Skip to content

CONTRIBUTORS.md

Latest commit

 

History

History
33 lines (25 loc) · 3.88 KB

CONTRIBUTORS.md

File metadata and controls

33 lines (25 loc) · 3.88 KB
Raw

Contributors

Open Ontologies exists thanks to everyone who has built, used, broken, and improved it.

Maintainer

Contributors

  • Jioh L. Jung (@ziozzang) — production user and contributor of substantial backend and runtime features (PR #11):

    • DuckDB SQL backbone alongside Postgres
    • OpenAI-compatible embeddings provider
    • Compile cache + TTL eviction + tool exposure filter
    • ontology_dirs config + onto_repo_list / onto_repo_load tools
    • Operational limits surfaced as [section] config
    • Docs alignment + resolver regression tests

  • Jason Smith (@rustforrecess) — diagnosed and fixed a real bug in the drift detector (PR #14):

    • Identified that anonymous restriction classes (and any blank-node IRIs returned by SPARQL) get freshly minted IDs on every parse, producing ~40 phantom add/remove pairs plus a Cartesian product of confidence-scored "renames" on detect(x, x) for typical OWL ontologies (Pizza tutorial repro). Shipped a surgical filter on the _: prefix in extract_vocabulary that bought time for the proper successor (RDFC 1.0 canonicalisation via Oxigraph 0.5.8 — landed in 2e329ee).
    • PR description quality (clear repro, minimal diff, full checklist of build/test/clippy/audit, CHANGELOG entry) is the model contributor experience.

  • Nicolas Geysse (@nicolas-geysse) — operational hardening of the HTTP transport, and the most rigorous issue reports this project has had (PR #78, PR #79):

    • serve-http had a graceful-shutdown CancellationToken that nothing ever cancelled, so the shutdown future pended forever, the process could only be killed, and the unix socket was left on disk afterwards. Fixed with ctrl-c and SIGTERM handling, a second-signal escape hatch for a stalled shutdown, and socket unlink on both paths.
    • An unauthenticated /health liveness route, correctly registered after the bearer layer so orchestrators can probe without credentials while /api and /mcp stay behind auth, plus a README table documenting that the HTTP surface is not read-only.
    • Seven issues filed against named commits, each citing file and line and stating the failure mode rather than the code smell. Several found silent-corruption paths rather than crashes, which is the harder class to see.

  • Ladislav Gazo (@lgazo) — opt-in persistent triple store (PR #81):

    • A [storage] backend selector for the main graph: memory (default, unchanged) or persistent, opening a RocksDB-backed Oxigraph store at <data_dir>/triplestore so triples survive a restart. CLI flag, env override and config, with correct precedence, plus a drop-and-reopen round-trip test.
    • Handled the case that would otherwise have quietly not worked: the one-shot CLI subcommands read the same setting, so load then query share state. Kept sandbox stores in-memory so only the main graph is ever persistent, and documented Oxigraph's single-writer-per-directory constraint rather than leaving it to be discovered.
    • He wrote this on his fork in June and never opened a PR; it was found during a fork-network audit. If you are sitting on something similar, open the PR. We would rather review it than find it.

How to be listed here

Open a PR. If it lands on main (in whole or in part), you'll be added with a short note describing your contribution. Bots and machine-generated commits are not credited as people.