPT also allow a single argument
	  as the starting (root) vertex.

	* Fix a bug in SPT_Dijkstra() which would ignore an "untrue" vertex
	  (such as '0') if it was any other vertex than the root vertex
	  (boolean context is dangerous, when you really mean "exists").

	* For "components" (strongly, biconnected, and connected) graphs
	  store the list of the original vertices as a vertex attribute
	  'subvertices' (so there is no need to do split(/\+/, ...) tricks),
	  the list is stored as a array reference.

	* Release as 0.69.

2005-11-23  Jarkko Hietaniemi  <>

	* SPT_Dijkstra() wasn't setting the vertex attributes of
	  the result graph, noticed by Susan Tang, only the edge
	  attributes were being set.  SPT_Bellman_Ford() was doing neither!

	* There was an actual typo in the SPT test case from Sedgewick,
	  a weight of 0.32 was mistyped as 0.22, this luckily didn't
	  affect the result graph but it of course affected the
	  resulting vertex 'weight' attributes.

	* Add tests to t/70_spt.t for the vertex and egde attributes
	  of the SPT_Dijkstra() and SPT_Bellman_Ford() results.

	* Minor documentation tweaks, most importantly clarify the
	  return value of the SPT_Dijkstra() and SPT_Bellman_Ford().

	* Document that Perl 5.6.0 is the minimum (because of weak references)
	  and also make require that (Makefile.PL was already doing
	  the probing using Scalar::Util qw(weaken)).

	* Add an early test (02_trap.t) for catching the development-time-only
	  setting of __DIE__ and __WARN__ handlers (as a result of this almost
	  all the numbered tests were renumbered, so the diff is falsely
	  gigantic).  (If the handlers were mistakenly left turned on,
	  a lot of later tests that checked the $@ got confusing failures.)

	* Release as 0.68.

2005-08-03  Jarkko Hietaniemi  <>

	* The 0.66 add_edge_get_id() fix was not yet quite right, Tels
	  found another problem with it.  Now with another fix, and
	  another test case (t/u_te_ae.t)

	* Documentation fixes from John P. Linderman.

	* Release as 0.67.

2005-07-20  Jarkko Hietaniemi  <>

	* Fix [ #13193] "Documentation error in set_edge_attributes"
	  and [ #13194] "Documentation error in set_edge_attributes"
	  (duplicate report)

	* Fixes for problems listed in [ #13195]
	  "add_vertex_get_id/add_edge_get_id() return wrong result on first call"
	  - add_edge_get_id() was returning an array reference instead
	    of the id with the first call (the array reference was the
	    ids of the vertices of the edge)
	  - add_vertex_get_id() was even more broken (a multivertexed
	    graph was using Graph::AdjacencyMap::Vertex for the vertex
	    map, not Graph::AdjacencyMap::Heavy)
	  - Added test t/u_te_me.t for the two above issues.
	  - document in which order multiedge ids are returned (random)
	  - require Data::Dumper only for deep_copy() and _dump()
	  (not changes for two listed items, "check directly multiedged
	   via a flag" and "remove returns for speed" because I have
	   issues with speed hacks without actual measurements, and even
	   if so would fear reduced maintainability)

	* Fix [ #13352] "Dijkstra heap logic"
	  Dijkstra was fine, the SPTHealElem cmp() routine was wrong
	  in having no tie breakers in case the weights compared equal.
	  Added test t/u_re_sd.t.

	* Release as 0.66.

2005-05-15  Jarkko Hietaniemi  <>

	* Tests added to 64_ref.t to verify that using different kinds
	  of blessed references as vertices works okay.	 Few bugs found
	  by these tests squashed.

	* Release as 0.65.

2005-05-14  Jarkko Hietaniemi  <>

	* Fix for [ #12509] "Errors using objects as nodes",
	  patch from the reporter of the bug, add t/u/bb_rv.t.

	* Fix for refvertexed isolated vertices not having overloaded cmp
	  and graph string presentation failing because of that.

	* The <NOTE>s needed to be B<NOTE>s.

	* Release as 0.64.

2005-04-16  Jarkko Hietaniemi  <>

	* After setting a vertex attribute one could not delete
	  non-attributed vertices, reported by Joseph Hamilton.

	* Inlining to speed up path_vertices() slightly.

	* Release as 0.63.

2005-04-10  Jarkko Hietaniemi  <>

	* The documentation of add_weighted_vertices was wrong:
	  the arguments are (v1, w1, v2, w2, ...) instead of (v1, v2, ..., w).

	* Made calling interfaces with an "options hash" like new()
	  and random_graph() more robust, now bails out earlier instead
	  of dieing mysteriously later with an "odd number of arguments"

	* Allow running under -d:DProf even when using random shuffling:
	  workaround for List::Util::shuffle and -d:DProf not working
	  together ([perl #32383]) by falling back to Fisher-Yates shuffle
	  if (any use of) the -d: is detected.

	* Allow calling random_graph() also as a class method:
	  Graph::random_graph(...) (the resulting graph will be a 'Graph').

	* in_degree() and out_degree() (and therefore vertex_degree())
	  were one too low for self-loop vertices in undirected graphs
	  (the self-loop edge was not counted).

	* Release as 0.62.

2005-03-27  Jarkko Hietaniemi  <>

	* [ #12023] from Macha Nikolski:
	  deleting an attributed vertex left the graph in a state
	  where has_vertex() returned correctly false but vertices()
	  still wrongly returned the freshly deleted vertex.

	* A few missing "See":s added to the pod.

	* Release as 0.61.

2005-03-25  Jarkko Hietaniemi  <>

	* Bug reported by Richard Ball: connected_component_by_index()
	  and connected_component_by_vertex() were starting their indexing
	  from one, not zero.

	* t/27_hyperedged.t was really testing for turning on
	  hypervertexedness (the actual functionality was being
	  tested correctly in t/32_hyperedge.t).

	* Release as 0.60.

2005-03-03  Jarkko Hietaniemi  <>

	* deep_copy_graph() could not handle code references since
	  Data::Dumper by default doesn't handle those.	 Now uses
	  the Deparse option for 5.8.x and later.

	* The removed interfaces add_graph() and delete_graph() still
	  had their documentation hanging around.

	* Release as 0.59.

2005-02-19  Jarkko Hietaniemi  <>

	* Document that using attributes does have a slowing down
	  effect on other graph operations
	  [ #11498]
	  "Performance problem: edge attributes slow source_vertices"
	  This is unlikely to get fixed any time soon, I am afraid,
	  this is one of those working-as-designed-and-correctly-but-
	  unfortunately-slow things.

	* Document that Graph 0.2xxx edges($v) is now edges_at($v)
	  [ #11494]

	* [ #11543]: self-edges reported twice by edges_at().

	* Declare/document that any attributes beginning with an underscore
	  are reserved for the internal use of Graph.

	* Various inlining optimizations: should run 5-10% faster
	  than the 0.57.

	* Release as 0.58.

2005-02-12  Jarkko Hietaniemi  <>

	* Further 10% speedup on 'make test' on top of 0.56 by inlining
	  various code paths related to finding edges, now 'make test'
	  is cumulatively about 15% faster than the 0.55 release.
	  The test case of [ #11465] is about 10 times faster.

	* Release as 0.57.

2005-02-12  Jarkko Hietaniemi  <>

	* Rewrite edges finding code (like edges_at()) to avoid a
	  quadratic algorithm.	Shame on me.  Luckily this extremely
	  slow path was not touched that often, but [ #11465]
	  shows one known bad case, source_vertices() for compat02
	  graphs.  The removal of the slow path sped up 'make test'
	  by about 5-10%.

	* Remove a voodoo keys() from vertices_at().

	* Document stubs for Graph::Directed and Graph::Undirected.
	* Tiny documentation tweaks.

	* Release as 0.56.

2005-01-22  Jarkko Hietaniemi  <>

	* Add unset_row(), get_row(), set_row(), and unset_row(), methods
	  to Graph::BitMatrix and make it public (remove the "internal use
	  only" warning from it).  Add t/82_bitmatrix.t.

	* Add vertex_degree() as an alias for degree().

	* One more alternative solution for spt.t from Koen.

	* I seem to have this drive to misspell people's names.
	  Sorry, Koen.

	* Release as 0.55.

2005-01-16  Jarkko Hietaniemi  <>

	* More bugs found in set_vertex_attribute(), fixed and tests
	  added.  (Basically the same failure pattern as with the
	  [ #9461]: after setting vertex attributes many of
	  the 'structural' methods such as predecessors() often returned
	  wrong results.)

	* More alternative solutions to spt.t, diameter.t, and dump.t,
	  found by the PRNG of Koen van der Drift in Mac OS X 10.3.7,
	  Perl 5.8.1.

	* Release as 0.54.

2005-01-14  Jarkko Hietaniemi  <>

	* The #9461 was still there.
	  But now we have a simple test case from Sebastian Nagel.
	  The real culprit seemed to be a misapplied optimisation.

	* Release as 0.53.

2005-01-12  Jarkko Hietaniemi  <>

	* Fix set_graph_attribute() documentation not to talk about $u, $v
	  (noticed by Kurt Jaeger).

	* A mysterious failure fixed by a mysterious fix: under some
	  circumstances it seems that an each() doesn't walk through
	  all the key-value pairs, the workaround is to reset the
	  each() iterator by a keys() call.  Not simple test code,
	  sadly, since the existing test code (see the case) is 13 kB
	  and non-trivial.
	  [ #9461]

	* Add a safety guard against a missing Scalar::Util::weaken
	  [ #9481]

	* Release as 0.52.

2005-01-09  Jarkko Hietaniemi  <>

	* Allow calling Makefile.PL with arguments other than --renum
	  (which is for internal use only, and therefore undocumented).
	  [ #9481]

	* Remove the add_graph() and delete_graph() interfaces, sorry
	  if you were already using them, but the current interface was
	  very poor and the concept ill-planned.  If you want to merge or
	  remove edges and vertices between your graph, you can probably
	  yourself implement the exactly right things to do.
	  [ #9493]

	* Document that one cannot assume Graphs are blessed hash references
	  (and the likely error message one will get if one so assumes).
	  [ #9505]

	* Fix Andras' last name (sorry).

	* Merge duplicate documentation of find_a_cycle(). 

	* Graph::AdjacencyMap::Base does not exist, fix Graph/
	  pod to comply.

	* Release as 0.51.

2005-01-01  Jarkko Hietaniemi  <>

	* The 0.50.

2004-10-30  Jarkko Hietaniemi  <>

	* Start wrapping up for the 0.50 release.

	* Start bothering beta testers.