Skip to content

FAQ and troubleshooting

Why does install fail on Rust 1.75 / 1.80?

Oxiland’s MSRV is 1.87 (edition 2024) so CI and Oxigraph 0.5.9 stay aligned. Upgrade with rustup update stable.

Why does pip install oxiland try to build from source?

Published packages are wheels only (no sdist). If pip cannot find a wheel for your platform/Python, it may attempt a source build and fail. Use CPython 3.10–3.14 on a platform with published wheels, or build from a git checkout with maturin (Python installation guide).

Tip 0.13.0 is the current package version.

Why not use Oxigraph directly?

If you do not need Redland-shaped APIs, inventories, or migration mapping, use Oxigraph. Oxiland adds a compatibility-oriented facade and evidence process on top of the same engine. See positioning.

Is Oxiland a network database?

No. In Rust and Python, a persistent model is an embedded local store inside the application process. Oxiland does not provide a server, authentication, replication, tenant isolation, or managed backups. See the Rust or Python production guide.

Is Oxiland “Redland-compatible”?

Only in evidence-scoped senses documented in the parity ledger. Tip 0.13 retains the 0.11 demonstrated Redland matrix for Rust, Python, and C source + librdf-compat packaging, and adds the ADR-029 suite-wide faster-than-Redland performance claim. Remaining behavioral gaps are listed in C ABI limitations. Oxiland is not a silent 100% librdf port and not an rdflib adapter. “Safe-API accounting” means inventoried symbols are classified, not that every Redland behavior is drop-in. See the C ABI guide.

Does Oxiland ship a C package on crates.io?

No. oxiland-capi is publish = false. Build it from a repository checkout (cargo build -p oxiland-capi). See C ABI.

What does “Verified” mean in the parity ledger?

It means the curated inventory rows for that subsystem have linked implementation and tests for the stated scope—not that every Redland symbol or differential fixture passes. Read the inventory JSON and milestone report.

Parse left data in my model / on disk

You used a progressive load. On failure, already-inserted quads remain (and on Fjall they are durable). In Python, use load_path(model, path, transactional=True) for atomic import or keep the default collecting=True for parse-then-insert behavior. In Rust, use load_transactional or load_collecting. See the Python data guide or Rust I/O guide.

Model::import_nquads_from_path merges into the existing model; it does not replace the store.

text/plain / .xml / guess returns Unsupported

Intentional. Pick an explicit Syntax or an unambiguous media type / extension (.nt, .ttl, application/rdf+xml, …).

Named graphs vanished / parse error on N-Quads

The default graph target rejects named-graph input. Rust callers can use GraphTarget::Dataset for TriG/N-Quads datasets. The Python API does not expose that target; load one compatible graph with graph= or create named graphs programmatically. See Python RDF I/O.

The CLI has the same dataset-import limitation. Restore multi-graph N-Quads backups with Rust Model::import_nquads_from_path or Python Model.import_nquads. Rust callers can also use Parser::parse_path_with_extension or configure GraphTarget::Dataset.

Wrong, unknown, or not-compiled storage backends

Backend selection is explicit and fails closed:

  • Unknown names raise Unsupported / UnsupportedError (“not recognized”).
  • Known but not compiled optional names (for example future adapters such as redb) raise a distinct “known but not compiled into this build” error—they never fall back to Fjall or memory.
  • Compiled backends for the default build are memory and fjall.

Use compiled_backends() (Rust and Python) to list what this build links. storage_backend_available(name) returns True only for compiled backends; other names error as above. CLI --storage and C librdf_new_storage follow the same registry.

Fjall store uses a lot of RAM

Fjall mode keeps a full Oxigraph in-memory working set for querying. Plan RAM for the dataset size; use streaming parse/serialize APIs for large files. See Performance.

Transaction methods do nothing / buffer forever (Python)

Call add / remove inside with model.transaction() as txn: — methods outside an entered context raise UnsupportedError.

Will my format-v1 store open after an upgrade?

Format v1 reopen is promised for patch releases in 0.4.x–0.13.x. Export N-Quads before any future format-v2 migration. See persistence and the support policy.

Can multiple processes write the same store path?

Do not use a shared writable directory as an inter-process coordination mechanism. Keep one owning application/service per store and expose controlled network concurrency above it. Model sharing and locks are in-process guarantees.

Where do I report bugs or security issues?

Performance guidance?

See Performance for published suites, host scope, and what is not claimable. Prefer streaming parse/serialize APIs for large data; avoid collecting full iterators. Fjall mode keeps a full in-memory working set—plan RAM accordingly.