cgraph - Man Page

abstract graph library


#include <graphviz/cgraph.h>




Agmemdisc_t AgMemDisc;
Agiddisc_t  AgIdDisc;
Agiodisc_t  AgIoDisc;
Agdisc_t    AgDefaultDisc;


Agraph_t	*agopen(char *name, Agdesc_t kind, Agdisc_t *disc);
int		agclose(Agraph_t *g);
Agraph_t	*agread(void *channel, Agdisc_t *);
Agraph_t	*agmemread(char *);
void		agreadline(int line_no);
void		agsetfile(char *file_name);
Agraph_t	*agconcat(Agraph_t *g, void *channel, Agdisc_t *disc)
int		agwrite(Agraph_t *g, void *channel);
int		agnnodes(Agraph_t *g),agnedges(Agraph_t *g), agnsubg(Agraph_t * g);
int		agisdirected(Agraph_t * g),agisundirected(Agraph_t * g),agisstrict(Agraph_t * g), agissimple(Agraph_t * g);


Agraph_t	*agsubg(Agraph_t *g, char *name, int createflag);
Agraph_t	*agidsubg(Agraph_t * g, unsigned long id, int cflag);
Agraph_t	*agfstsubg(Agraph_t *g), agnxtsubg(Agraph_t *);
Agraph_t	*agparent(Agraph_t *g);
int		agdelsubg(Agraph_t * g, Agraph_t * sub);    /* same as agclose() */


Agnode_t	*agnode(Agraph_t *g, char *name, int createflag);
Agnode_t	*agidnode(Agraph_t *g, ulong id, int createflag);
Agnode_t	*agsubnode(Agraph_t *g, Agnode_t *n, int createflag);
Agnode_t	*agfstnode(Agraph_t *g);
Agnode_t	*agnxtnode(Agraph_t *g, Agnode_t *n);
Agnode_t	*agprvnode(Agraph_t *g, Agnode_t *n);
Agnode_t	*aglstnode(Agraph_t *g);
int		agdelnode(Agraph_t *g, Agnode_t *n);
int		agdegree(Agraph_t *g, Agnode_t *n, int use_inedges, int use_outedges);
int		agcountuniqedges(Agraph_t * g, Agnode_t * n, int in, int out);


Agedge_t	*agedge(Agraph_t* g, Agnode_t *t, Agnode_t *h, char *name, int createflag);
Agedge_t	*agidedge(Agraph_t * g, Agnode_t * t, Agnode_t * h, unsigned long id, int createflag);
Agedge_t	*agsubedge(Agraph_t *g, Agedge_t *e, int createflag);
Agnode_t	*aghead(Agedge_t *e), *agtail(Agedge_t *e);
Agedge_t	*agfstedge(Agraph_t* g, Agnode_t *n);
Agedge_t	*agnxtedge(Agraph_t* g, Agedge_t *e, Agnode_t *n);
Agedge_t	*agfstin(Agraph_t* g, Agnode_t *n);
Agedge_t	*agnxtin(Agraph_t* g, Agedge_t *e);
Agedge_t	*agfstout(Agraph_t* g, Agnode_t *n);
Agedge_t	*agnxtout(Agraph_t* g, Agedge_t *e);
int		agdeledge(Agraph_t *g, Agedge_t *e);
Agedge_t	*agopp(Agedge_t *e);
int		ageqedge(Agedge_t *e0, Agedge_t *e1);

String Attributes

Agsym_t	*agattr(Agraph_t *g, int kind, char *name, const char *value);
Agsym_t	*agattrsym(void *obj, char *name);
Agsym_t	*agnxtattr(Agraph_t *g, int kind, Agsym_t *attr);
char		*agget(void *obj, char *name);
char		*agxget(void *obj, Agsym_t *sym);
int		agset(void *obj, char *name, char *value);
int		agxset(void *obj, Agsym_t *sym, char *value);
int		agsafeset(void *obj, char *name, char *value, char *def);
int		agcopyattr(void *, void *);


void		*agbindrec(void *obj, char *name, unsigned int size, move_to_front);
Agrec_t		*aggetrec(void *obj, char *name, int move_to_front);
int		agdelrec(Agraph_t *g, void *obj, char *name);
void		aginit(Agraph_t * g, int kind, char *rec_name, int rec_size, int move_to_front);
void		agclean(Agraph_t * g, int kind, char *rec_name);


int			*agpopdisc(Agraph_t *g);
void		agpushdisc(Agraph_t *g, Agcbdisc_t *disc);
int			agcallbacks(Agraph_t * g, int flag);


void		*agalloc(Agraph_t *g, size_t request);
void		*agrealloc(Agraph_t *g, void *ptr, size_t oldsize, size_t newsize);
void		agfree(Agraph_t *g, void *ptr);


char		*agstrdup(Agraph_t *, char *);
char		*agstrdup_html(Agraph_t *, char *);
int		aghtmlstr(char *);
char		*agstrbind(Agraph_t * g, char *);
int		strfree(Agraph_t *, char *);
char		*agcanonStr(char *);
char		*agstrcanon(char *, char *);
char		*agcanon(char *, int);

Generic Objects

Agraph_t	*agraphof(void*);
Agraph_t	*agroot(void*);
int		agcontains(Agraph_t*, void*);
char		*agnameof(void*);
void		agdelete(Agraph_t *g, void *obj);
int		agobjkind(void *obj);
Agrec_t		*AGDATA(void *obj);
ulong		AGID(void *obj);
int		AGTYPE(void *obj);

Error Reporting

typedef enum { AGWARN, AGERR, AGMAX, AGPREV } agerrlevel_t;
typedef int (*agusererrf) (char*);
agerrlevel_t	agerrno;
agerrlevel_t	agseterr(agerrlevel_t);
char		*aglasterr(void);
int		agerr(agerrlevel_t level, char *fmt, ...);
void		agerrorf(char *fmt, ...);
void		agwarningf(char *fmt, ...);
int		agerrors(void);
agusererrf	agseterrf(agusererrf);


Libcgraph supports graph programming by maintaining graphs in memory and reading and writing graph files. Graphs are composed of nodes, edges, and nested subgraphs. These graph objects may be attributed with string name-value pairs and programmer-defined records (see Attributes).

All of Libcgraph's global symbols have the prefix ag (case varying). In the following, if a function has a parameter int createflag and the object does not exist, the function will create the specified object if createflag is non-zero; otherwise, it will return NULL.

Graph and Subgraphs

A “main” or “root” graph defines a namespace for a collection of graph objects (subgraphs, nodes, edges) and their attributes. Objects may be named by unique strings or by integer IDs.

agopen creates a new graph with the given name and kind. (Graph kinds are Agdirected, Agundirected, Agstrictdirected, and Agstrictundirected. A strict graph cannot have multi-edges or self-arcs.) The final argument points to a discpline structure which can be used to tailor I/O, memory allocation, and ID allocation. Typically, a NULL value will be used to indicate the default discipline AgDefaultDisc. agclose deletes a graph, freeing its associated storage. agread, agwrite, and agconcat perform file I/O  using the graph file language described below. agread constructs a new graph while agconcat merges the file contents with a pre-existing graph.  Though I/O methods may be overridden, the default is that the channel argument is a stdio FILE pointer.  agmemread attempts to read a graph from the input string. agsetfile and agreadline are helper functions that simply set the current file name and input line number for subsequent error reporting.

The functions agisdirected, agisundirected, agisstrict, and agissimple can be used to query if a graph is directed, undirected, strict (at most one edge with a given tail and head), or simple (strict with no loops), respectively,

agsubg finds or creates a subgraph by name. agidsubg allows a programmer to specify the subgraph by a unique integer ID. A new subgraph is initially empty and is of the same kind as its parent.  Nested subgraph trees may be created.  A subgraph's name is only interpreted relative to its parent. A program can scan subgraphs under a given graph using agfstsubg and agnxtsubg.  A subgraph is deleted with agdelsubg (or agclose). The agparent function returns the immediate parent graph of a subgraph, or itself if the graph is already a root graph.

By default, nodes are stored in ordered sets for efficient random access to insert, find, and delete nodes. The edges of a node are also stored in ordered sets. The sets are maintained internally as splay tree dictionaries using Phong Vo's cdt library.

agnnodes, agnedges, and agnsubg return the sizes of node, edge and subgraph sets of a graph.   The function agdegree returns the size of the edge set of a nodes, and takes flags to select in-edges, out-edges, or both. The function agcountuniqedges returns the size of the edge set of a nodes, and takes flags to select in-edges, out-edges, or both. Unlike agdegree, each loop is only counted once.


A node is created by giving a unique string name or programmer defined integer ID, and is represented by a unique internal object. (Node equality can checked by pointer comparison.)

agnode searches in a graph or subgraph for a node with the given name, and returns it if found. agidnode allows a programmer to specify the node by a unique integer ID. agsubnode performs a similar operation on an existing node and a subgraph.

agfstnode and agnxtnode scan node lists. agprvnode and aglstnode are symmetric but scan backward. The default sequence is order of creation (object timestamp.) agdelnode removes a node from a graph or subgraph.


An abstract edge has two endpoint nodes called tail and head where all outedges of the same node have it as the tail value and similarly all inedges have it as the head. In an undirected graph, head and tail are interchangeable. If a graph has multi-edges between the same pair of nodes, the edge's string name behaves as a secondary key.

agedge searches in a graph or subgraph for an edge between the given endpoints (with an optional multi-edge selector name) and returns it if found or created. Note that, in undirected graphs, a search tries both orderings of  the tail and head nodes. If the name  is NULL, then an anonymous internal value is generated. agidedge allows a programmer to create an edge by giving its unique integer ID. agsubedge performs a similar operation on an existing edge and a subgraph. agfstin, agnxtin, agfstout, and  agnxtout visit directed in- and out- edge lists, and ordinarily apply only in directed graphs. agfstedge and agnxtedge visit all edges incident to a node.  agtail and aghead get the endpoint of an edge. agdeledge removes an edge from a graph or subgraph.

Note that an abstract edge has two distinct concrete representations: as an in-edge and as an out-edge. In particular, the pointer as an out-edge is different from the pointer as an in-edge. The function ageqedge  canonicalizes the pointers before doing a comparison and so can be used to test edge equality. The sense of an edge can be flipped using agopp.

Internal Attributes

Programmer-defined values may be dynamically attached to graphs, subgraphs, nodes, and edges. Such values are either character string data (for I/O) or uninterpreted binary records (for implementing algorithms efficiently).

String Attributes

String attributes are handled automatically in reading and writing graph files.  A string attribute is identified by name and by an internal symbol table entry (Agsym_t) created by Libcgraph. Attributes of nodes, edges, and graphs (with their subgraphs) have separate namespaces.  The contents of an Agsym_t have a char* name for the attribute's name, a char* defval field for the attribute's default value, and an int id field containing the index of the attribute's specific value for an object in the object's array of attribute values.

agattr creates or looks up attributes. kind may be AGRAPH, AGNODE, or AGEDGE. If value is (char*)0), the request is to search for an existing attribute of the given kind and name. Otherwise, if the attribute already exists, its default for creating new objects is set to the given value; if it does not exist, a new attribute is created with the given default, and the default is applied to all pre-existing objects of the given kind. If g is NULL, the default is set for all graphs created subsequently. agattrsym is a helper function that looks up an attribute for a graph object given as an argument. agnxtattr permits traversing the list of attributes of a given type.  If NULL is passed as an argument it gets the first attribute; otherwise it returns the next one in succession or returns NULL at the end of the list. agget and agset allow fetching and updating a string attribute for an object taking the attribute name as an argument.  agxget and agxset do this but with an attribute symbol table entry as an argument (to avoid the cost of the string lookup).  Note that agset will fail unless the attribute is first defined using agattr.  agsafeset is a convenience function that ensures the given attribute is declared before setting it locally on an object.

It is sometimes convenient to copy all of the attributes from one object to another. This can be done using agcopyattr. This fails and returns non-zero of argument objects are different kinds, or if all of the attributes of the source object have not been declared for the target object.


Libcgraph performs its own storage management of strings as  reference-counted strings. The caller does not need to dynamically allocate storage.

agstrdup returns a pointer to a reference-counted copy of the argument string, creating one if necessary. agstrbind returns a pointer to a reference-counted string if it exists, or NULL if not. All uses of cgraph strings need to be freed using agstrfree in order to correctly maintain the reference count.

The cgraph parser handles HTML-like strings. These should be  indistinguishable from other strings for most purposes. To create an HTML-like string, use agstrdup_html. The aghtmlstr function can be used to query if a string is an ordinary string or an HTML-like string.

agcanonStr returns a pointer to a version of the input string canonicalized for output for later re-parsing. This includes quoting special characters and keywords. It uses its own internal buffer, so the value will be lost on the next call to agcanonStr. agstrcanon is an unsafe version of agcanonStr, in which the application passes in a buffer as the second argument. Note that the buffer may not be used; if the input string is in canonical form, the function will just return a pointer to it. For both of the functions, the input string must have been created using agstrdup or agstrdup_html. Finally, agcanonStr is identical with agcanonStr except it can be used with any character string. The second argument indicates whether or not the string should be canonicalized as an HTML-like string.


Uninterpreted records may be attached to graphs, subgraphs, nodes, and edges for efficient operations on values such as marks, weights, counts, and pointers needed by algorithms.  Application programmers define the fields of these records, but they must be declared with a common header as shown below.

typedef struct {
    Agrec_t        header;
    /* programmer-defined fields follow */
} user_data_t;

Records are created and managed by Libcgraph. A programmer must explicitly attach them to the objects in a graph, either to individual objects one at a time via agbindrec, or to all the objects of the same class in a graph via aginit. (Note that for graphs, aginit is applied recursively to the graph and its subgraphs if rec_size is negative (of the actual rec_size.)) The name argument of a record distinguishes various types of records, and is programmer defined (Libcgraph reserves the prefix _ag). If size is 0, the call to agbindrec is simply a lookup. The function aggetrec can also be used for lookup. agdelrec deletes a named record from one object. agclean does the same for all objects of the same class in an entire graph.

Internally, records are maintained in circular linked lists attached to graph objects. To allow referencing application-dependent data without function calls or search, Libcgraph allows setting and locking the list pointer of a graph, node, or edge on a particular record. This pointer can be obtained with the macro AGDATA(obj). A cast, generally within a macro or inline function, is usually applied to convert the list pointer to an appropriate programmer-defined type.

To control the setting of this pointer, the move_to_front flag may be TRUE or FALSE. If move_to_front is TRUE, the record will be locked at the head of the list, so it can be accessed directly by AGDATA(obj). The lock can be subsequently released or reset by a call to aggetrec.


(This section is not intended for casual users.) Programmer-defined disciplines customize certain resources- ID namespace, memory, and I/O - needed by Libcgraph. A discipline struct (or NULL) is passed at graph creation time.

struct Agdisc_s {            /* user's discipline */
    Agmemdisc_t            *mem;
    Agiddisc_t            *id;
    Agiodisc_t            *io;
} ;

A default discipline is supplied when NULL is given for any of these fields.

Id Discipline

An ID allocator discipline allows a client to control assignment of IDs (uninterpreted integer values) to objects, and possibly how they are mapped to and from strings.

struct Agiddisc_s {             /* object ID allocator */
    void *(*open) (Agraph_t * g, Agdisc_t*);       /* associated with a graph */
    long (*map) (void *state, int objtype, char *str, unsigned long *id, int createflag);
    long (*alloc) (void *state, int objtype, unsigned long id);
    void (*free) (void *state, int objtype, unsigned long id);
    char *(*print) (void *state, int objtype, unsigned long id);
    void (*close) (void *state);

open permits the ID discipline to initialize any data structures that it maintains per individual graph. Its return value is then passed as the first argument (void *state) to all subsequent ID manager calls.

alloc informs the ID manager that Libcgraph is attempting to create an object with a specific ID that was given by a client. The ID manager should return TRUE (nonzero) if the ID can be allocated, or FALSE (which aborts the operation).

free is called to inform the ID manager that the object labeled with the given ID is about to go out of existence.

map is called to create or look-up IDs by string name (if supported by the ID manager).  Returning TRUE (nonzero) in all cases means that the request succeeded (with a valid ID stored through result.  There are four cases:

name != NULL and createflag == 1: This requests mapping a string (e.g. a name in a graph file) into a new ID. If the ID manager can comply, then it stores the result and returns TRUE. It is then also responsible for being able to print the ID again as a string.  Otherwise the ID manager may return FALSE but it must implement the following (at least for graph file reading and writing to work):

name == NULL and createflag == 1: The ID manager creates a unique new ID of its own choosing.  Although it may return FALSE if it does not support anonymous objects, but this is strongly discouraged (to support "local names" in graph files.)

name != NULL and createflag == 0: This is a namespace probe.  If the name was previously mapped into an allocated ID by the ID manager, then the manager must return this ID. Otherwise, the ID manager may either return FALSE, or may store any unallocated ID into result. (This is convenient, for example, if names are known to be digit strings that are directly converted into integer values.)

name == NULL and createflag == 0: forbidden.

print is allowed to return a pointer to a static buffer; a caller must copy its value if needed past subsequent calls. NULL should be returned by ID managers that do not map names.

The map and alloc calls do not pass a pointer to the newly allocated object.  If a client needs to install object pointers in a handle table, it can obtain them via  new object callbacks.

IO Discipline

The I/O discipline provides an abstraction for the reading and writing of graphs.

struct Agiodisc_s {
    int        (*fread)(void *chan, char *buf, int bufsize);
    int        (*putstr)(void *chan, char *str);
    int        (*flush)(void *chan);    /* sync */
} ;

Normally, the FILE structure and its related functions are used for I/O. At times, though, an application may need to use a totally different type of character source. The associated state or stream information is provided by the chan argument to agread or agwrite. The discipline function fread and putstr provide the corresponding functions for read and writing.

Memory Discipline

Memory management in Libcgraph is handled on a per graph basis using the memory discipline.

struct Agmemdisc_s {    /* memory allocator */
    void    *(*open)(Agdisc_t*);        /* independent of other resources */
    void    *(*alloc)(void *state, size_t req);
    void    *(*resize)(void *state, void *ptr, size_t old, size_t req);
    void    (*free)(void *state, void *ptr);
    void    (*close)(void *state);
} ;

The open function is used to initialize the memory subsystem, returning state information that is passed to the calls to   alloc, resize, and free. The semantics of these should be comparable to the standard C library functions malloc, realloc, and free, except that new space created by agalloc  and agrealloc should be zeroed out. The close function is used to terminate the memory subsystem, freeing any additional open resources. For actual allocation, the library uses the functions agalloc, agrealloc, and agfree, which provide simple wrappers for the underlying discipline functions alloc, resize, and free.

When Libcgraph is compiled with Vmalloc (which is not the default), each graph has its own heap. Programmers may allocate application-dependent data within the same heap as the rest of the graph.  The advantage is that a graph can be deleted by atomically freeing its entire heap without scanning each individual node and edge.


An Agcbdisc_t defines callbacks to be invoked by Libcgraph when initializing, modifying, or finalizing graph objects. Disciplines are kept on a stack.  Libcgraph automatically calls the methods on the stack, top-down.  Callbacks are installed with agpushdisc, uninstalled with agpopdisc, and  can be held pending or released via agcallbacks.

Generic Objects

agroot takes any graph object (graph, subgraph, node, edge) and returns the root graph in which it lives. agraphof does the same, except it  is the identity function on graphs and subgraphs. Note that there is no  function to return the least subgraph containing an object, in part because  this is not well-defined as nodes and edges may be in incomparable subgraphs.

agcontains(g,obj) returns non-zero if obj is a member  of (sub)graph g. agdelete(g,obj) is equivalent  to agclose, agdelnode, and agdeledge for obj being a  graph, node or edge, respectively. It returns -1 if obj does not  belong to g.

AGDATA, AGID, and AGTYPE are macros returning the specified fields of the argument object. The first is described in the Records section above. The second returns the unique integer ID associated with the object. The last returns AGRAPH, AGNODE, and AGEDGE depending on the type of the object.

agnameof returns a string descriptor for the object. It returns the name of the node or graph, and the key of an edge.  agobjkind is a synonym for AGTYPE.

Error Reporting

The library provides a variety of mechanisms to control the reporting of errors and warnings. At present, there are basically two types of messages: warnings and errors. A message is only written if its type has higher priority than a programmer-controlled minimum, which is AGWARN by default. The programmer can set this value using agseterr, which returns the previous value. Calling agseterr(AGMAX) turns off the writing of messages.

The function agerr if the main entry point for reporting an anomaly. The first argument indicates the type of message. Usually, the first argument in AGWARN or AGERR to indicate warnings and errors, respectively. Sometimes additional context information is only available in functions calling the function where the error is actually caught. In this case, the calling function can indicate that it is continuing the current error by using AGPREV as the first argument. The remaining arguments to agerr are the same as the arguments to printf.

The functions agwarningf and agerrorf are shorthand for agerr(AGERR,...) and agerr(AGWARN,...), respectively.

Some applications desire to directly control the writing of messages. Such an application can use the function agseterrf to register the function that the library should call to actually write the message. The previous error function is returned. By default, the message is written to stderr.

Errors not written are stored in a log file. The last recorded error can be retreived by calling aglasterr.

The function agerrors returns non-zero if errors have been reported.

Example Program

#include <stdio.h>
#include <cgraph.h>

typedef struct {Agrec_t hdr; int x,y,z;} mydata;

void main(int argc, char **argv)
    Agraph_t    *g, *h;
    Agnode_t    *v;
    Agedge_t    *e;
    Agsym_t     *attr;
    Dict_t      *d;
    int         cnt;
    mydata      *p;

    if (g = agread(stdin,NIL(Agdisc_t*))) {
        cnt = 0; attr = 0;
        while (attr = agnxtattr(g, AGNODE, attr)) cnt++;
        printf("The graph %s has %d attributes\n",agnameof(g),cnt);

        /* make the graph have a node color attribute, default is blue */
        attr = agattr(g,AGNODE,"color","blue");

        /* create a new graph of the same kind as g */
        h = agopen("tmp",g->desc, NULL);

        /* this is a way of counting all the edges of the graph */
        cnt = 0;
        for (v = agfstnode(g); v; v = agnxtnode(g,v))
            for (e = agfstout(g,v); e; e = agnxtout(g,e))

        /* attach records to edges */
        for (v = agfstnode(g); v; v = agnxtnode(g,v))
            for (e = agfstout(g,v); e; e = agnxtout(g,e)) {
                p = (mydata*) agbindrec(e,"mydata",sizeof(mydata),TRUE);
                p->x = 27;  /* meaningless data access example */
                ((mydata*)(AGDATA(e)))->y = 999; /* another example */

Example Graph Files

digraph G {
    a -> b;
    c [shape=box];
    a -> c [weight=29,label="some text"];
    subgraph anything {
        /* the following affects only x,y,z */
        node [shape=circle];
        a; x; y -> z; y -> z;  /* multiple edges */

strict graph H {
    n0 -- n1 -- n2 -- n0;  /* a cycle */
    n0 -- {a b c d};       /* a star */
    n0 -- n3;
    n0 -- n3 [weight=1];   /* same edge because graph is strict */

See Also



It is difficult to change endpoints of edges, delete string attributes or modify edge keys.  The work-around is to create a new object and copy the contents of an old one (but new object obviously has a different ID, internal address, and object creation timestamp).

The API lacks convenient functions to substitute programmer-defined ordering of nodes and edges but in principle this can be supported.

The library is not thread safe.


Stephen North,, AT&T Research.

Referenced By

gvpr(1), gvpr(3).

28 FEBRUARY 2013