Node Shapes

The geometries of all node shapes are affected by the node attributes fixedsize, fontname, fontsize, height, label, style and width.

Polygon-based Nodes

The possible polygon-based shapes are displayed below.
box polygon ellipse circle
point egg triangle plaintext
diamond trapezium parallelogram house
pentagon hexagon septagon octagon
doublecircle doubleoctagon tripleoctagon invtriangle
invtrapezium invhouse Mdiamond Msquare
Mcircle rect rectangle
As the figures suggest, the shapes rect and rectangle are synonyms for box. Also, unlike the rest, we have shown the plaintext node without style=filled to indicate its normal use. If fill were turned on, the label text would appear in a filled rectangle.

The geometries of polygon-based shapes are also affected by the node attributes regular, peripheries and orientation. If shape="polygon", the attributes sides, skew and distortion are also used. If unset, they default to 4, 0.0 and 0.0, respectively. In addition, the 3 M* shapes support auxiliary labels using the toplabel and bottomlabel attributes. On the other hand, the point shape is special in that it is only affected by the peripheries, width and height attributes.

Record-based Nodes

These are specified by shape values of "record" and "Mrecord". The structure of a record-based node is determined by its label, which has the following schema:
rlabel=field ( '|' field )*
where field=fieldId or '{' rlabel '}'
and fieldId= ( '<' string '>') ( string }
Literal braces, vertical bars and angle brackets must be escaped. Spaces are interpreted as separators between tokens, so they must be escaped if you want spaces in the text.

The first string in fieldId gives a name to the field and can be combined with the node name to denote the end of an edge. The second string is used as the text for the field; it supports the usual escape sequences \n, \l and \r.

Visually, a record is a box, with fields represented by alternating rows of horizontal or vertical subboxes. The Mrecord shape is identical to a record shape, except that the outermost box has rounded corners. Flipping between horizontal and vertical layouts is done by nesting fields in braces "{...}". The top-level orientation in a record is horizontal. Thus, a record with label "A | B | C | D" will have 4 fields oriented left to right, while "{A | B | C | D}" will have them from top to bottom and "A | { B | C } | D" will have "B" over "C", with "A" to the left and "D" to the right of "B" and "C".

As an example of a record node, the dot input

digraph structs { node [shape=record]; struct1 [label="<f0> left|<f1> mid\ dle|<f2> right"]; struct2 [label="<f0> one|<f1> two"]; struct3 [label="hello\nworld |{ b |{c|<here> d|e}| f}| g | h"]; struct1:f1 -> struct2:f0; struct1:f2 -> struct3:here; } yields the figure

If we change node struct1 to have shape Mrecord, it then looks like:

Styles for Nodes

The style attribute can be used to modify the appearance of a node. At present, there are 4 style values recognized: filled, invisible, diagonals and rounded.
filled
This value indicates that the node's interior should be filled. The color used is the node's fillcolor or, if that's not defined, its color. For unfilled nodes, the interior of the node is transparent to whatever color is the current graph or cluster background color. Note that point shapes are always filled.

Thus, the code

digraph G { rankdir=LR node [shape=box, color=blue] node1 [style=filled] node2 [style=filled, fillcolor=red] node0 -> node1 -> node2 } yields the figure
invisible
Setting this style causes the node not to be displayed at all. Note that the node is still used in laying out the graph.
diagonals
The diagonals style causes small chords to be drawn near the vertices of the node's polygon or, in case of circles and ellipses, two chords near the top and the bottom of the shape. The special node shapes Msquare, Mcircle, and Mdiamond are simply an ordinary square, circle and diamond with the diagonals style set.
rounded
The rounded style causes the polygonal corners to be smoothed. Note that this style also applies to record-based nodes. Indeed, the Mrecord shape is simply shorthand for setting this style. At present, the rounded and filled styles are mutually exclusive.

As an example of rounding, dot uses the graph

digraph R { rankdir=LR node [style=rounded] node1 [shape=box] node2 [shape=diamond] node3 [shape=record, label="{ a | b | c }"] node1 -> node2 -> node3 } to produce the figure

Additional styles may be available with a specific code generator.

HTML-Like Labels

NOTE:This feature is only available on versions of Graphviz that are newer than mid-November 2003. In particular, it is not part of release 1.10.

If the value of a label attribute (label for nodes, edges, clusters, and graphs, and the headlabel and taillabel attributes of an edge) is given as an HTML string, that is, delimited by <...> rather than "...", the label is interpreted as an HTML description. At their simplest, such labels can describe multiple lines of variously aligned text as provided by ordinary string labels. More generally, the label can specify a table similar to those provided by HTML, with different graphical attributes at each level.

NOTE: The features and syntax supported by these labels is inspired by HTML. However, there are many aspects that are relevant to Graphviz labels that are not in HTML and, conversely, HTML allows various constructs which are meaningless in Graphviz. We will generally refer to these labels as "HTML labels" rather than the cumbersome "HTML-like labels" but the reader is warned that these are not really HTML. The grammar below describes precisely what Graphviz will accept.

Although HTML labels are not strictly speaking a shape, they can be viewed as a generalization of the record shapes described above. In particular, if a node has set its shape attribute to plaintext, the HTML label will be the node's shape. On the other hand, if the node has any other shape (except point), the HTML label will be embedded within the node the same way an ordinary label would be.

The following is an abstract grammar for HTML labels. Terminals are shown in bold font and nonterminals in italics. Vertical bars | separate alternatives.
label : text
| table
text : lines
| lines string
| string
lines : string <BR/>
| lines string <BR/>
table : <TABLE> rows </TABLE>
rows : row
| rows row
row : <TR> cells </TR>
cells : cell
| cells cell
cell : <TD> label </TD>

Above, a string is any collection of printable characters, including all spaces. Note that outside of the body of a <TD> element, whitespace characters are ignored; within a <TD> element, spaces are preserved but all other white space characters are discarded. HTML comments are allowed within an HTML string. They can occur anywhere provided that if they contain part of an HTML element, they just contain the entire element.

Each of the HTML elements has a set of optional attributes. 

<TABLE
  ALIGN="CENTER|LEFT|RIGHT"
  BGCOLOR="color"
  BORDER="value"
  CELLBORDER="value"
  CELLPADDING="value"
  CELLSPACING="value"
  FIXEDSIZE="FALSE|TRUE"
  HEIGHT="value"
  HREF="value"
  PORT="portName"
  VALIGN="MIDDLE|BOTTOM|TOP"
  WIDTH="value"
>


<TD
  ALIGN="CENTER|LEFT|RIGHT"
  BGCOLOR="color"
  BORDER="value"
  CELLPADDING="value"
  CELLSPACING="value"
  COLSPAN="value"
  FIXEDSIZE="FALSE|TRUE"
  HEIGHT="value"
  HREF="value"
  PORT="portName"
  ROWSPAN="value"
  VALIGN="MIDDLE|BOTTOM|TOP"
  WIDTH="value"
>


<BR
  ALIGN="CENTER|LEFT|RIGHT"
>

ALIGN

specifies horizontal placement. When an object is allocated more space than required, this value determines where the extra space is placed left and right of the object.

BGCOLOR="color"
sets the color of the background. This color can be overridden by a BGCOLOR attribute in descendents.
BORDER="value"
specifies the width of the border around the object in points. A value of zero indicates no border. The default is 1. If set in a table, and CELLBORDER is not set, this value is also used for all cells in the table. It can be overridden by a BORDER tag in a cell.
CELLBORDER="value"
specifies the width of the border for all cells in a table. It can be overridden by a BORDER tag in a cell.
CELLPADDING="value"
specifies the space, in points, between a cell's border and its content. The default is 2.
CELLSPACING="value"
specifies the space, in points, between cells in a table and between a cell and the table's border. The default is 2.
COLSPAN="value"
specifies the number of columns spanned by the cell. The default is 1.
FIXEDSIZE
specifies whether the values given by the WIDTH and HEIGHT attributes are enforced.

HEIGHT="value"
specifies the mininum height, in points, of the object. The height includes the contents, any spacing and the border. Unless FIXEDSIZE is true, the height will be expanded to allow the contents to fit.
HREF="value"
attaches a URL to the object.
PORT="value"
attaches a port name to the object. This can be used to modify the head or tail of an edge, so that the end attaches directly to the object.
ROWSPAN="value"
specifies the number of rows spanned by the cell. The default is 1.

VALIGN

specifies vertical placement. When an object is allocated more space than required, this value determines where the extra space is placed above and below the object.

WIDTH="value"
specifies the mininum width, in points, of the object. The width includes the contents, any spacing and the border. Unless FIXEDSIZE is true, the width will be expanded to allow the contents to fit.

There is some inheritance among the attributes. If a table specifies a CELLPADDING, CELLBORDER or BORDER value, this value is used by the table's cells unless overridden. If a cell or table specifies a BGCOLOR, this will be the background color for all of its descendents. Of course, if a background or fill color is specified for the graph object owning the label, this will be the original background for the label. Finally, the pencolor or color of the graph object will be used as the border color, and the object's fontname, fontcolor and fontsize attributes are used for drawing text.

Because of certain limitations in handling tables in a device-independent manner, when BORDER is 1 and both table and cell borders are on and CELLSPACING is less than 2, anomalies can arise in the output, such as gaps between sides of borders which should be abutting or even collinear. The user can usual get around this by increasing the border size or the spacing, or turning off the table border.

As an example of HTML labels, the dot input

digraph structs { node [shape=plaintext] struct1 [label=< <TABLE BORDER="0" CELLBORDER="1" CELLSPACING="0"> <TR><TD>left</TD><TD PORT="f1">mid dle</TD><TD PORT="f2">right</TD></TR> </TABLE>>]; struct2 [label=< <TABLE BORDER="0" CELLBORDER="1" CELLSPACING="0"> <TR><TD PORT="f0">one</TD><TD>two</TD></TR> </TABLE>>]; struct3 [label=< <TABLE BORDER="0" CELLBORDER="1" CELLSPACING="0" CELLPADDING="4"> <TR> <TD ROWSPAN="3">hello<BR/>world</TD> <TD COLSPAN="3">b</TD> <TD ROWSPAN="3">g</TD> <TD ROWSPAN="3">h</TD> </TR> <TR> <TD>c</TD><TD PORT="here">d</TD><TD>e</TD> </TR> <TR> <TD COLSPAN="3">f</TD> </TR> </TABLE>>]; struct1:f1 -> struct2:f0; struct1:f2 -> struct3:here; } produces the HTML analogue of the record example above

As usual, an HTML specification is more verbose. On the other hand, HTML labels are much more general, as the following example shows:

The source for this graph can be found here.

User-defined Nodes

There is a third type of node shape which is specified by the user. Typically, these shapes rely on the details of a concrete graphics format. At present, shapes can be described using PostScript, via a file or add-on library, for use in PostScript output, or shapes can be specified by a bitmap-image file for use with SVG or bitmap (jpeg, gif, etc.) output. More information can be found on the page How to create custom shapes.