feat(graphviz): node- and edge-specific custom attributes

[peter-gy]

Description of changes

Follow-up to #8510.

Adds support to optionally specifying custom Graphviz node attributes and edge attributes based on arbitrary logic described by user-defined callbacks: node_attr_callback and edge_attr_callback.

With this, more complex visualization needs can be addressed, such as:

• I want to use a custom shape and fill color for all the Field nodes
• I want to have custom edge style and color between Field and Comparison nodes

Consider the example below.

import ibis
import ibis.expr.operations as ops

left = ibis.table(dict(a="int64", b="string"), name="left")
right = ibis.table(dict(b="string", c="int64", d="string"), name="right")
expr = (
    left.inner_join(right, "b")
    .select(left.a, b=right.c, c=right.d)
    .mutate(
        arrays=ibis.array([1, 2, 3]),
    )
)

def get_node_attributes(node):
    if isinstance(node, ops.Literal):
        return {"shape": "hexagon", "style": "filled", "fillcolor": "yellow"}

def get_edge_attributes(u, v):
    if isinstance(u, ops.Field) and isinstance(v, ops.Comparison):
        return {"style": "dashed", "color": "red"}

    if isinstance(v, ops.Field):
        return {"arrowhead": "box", "color": "blue"}


expr.visualize(
    label_edges=True,
    node_attr={"color": "green", "fontname": "Roboto Mono"},
    node_attr_callback=get_node_attributes,
    edge_attr={"fontsize": "12", "fontname": "Comic Sans MS"},
    edge_attr_callback=get_edge_attributes,
)

[peter-gy]

An API design consideration we have here is whether we want to merge node_attr and node_attr_callback as well as edge_attr and edge_attr_callback to be of type dict[str,str] | NodeAttributeCallback | None and dict[str,str] | EdgeAttributeCallback | None respectively so that we don’t impose additional cognitive load on users with parameter variations.

What do you think?

[cpcloud]

Pretty fancy graph there!

Keeping the separation between "default attributes for every node" and "a function that can be used to override specific nodes" seems like a good separation to me.

While I don't love the _callback naming, I'm not sure there's something that significantly better.

[peter-gy]

Pretty fancy graph there!

Keeping the separation between "default attributes for every node" and "a function that can be used to override specific nodes" seems like a good separation to me.

While I don't love the _callback naming, I'm not sure there's something that significantly better.

I agree with the need for the separation.

What about _getter as a suffix?

[cpcloud]

getter SGTM!

[cpcloud]

Niiiiice!