- SPARQL: The property path evaluator was sometime emitting duplicates when evaluating disjunctive patterns (`a|b`). It is now fixed.
- SPARQL: If written explicitly in the SPARQL query, the regular expressions are now compiled once and not for each row.
- SPARQL: Property path evaluation with both start and end variables bound has been optimized.
- SPARQL: Casts to `xsd:decimal` from `xsd:float` and `xsd:double` now properly fails on overflow instead of overflowing the internal 128-bits representation.
This follows [XPath casting rules](https://www.w3.org/TR/xpath-functions/#casting-to-decimal).
- Rust: The minimal supported Rust version is set at 1.60 and enforced using the CI.
- Python: Local builds will now target the specific Python version instead of [abi3](https://docs.python.org/3/c-api/stable.html).
abi3 wheels are still released on Pypi alongside new Python-version specific builds for Linux+GNU.
- SPARQL: Fixes a panic when the estimated upper bound of the results size was overflowing a `usize`.
- Python: Uses `typing.IO` in Python stubs instead of narrower interfaces.
- Upgrades RocksDB to 7.9.2, `quick-xml` to 0.27 and `pyo3` to 0.18.
- SPARQL: Evaluation of zero-length paths with both start and end unbounded now return only terms defined in the current graph but not terms only defined in the query.
For example, in `SELECT * WHERE { VALUES ?s { 1 } { ?s ex:p? ?o } }` the literal `1` won't be returned anymore if not in the queried graph.
- Python: type stubs are now compatible with Python 3.7 and Mypy strict mode.
- RDF/XML: allows entities declared using other entities.
- Server: The `/store` endpoints now has a `no_transaction` HTTP option for `POST` and `PUT` request to get better performances at the cost of transactional guarantees.
- Server: The `/store` endpoints now has a `lenient` HTTP option for `POST` and `PUT` request to ignore syntax errors (requires the `no_transaction` option).
- Server: allows path that are not valid UTF-8 in file path CLI arguments.
- Rust: `From<spargebra::Query>` to `oxigraph::Query` (thanks to @hobofan).
- [RDF-star](https://w3c.github.io/rdf-star/cg-spec) support. `Triple` is now a possible `Term`. Serialization formats and SPARQL support have been updated to match the [latest version of the specification draft](https://w3c.github.io/rdf-star/cg-spec/2021-07-01.html).
- Fast data bulk load with the `Store``bulk_load_dataset` and `bulk_load_graph` methods and a special command line option of the server.
- It is now possible to quickly backup the database using the `backup` method.
- Rust: `*Syntax::from_extension` to easy guess a graph/dataset/sparql result format from a file extension.
- Rust: Custom SPARQL functions are now supported using `QueryOptions::with_custom_function`.
- Rust: Simple in-memory graph (`Graph`) and dataset (`Dataset`) data structures with canonicalization.
- Nightly build of the server binary and docker image, and of pyoxigraph wheels.
-`Store` operations are now transactional using the "repeatable read" isolation level:
the store only exposes changes that have been "committed" (i.e. no partial writes) and the exposed state does not change for the complete duration of a read operation (e.g. a SPARQL query) or a read/write operation (e.g. a SPARQL update).
the `Store``transaction` method now allows to do read/write transactions.
-`RDF-star <https://w3c.github.io/rdf-star/cg-spec>`_ is now supported (including serialization formats and SPARQL-star). :py:class:`.Triple` can now be used in :py:attr:`.Triple.object`, :py:attr:`.Triple.object`, :py:attr:`.Quad.subject` and :py:attr:`.Quad.object`.
- SPARQL: It is now possible to compare `rdf:langString` literals with the same language tag.
- SPARQL: The parser now validates more carefully the inputs following the SPARQL specification and test suite.
- SPARQL: Variable scoping was buggy with "FILTER EXISTS". It is now fixed.
- Rust: RDF model, SPARQL parser and SPARQL result parsers have been moved to stand-alone reusable libraries.
- Rust: HTTPS is not supported by default with the `http_client` option. You need to enable the `native-tls` or the `rustls` feature of the `oxhttp` crate to enable a TSL layer.
- Rust: The error types have been cleaned.
Most of the `Store` methods now return a `StorageError` that is more descriptive than the previous `std::io::Error`.
The new error type all implements `Into<std::io::Error>` for easy conversion.
- Rust: There is now a `Subject` struct that is the union of `NamedNode`, `BlankNode` and `Triple`.
It is The used type of the `subject` field of the `Triple` and `Quad` structs.
- Rust: The SPARQL algebra is not anymore publicly exposed in the `oxigraph` crate. The new `oxalgebra` crate exposes it.
- Rust: `UpdateOptions` API have been rewritten. It can now be built using `From<QueryOptions>` or `Default`.
- Server: The command line API has been redesign. See the [server README](server/README.md) for more information.
- Server: The HTTP implementation is now provided by [`oxhttp`](https://github.com/oxigraph/oxhttp).
- Server: The HTTP response bodies are now generated on the fly instead of being buffered.
- Python: The `SledStore` and `MemoryStore` classes have been removed in favor of the `Store` class.
- JS: The `MemoryStore` class has been renamed to `Store`.
- JS: The [RDF/JS `DataFactory` interface](http://rdf.js.org/data-model-spec/#datafactory-interface) is now implemented by the `oxigraph` module itself and the `MemoryStore.dataFactory` property has been removed.
- The implementation of SPARQL evaluation has been improved for better performances (especially joins).
- The TLS implementation used in SPARQL HTTP calls is now [rustls](https://github.com/rustls/rustls) and not [native-tls](https://github.com/sfackler/rust-native-tls). The host system certificate registry is still used.
- Spargebra: The basic RDF terms are now the ones of the `oxrdf` crate.
### Removed
-`SledStore` and `MemoryStore`. There is only the `RocksDbStore` anymore that is renamed to `Store`.
-`oxigraph_wikibase` is now stored in [its own repository](https://github.com/oxigraph/oxigraph-wikibase).
- Rust: `From` implementations between `oxigraph` terms and `rio_api` terms.
Many thanks to [Thad Guidry](https://github.com/thadguidry), [James Overton](https://github.com/jamesaoverton) and [Jeremiah](https://github.com/jeremiahpslewis) who sponsored the project during the development of this version.
- The HTTP server allows to query the union of all graphs using the `union-default-graph` query parameter and to use the union graph for update `WHERE` clauses using the `using-union-graph` parameter.
- Exposes Sled flush operation (useful for platforms without auto-flush like Windows or Android).
- Fixes a possible out of bound panic in SPARQL query evaluation.
- [SPARQL 1.1 Update](https://www.w3.org/TR/sparql11-update/) support for Rust, Python and JavaScript. All store-like classes now provide an `update` method.
- [SPARQL 1.1 Graph Store HTTP Protocol](https://www.w3.org/TR/sparql11-http-rdf-update/) partial support in `oxigraph_server`. This protocol is accessible under the `/store` path.
- A simple built-in HTTP client. In the Rust library, is disabled by default behind the `http_client` feature. It powers SPARQL federation and SPARQL UPDATE `LOAD` operations.
-`std::str::FromStr` implementations to `NamedNode`, `BlankNode`, `Literal`, `Term` and `Variable` allowing to easily parse Turtle/SPARQL serialization of these terms.
-`(Memory|RocksDB|Sled)Store::prepare_query` methods. It is possible to cache SPARQL query parsing using the `Query::parse` function and give the parsed query to the `query` method.
- Loading data into `oxigraph_server` is now possible using `/store` and not anymore using `/`.
For example, you should use now `curl -f -X POST -H 'Content-Type:application/n-quads' --data-binary "@MY_FILE.nq" http://localhost:7878/store` to add the N-Quads file MY_FILE.nt to the server dataset.
The disk data format has been changed between Oxigraph 0.1 (version 0) and Oxigraph 0.2 (version 1). Data is automatically migrated from the version 0 format to the version 1 format when opened with Oxigraph 0.2.
-`QueryOptions::with_default_graph` now takes an `impl Into<GraphName>` instead of an `impl Into<NamedNode>`.
-`QueryOptions::with_named_graph` now takes an `impl Into<NamedOrBlankNode>` instead of an `impl Into<NamedNode>`.
-`pyoxigraph``query` methods now takes two new parameters, `default_graph` and `named_graphs`. `default_graph_uris` and `named_graph_uris` parameters are deprecated.
