Graphviz's Four Layout Engines: Pick Wrong and Your Diagram Becomes Unreadable
Mastering Layout Engines in Graphviz: Dot vs. Neato vs. Twopi vs. Circo

Graphviz's rendering power comes from four specialized layout engines: dot for hierarchical directed graphs, neato for force-directed networks, twopi for radial hierarchies, and circo for circular topologies. Choosing the wrong one produces cluttered node overlaps and crossing lines. A comparative code example shows how switching the engine attribute transforms the same dataset, and a browser-based editor lets you preview each layout instantly.
Selecting the wrong layout engine can result in cluttered node overlaps, crossing lines, and unreadable architecture maps.
- meh2frdf
I’ve been working in this area recently, none of the layout engine are very good, they essentially apply the same algorithm to all the things when really a graph is often a collection of different patterns and what makes sense is often semantics driven. So yes built my own, the use case is representing cloud and enterprise architectures. Have a decent layout engine now, and created a declarative dsl for the agent to interact with, but the agent sucks, feels like we need to hand roll a lot of features/skills
- AceJohnny2
Tangentially, I must highlight the underrecognized gvpr [1], the Graphviz Pattern Scanning and Processing Language and part of any Graphviz installation. In short, it is an AWK-like program to process your .dot files.
Combined with the fact that the DOT syntax allows for gradual precision, in that you can add directives to an existing DOT document to add details (shape, positioning...), and you can do pre-processing of a graph before passing it to the final layout and rendering engine.
(In fact, you could probably implement all of the linked layout engines in gvpr)
I used this at some point to generate a Factorio processing graph, organizing each layer by its minimum production depth (ie ores at layer 1, plates at layer 2, etc...)
- aezart
I occasionally use graphviz and pretty much always wind up frustrated with it. The syntax is pedantic and the resulting output is pretty ugly for all but the simplest graphs.