TALA autolayout algorithm released as open-source under MPL-2.0

TALA, the orthogonal autolayout algorithm built for architecture diagrams, has been released as open-source under the MPL-2.0 license. The source is now bundled with D2 starting at version v0.9.0 and can be invoked with the --layout=tala flag or tried in the online playground. This makes the algorithm’s implementation visible and modifiable for anyone using D2 to generate diagrams.

Adopting TALA requires understanding its layout objectives, such as symmetry, median distance, flow, and clustering, which influence the aesthetic score. The algorithm uses a seeded stochastic search (default three seeds) and picks the best-scoring layout, so deterministic results depend on fixing the seed set. Runtime grows nonlinearly with diagram size, making it slower than Dagre or ELK for large graphs, and users may need to benchmark performance via the provided link. Additionally, TALA is less effective at producing long, unidirectional DAG layouts compared to the existing Dagre and ELK options.

Because of its stochastic nature, adding a single node can lead to a markedly different diagram, whereas Dagre and ELK tend to preserve the existing structure. Users who prioritize stable incremental updates may find this unpredictability a drawback for evolving diagrams. For diagrams where node positions must be locked or partially specified, TALA supports custom coordinates and hybrid modes, but the core routing still relies on its stochastic search.

Opening the code under MPL-2.0 invites the community to improve the algorithm’s scalability, reduce randomness, or extend its aesthetic criteria. Contributions could address the noted tradeoffs, potentially making TALA more suitable for larger or more directional graphs. Until such improvements appear, engineers should weigh the layout quality benefits against the performance and stability costs when choosing TALA for their diagramming workflows.