../../README.md "OGDF" » ../dev-guide.md "Developer's Guide" » ../porting.md "Porting Guide" » Catalpa
OGDF now requires C++11 features. Make sure you do not use another standard when compiling your user programs.
If you want to compile for debug mode (even if it is user code, linking against OGDF), you do not need to define OGDF_DEBUG
any longer. This is automatically defined using CMake.
It might be necessary that you need to include additional header files for code that worked before. For example, #include <ogdf/basic/NodeArray.h>
does not ensure that you also have an EdgeArray
.
We also removed header files that contained declarations of classes whose implementation has already been removed in former releases. This includes FixedUpwardEmbeddingInserter.h
.
The module
subdirectory has been dissolved. Please look for the module classes in the subdirectories of their implementations. LayoutModule.h
has been moved into the basic
subdirectory.
The directory structure for internal headers (for example, helper classes) has changed.
In former versions of the OGDF some symbols were added to the global namespace. This includes but is not limited to ifstream
, ofstream
, cout
, ios
, endl
, and numeric_limits
. Thus, you may be required to explicitly use the std
namespace.
Also many std
functions and classes that were also available in the ogdf
namespace are now only available in the std
namespace. For example, string-to-number conversion functions like std::stoul
cannot be called using ogdf::stoul
any longer.
When a AlgorithmFailureException
was only thrown in debug mode, it is now replaced by an assertion (OGDF_ASSERT
). PreconditionViolationException
have been completely replaced by assertions.
All enums are now enum classes or const static. Since the enumerators are scoped, the distinguishing prefixes of the enumerators have been dropped. Unscoped enums are now forbidden in the OGDF.
We have removed Symphony and Cgl from the included COIN-OR package. The only LP solver we ship is hence Clp.
If your code used Cgl or Symphony, you have to link your objects to the original versions of these libraries now.
Some macros have been removed because they are not necessary any longer.
This includes
OGDF_NEW
(simply use new
instead)ABACUS_LP_OSI
(Abacus always uses COIN)OGDF_NO_COMPILER_TLS
(thread-local storage is a C++11 feature)OGDF_DECL_THREAD
(simply use thread_local
instead)We also have removed most of the iteration macros since C++11 offers range-based for loops. This includes the removal of forall_nodes
and forall_edges
.
should be replaced by
OGDF_ASSERT_IF
was replaced by OGDF_HEAVY_ASSERT
. The debug level argument was dropped.
There must now be a semicolon after usage of the macro OGDF_ASSERT
and OGDF_HEAVY_ASSERT
.
To access the st-numbering functions, include ogdf/basic/STNumbering.h
(instead of ogdf/basic/extended_graph_alg.h
).
The respective functions were renamed:
Former | New |
---|---|
stNumber | computeSTNumbering |
testSTnumber | isSTNumbering |
The DTreeMultilevelEmbedder
header file is moved from include/ogdf/internal/energybased/
to include/ogdf/energybased
because it is not internal.
Its internal header files however have moved to include/ogdf/energybased/dtree
.
The class MixedForceLayout
was removed. Instead of calling this layout algorithm, call the FastMultipoleEmbedder
directly followed by a call to the SpringEmbedderGridVariant
.
The (formerly internal) class TricComp
was renamed to Triconnectivity
and can now be found in include/ogdf/graphalg/Triconnectivity.h
.
The method writeGML(const char*, GraphAttributes&, bool)
was deleted. Use writeGML(ostream &os, const GraphAttributes&)
instead.
The random planar graph generators got a random
prefix. All directed graph generators are now called Digraph
instead of DiGraph
to keep consistency in capitalization across the OGDF. This affects the following generators:
Former | New |
---|---|
randomDiGraph | randomDigraph |
planarConnectedGraph | randomPlanarConnectedGraph |
planarBiconnectedGraph | randomPlanarBiconnectedGraph |
planarBiconnectedDiGraph | randomPlanarBiconnectedDigraph |
upwardPlanarBiconnectedDiGraph | randomUpwardPlanarBiconnectedDigraph |
planarCNBGraph | randomPlanarCNBGraph |
planarTriconnectedGraph | randomPlanarTriconnectedGraph |
Additionally, if you require only either deterministic or randomized generators, you can now include include/ogdf/basic/graph_generators/deterministic.h
and include/ogdf/basic/graph_generators/randomized.h
respectively.
The parameter BlockOrder* order
was deleted from the constructors of Block
in BlockOrder.h
. Use the constructors Block(edge e)
and Block(node v)
instead.
NodeElement::adjEdges
was renamed to NodeElement::adjEntries
. This means you have to change for(adjEntry adj : v->adjEdges)
to for(adjEntry adj : v->adjEntries)
.
The following getter-methods were moved from Graph
to NodeElement
. All of these methods take a single list as the only parameter now.
adjEntries()
(renamed to allAdjEntries()
)adjEdges()
inEdges()
outEdges()
Hiding and restoring edges was not a safe operation in former releases. Replace code like
by
All edges are restored by hiddenSet.restore()
or automatically by the HiddenEdgeSet
destructor.
The method collaps
has been renamed to collapse
.
The static method nextPower2
was slightly changed and moved to Math
. The computed result is no longer strictly larger than the second parameter. nextPower2(0)
is now 0
instead of 1
.
The newEdge()
methods to add edges at predefined positions in the adjacency list have been removed from GraphCopy
. You can instead use the respective newEdge()
function inherited from Graph
, followed by GraphCopy::setEdge()
.
The methods rbegin()
and rend()
were removed from all GraphArrayIterator
-classes, i.e. from AdjEntryArray
, ClusterArray
, EdgeArray
, FaceArray
and NodeArray
. Iterating over their elements in any order can be done by iterating over their respective keys in this order.
The methods rbegin()
and rend()
of all container classes now return reverse iterators. succ()
, operator++()
, &operator++()
and cyclicSucc()
of reverse iterators behave like pred()
, operator--()
, &operator--()
and cyclicPred()
of normal iterators respectively, and vice versa.
The classes Stack
, StackPure
and BoundedStack
are deleted. Use ArrayBuffer
instead (which provides the functionality of all three classes (and more) and outperforms all implementations). Change pop()
in your implementations to popRet()
.
The deprecated methods List::exchange()
and ListPure::exchange()
were removed. Use List::swap()
and ListPure::swap()
instead.
The rbegin()
-methods of the classes SList
, SListPure
, Queue
and QueuePure
were replaced by backIterator()
-methods. The name rbegin()
was misleading since the methods do not return reverse iterators.
BoyerMyrvoldSubgraph
was renamed to PlanarSubgraphBoyerMyrvold
and FastPlanarSubgraph
to PlanarSubgraphFast
. MaximalPlanarSubgraphSimple
now supports arbitrary start heuristics. As such, the doMaximize
option was removed from PlanarSubgraphBoyerMyrvold
(formerly BoyerMyrvoldSubgraph
). MaximalPlanarSubgraphSimple::clone()
should be used to obtain a copy of the respective instance since MaximalPlanarSubgraphSimple(const MaximalPlanarSubgraphSimple &mps)
is no longer available.
Support for edge weights was added to MaximumCPlanarSubgraph
. The signature of call
includes the new parameter EdgeArray<int> *pCost
that can be set to nullptr
for uniform weight.
The Module PlanarSubgraphModule
is now a template. The following implementations are templates themselves: PlanarSubgraphCactus
, MaximalPlanarSubgraphSimple
, PlanarSubgraphEmpty
, PlanarSubgraphFast
, MaximumPlanarSubgraph
. All other implementations of the module inherit from PlanarSubgraphModule<int>
.
MaximumCPlanarSubgraph
supports advanced calls that also return the edges required to connect the input graph. To avoid conflicts with the method defined by CPlanarSubgraphModule
, MaximumCPlanarSubgraph::call
was renamed to MaximumCPlanarSubgraph::callAndConnect
.
The signature of call()
changed. The last two parameters List<NodePair>& npEdges
and List<edge>& newEdges
became the single parameter List<edge>& origEdges
that contains all edges in the original that still need to be inserted. The original graph must now contain all edges prior to the call. This also means that the input graph is no longer modified (except for embedding it).
NodeComparer
was removed. Instead of NodeComparer<T> foo(array, asc)
use GenericComparer<node, T, asc> foo(array)
. IndexComparer
also was removed. Use GenericComparer<node, int>([](node v) { return v->index(); });
instead. If you need a default constructor you may use OGDF_DECLARE_COMPARER
.
In an attempt to unify the existing heap classes, BinaryHeap
and BinaryHeap2
were replaced by a new BinaryHeap
. All heap implementations are derived from the common interface HeapBase
. Additionally, the classes MinPriorityQueue
and PQueue
were removed.
The following heap implementations were introduced:
BinaryHeap
BinomialHeap
FibonacciHeap
PairingHeap
RadixHeap
Priority queues might be realized using the newly introduced PriorityQueue
independently of the desired heap implementation. In contrast to PriorityQueue
that merely stores directly comparable elements, PrioritizedQueue
is used to store elements with priorities assigned upon insertion. For even simpler usage see PrioritizedMapQueue
that keeps track of handles for the inserted elements but requires elements to be unique.
The tests in test/src/basic/heap.cpp
show exemplary usage of the new classes.
BinaryHeap
was replaced by PrioritizedQueue
. Accessing elements at arbitrary positions is no longer supported.
Former | New |
---|---|
BinaryHeap::clear | PrioritizedQueue::clear |
BinaryHeap::decPriority | PrioritizedQueue::decrease |
BinaryHeap::empty | PrioritizedQueue::empty |
BinaryHeap::extractMin | PrioritizedQueue::topElement & PrioritizedQueue::pop |
BinaryHeap::getMin | PrioritizedQueue::topElement |
BinaryHeap::insert | PrioritizedQueue::push |
BinaryHeap::operator[] | - |
BinaryHeap::pop | PrioritizedQueue::pop |
BinaryHeap::push | PrioritizedQueue::push |
BinaryHeap::size | PrioritizedQueue::size |
BinaryHeap::top | PrioritizedQueue::topElement |
BinaryHeap2
was replaced by PrioritizedMapQueue
. Querying the capacity of the heap is no longer supported.
Former | New |
---|---|
BinaryHeap2::capacity | - |
BinaryHeap2::clear | PrioritizedMapQueue::clear |
BinaryHeap2::decreaseKey | PrioritizedMapQueue::decrease |
BinaryHeap2::empty | PrioritizedMapQueue::empty |
BinaryHeap2::extractMin | PrioritizedMapQueue::topElement & PrioritizedMapQueue::pop |
BinaryHeap2::getMinimumKey | PrioritizedMapQueue::topPriority |
BinaryHeap2::getPriority | PrioritizedMapQueue::priority |
BinaryHeap2::insert | PrioritizedMapQueue::push |
BinaryHeap2::size | PrioritizedMapQueue::size |
BinaryHeap2::topElement | PrioritizedMapQueue::topElement |
PQueue
was replaced by PrioritizedQueue
.
Former | New |
---|---|
PQueue::del_min | PrioritizedQueue::pop |
PQueue::find_min | PrioritizedQueue::top |
PQueue::insert | PrioritizedQueue::push |
MinPriorityQueue
was replaced by PrioritizedQueue
. Note that PrioritizedQueue::size
returns the number of elements in the heap instead of the capacity.
Former | New |
---|---|
MinPriorityQueue::count | PrioritizedQueue::size |
MinPriorityQueue::decreasePriority | PrioritizedQueue::decrease |
MinPriorityQueue::empty | PrioritizedQueue::empty |
MinPriorityQueue::getMin | PrioritizedQueue::top |
MinPriorityQueue::insert | PrioritizedQueue::push |
MinPriorityQueue::pop | PrioritizedQueue::pop |
MinPriorityQueue::size | - |
Layout algorithms no longer support GraphConstraints
. Use call(GraphAttributes)
instead of call(GraphAttributes, GraphConstraints)
. Graph constraints were removed entirely. Removed classes: Constraint
, ConstraintManager
, and GraphConstraints
.
edgeLengths()
, numberOfBends()
, angularResolution()
and numberOfCrossings()
now return ArrayBuffer
s with values for each edge/angle/node. The mininum, maximum, mean and standard deviation can no longer be obtained using function parameters but by applying Math
-functions to the returned values.
Moreover, angularResolution()
was renamed to angles()
.
Methods that take filenames (instead of streams) are removed. For example, instead of using
you have to create the streams manually, i.e.,
There are also helper functions GraphIO::read()
and GraphIO::write()
that accept filenames for GraphAttribute
s and Graph
s, so you can also write
Do not confuse the new filename-based GraphIO::read()
helper function with our new (experimental) generic reader function of the same name. However, this generic reader even allows to write
The keywords for nodeshapes changed for some shapes.
Shape | Former | New |
---|---|---|
Shape::RoundedRect | "rect" | "roundedrect" |
Shape::InvParallelogram | "parallelogram" | "invparallelogram" |
Shape::Image | "box" | "image" |
This can result in these node shapes not being recognized by other programs when parsed, but provides better OGDF to OGDF parsing and writing.
edgeArrow is now "string" instead of "int".
Edge weights are written into the weight
attribute instead of the additional thickness
value.
When parsing bends the first (last) point of the bend is deleted, if it is inside the source (target) node, i.e., if its distance to the source (target) node is zero.
stroketype, stipple and node type are now "string" instead of "int".
The line color is written as "outline" rather than "line".
Edge weight has been moved out of the graphics definition.
GML files use one '\t' instead of two ' ' as indentChar now.
The GraphAttributes::edgeSubGraphs
feature allows to define up to 32 subgraphs by assigning a subset of these subgraphs to each edge. These subsets are internally stored in a bitmask. In former versions of OGDF, the GML representation of this attribute was simply subgraph <bitmask>
(in the definition scope of an edge). This has been changed to output one line containing subgraph <subgraph-id>
for each subgraph the edge is contained in. Note that this change implies that GML files exported by former versions of OGDF and using this feature are not recognized correctly. One can fix this manually, e.g., subgraph 5
now becomes subgraph 0 subgraph 2
.
x,y,z and size are now "double" instead of "float".
NodeType is now "int" instead of "string".
The (misspelled) keyword "avaliable-for" is changed to "subgraphs".
If custom node IDs are supplied, they are written into an extra field "nodeid" instead of the node's "id" attribute, to keep source and target references for edges consistent.
Support for the OGML
-format was dropped entirely. This includes several methods of GraphIO
and some classes that were used only by the OGML
reader/writer.
The list of deleted classes includes
LineBuffer
OgmlParser
UmlToGraphConverter
XmlParser
XmlScanner
This section applies if you happened to use the GmlParser
class directly (instead of GraphIO
).
The class has been moved and renamed to gml::Parser
. The constructor with filename argument is gone now. Use the constructor for input streams instead. Error messages are now output via the central logger; the ability to directly query the last error has been removed.
The filesystem functions previously found in ogdf/basic/filesystem.h
have been removed since there are maintained and portable alternatives like tinydir. C++17 also offers filesystem functions via #include <filesystem>
.
CombinatorialEmbedding::splitFace(adjEntry, node)
and CombinatorialEmbedding::splitFace(node, adjEntry)
are used to insert degree-0 nodes without having to call CombinatorialEmbedding::computeFaces()
again. In previous version both methods could be used with non-isolated nodes if the last adjacency entry belonged to the same face as the other adjacency entry. This is no longer supported and the node has to be isolated. To reflect this functional change the respective versions of splitFace
were renamed to addEdgeToIsolatedNode
. Use myEmbedding.splitFace(v->lastAdj(), adj)
instead of myEmbedding.splitFace(v, adj)
if v
isn't isolated (and you are sure that the adjacency entries lie on the same face!). Use myEmbedding.addEdgeToIsolatedNode(v, adj)
if v
is isolated.
The macros mentioned above have been removed. You can check for the SSE2 CPU feature directly using ogdf::System::cpuSupports(ogdf::CPUFeature::SSE2)
.
The template class ModuleOption
was removed. std::unique_ptr
should be used instead.
The static methods angle
and angleDegrees
of DPoint
and GenericPoint
are now non-static. That means, old calls of
now become
Some types / constants of PoolMemoryAllocator
were deleted or renamed.
Former | New |
---|---|
BlockChainPtr | - |
eBlockSize | BLOCK_SIZE |
eMinBytes | MIN_BYTES |
ePoolVectorLength | - |
eTableSize | TABLE_SIZE |
MemElemEx | - |
MemElemExPtr | - |
PoolVector | - |
NonPlanarCore<TCost>::mincut(edge e)
now returns an NonPlanarCore<TCost>::CutEdge
which is a struct with the two parameters edge e
, the edge itself and a bool dir
which indicates the direction of the edge, i.e. indicates if the edge goes from s to t or the other way round. Furthermore the NonPlanarCore
can now handle weighted edges and therefore is templated with the type of the weights.
Some very simple example layout algorithms (using the ModularMultilevelMixer
) were removed. This includes MMMExampleFast
, MMMExampleNice
, and MMMExampleNoTwist
. The respective code can still be found in doc/examples/layout/multilevelmixer.cpp
.
The enumerators from FMMMLayout
are now in a new class FMMMOptions
. Hence, for example, FMMMLayout::apInteger
became FMMMOptions::AllowedPositions::Integer
and FMMMLayout::gcNonUniformProbLowerMass
became FMMMLayout::GalaxyChoice::NonUniformProbLowerMass
.
The basic data structure DisjointSets
is highly customizable using template parameters. The enumerator names to do so have now changed.
Former | New |
---|---|
NL | LinkOptions::Naive |
LI | LinkOptions::Index |
LS | LinkOptions::Size |
LR | LinkOptions::Rank |
PC | CompressionOptions::PathCompression |
PS | CompressionOptions::PathSplitting |
PH | CompressionOptions::PathHalving |
R1 | CompressionOptions::Type1Reversal |
CO | CompressionOptions::Collapsing |
NF | CompressionOptions::Disabled |
NI | InterleavingOptions::Disabled |
Rem | InterleavingOptions::Rem |
TvL | InterleavingOptions::Tarjan |
IR0 | InterleavingOptions::Type0Reversal |
IPSPC | InterleavingOptions::SplittingCompression |
The following arrays (containing the string expansions of the settings) are removed:
linkOptionNames
compressionOptionNames
interleavingOptionNames
The function isForest()
in basic/simple_graph_alg.h
has been deprecated in favor of isArborescenceForest()
. isForest()
formerly returned true
even when the graph contained multi-edges or self-loops. This is no longer the case for isArborescenceForest()
.
Furthermore, isFreeForest()
has been deprecated in favor of isAcyclicUndirected()
.
The function connectedIsolatedComponents()
in basic/simple_graph_alg.h
has been deprecated in favor of connectedComponents()
. connectedIsolatedComponents(graph, isolatedNodes, component)
can be rewritten as connectedComponents(graph, component, &isolatedNodes)
.
The overloaded functions of makeParallelFreeUndirected()
in basic/simple_graph_alg.h
have been deprecated in favor of a single function that takes pointers as parameters (their default being nullptr
). makeParallelFreeUndirected(graph, parEdges, cardPositive, cardNegative)
can be rewritten as makeParallelFreeUndirected(graph, &parEdges, &cardPositive, &cardNegative)
.
The old CliqueFinder
was replaced by CliqueFinderSPQR
which is a subclass of CliquerFinderModule
. Previously, a call to CliqueFinder
using Graph G
looked as follows:
It can be replaced by the following code:
The breaking changes made to the CliqueFinder
include:
CliqueFinderModule
via call()
, not via the constructor.call()
takes List<List<node>*>
as a parameter instead of List<List<node>>
.m_density
is a double in [0..1] instead of an int in [0..100].m_postProcess
is a bool instead of an enum value.writeGraph()
was removed in favor of the static function cliqueGraphAttributes()
, which can be used to modify a GraphAttributes
object in order to pass it to GraphIO:write
.Moreover, CliqueReplacer::replaceByStar()
takes List<List<node>*>
as a parameter instead of List<List<node>>
.
The function template template<class E> doDestruction(const E *)
has been removed in favor of !
std::is_trivially_destructible<E>::value
. Hence, lists of elements with trivial destructors are now deallocated in constant time automatically, without the necessity of writing a template specialization.
The methods setDirected
, setStrokeType
, and setFillPattern
have been removed from GraphAttributes
. Use the respective getters (directed
, strokeType
, and fillPattern
) instead to obtain a non-const reference to the underlying value.
The method initAttributes
was renamed to addAttributes
as it does not disable previously enabled attributes.
ClusterGraphAttributes
was adapted to work similarly to GraphAttributes
. Attributes can be enabled using the ClusterGraphAttributes
-flags clusterGraphics
, clusterStyle
, clusterLabel
and clusterTemplate
.
The methods setStrokeType
, and setFillPattern
have been removed from GraphClusterAttributes
. Use the respective getters (strokeType
and fillPattern
) instead to obtain a non-const reference to the underlying value.
GraphCopyAttributes
was removed, it must be replaced by GraphAttributes
. Calls to copyAttr.transform()
must be replaced by copyAttr.transferToOriginal(origAttr)
. Calls to copyAttr.getWidth()
must be replaced by graphCopy.isDummy(v) ? 0.0 : copyAttr.width(v)
(analogously for getHeight()
).
Some set classes were removed / renamed.
Former | New |
---|---|
FaceSetSimple | FaceSet<false> |
FaceSetPure | FaceSet<false> |
FaceSet | FaceSet<> (defaults to FaceSet<true> ) |
NodeSetSimple | NodeSet<false> |
NodeSetPure | NodeSet<false> |
NodeSet | NodeSet<> (defaults to NodeSet<true> ) |
Math
is no longer a class with static methods but a namespace.
The classes DPoint
and DVector
are now merged. Note that DVector::length()
is now DPoint::norm()
. Furthermore, DVector::operator^
is now DPoint::determinant
.
The class DScaler
and the methods ogdf::DRound
and DPolyline::convertToInt
were deleted.
The classes DLine
and DSegment
were rewritten in such a way that DLine
represents infinite lines and DSegment
represents finite line segments with start and end points. The methods to access the sides of a DRect
now return DSegment
s instead of DLine
s, and they were renamed appropriately.
Former | New |
---|---|
DLine::intersectionOfLines() | DLine::intersection() |
DSegment::intersectionOfLines() | DLine::intersection() |
DRect::bottomLine() | DRect::bottom() |
DRect::leftLine() | DRect::left() |
DRect::rightLine() | DRect::right() |
DRect::topLine() | DRect::top() |
The methods intersection()
, verIntersection()
and horIntersection()
of DLine
and DSegment
now return an IntersectionType
instead of a bool
.
The class HyperGraph
was removed. Use Hypergraph
instead.
The classes EFreeList
, EList
and EStack
were removed. Use List
and Stack
instead.