Skip to content

Publication

We intend to publish rostree as a peer-reviewed software paper.

The aim is the same as the project's: a framework other people can build on, and a clear explanation of the algorithms behind dependency analysis in ROS 2 — why a ROS dependency graph has to be treated as a graph rather than a tree, and how to answer questions about it (why, who depends on this, what changed) in time linear in its size. A paper puts that explanation through review and gives researchers who use rostree something to cite.

This page records the plan in the open, including what is not ready yet.

Candidate venues

As of November 2026:

Venue Format Fit Status
Frontiers — Robotics Software Engineering Vol. 2 Journal article, Technology and Code type Scope names ROS, software architecture, testing and CI; linked to the RoSE workshop Open; manuscripts due 18 December 2026. Publishing fee applies
RoSE — International Workshop on Robotics Software Engineering Workshop paper; past editions had a short tool-paper track Software engineering for robotics — exactly this 2026 edition held with ICRA (Vienna, June); 2027 not yet announced
JOSS — Journal of Open Source Software Short paper, open review on GitHub Requires an "obvious research application"; rejects minor utilities Possible, but the research use would have to be argued
ROSCon Talk, not a paper The ROS community itself 2026 calls closed; 2027 to come

What the paper would describe

  • The problem, measured: printing a ROS dependency DAG as a tree grows exponentially with shared dependencies — 60,981 lines for one package of a 146-package workspace, against 253 when each package is expanded once.
  • The algorithms, as How the algorithms work explains them: expand-once tree building, breadth-first graph resolution with an explicit depth frontier, all-shortest-paths for why, reverse dependencies that filter the answer rather than the walk, three-colour cycle detection, layered DAG layout in the browser, and REP 149 condition evaluation without eval.
  • The design: a library first, with the CLI, TUI and HTML viewer as clients of the same API; an interactive graph that works with no network; filters that always say what they removed.

What has to happen first

A paper makes claims, and some of rostree's cannot be backed yet.

  • [x] Public repository and issue tracker
  • [ ] One licence. The LICENSE file is BSD 3-Clause, with copyright held by "The rosdep_viz authors" (the project's first name); pyproject.toml, and so PyPI, say MIT. Both are permissive, but a paper — and anyone who redistributes rostree — needs them to agree. The citation files leave the licence out until they do.
  • [x] Automated tests and CI, with coverage reported on every pull request
  • [x] Documentation site, including a public API covered by semantic versioning
  • [x] Tagged releases on PyPI, and a changelog
  • [x] Citation metadata: CITATION.cff and codemeta.json in the repository
  • [ ] Tests against a real ROS 2 installation. Every test runs on generated workspaces today. Running the suite in a ros:jazzy container is the top item on the roadmap, and a paper should not go out before it.
  • [ ] Evaluation on real, public workspaces — timings and output sizes on large open-source ROS 2 code bases, rather than the synthetic workspace the current measurements use.
  • [ ] An archived release with a DOI (for example through Zenodo), so the paper cites an exact, permanent version.

Use of AI tools

Much of rostree's code, tests and documentation from version 0.3.0 on was written with Claude Code, Anthropic's AI coding agent, working through pull requests on this repository. Those pull requests say so in their descriptions. The commits on the main branch do not reliably record it — most were squash-merged without the co-author line — so the pull requests, not the commit log, are the record.

The code review in docs/review.md that preceded 0.3.0 was also AI-generated, as the commit that added it says.

Any submission will disclose this in the form the venue asks for.

How to cite rostree

Until there is a paper, cite the software itself. GitHub's Cite this repository button (on the repository's front page) reads CITATION.cff and produces APA and BibTeX. Please include the version you used.

How you can help

The most useful contribution to the paper is evidence: run rostree on a workspace you work on and tell us what it got right or wrong — an issue describing your workspace layout is enough. Reports from real code bases are exactly what the evaluation needs, and the kind of thing no generated fixture can stand in for.