nordisk/graph-ar
Commits
369c519cf8b1bench: BENCH.md and the raw summary in the repositorycd4e15ee55a9bench: types.sh checks every value type between native and C++ (prep-types, typed checksums); design.md gaps follow list<T>, date, timestamp, ORC4dc865318c4bGREEN: list<T> for every scalar type, date and timestamp values in the row builders, ZSTD ORC chunks (78/0, 83/0 with orc)8097c2b8973dTEST: list<T>, date and timestamp through info files, row builders, table writers and every format; ORC chunks compressed (RED: DataType::List is list<string> only, no date/timestamp values, ORC uncompressed, ORC lists panic)2d757cb7e9f9bench: BENCH_THREADS caps the native decode threadseecfc388d94fdocs: perf.md without the in-house image named85f31fc56c3bench/perf: the stage bench moves out of the crate manifest (znippy-zoomies and memmap2 are no longer dev-dependencies)bade5737e3c0docs: perf.md — fill in the 100M-edge after numbers427b2b9a9ef8Merge perf/graphar-fast-2: faster writer, delta-packed ids, lookup pushdown, parallel decode, perf docs9ecd9a878567docs: perf.md — algorithms, bench tables before/after, the static search tree answeraad4ced3d29cperf: Vec-backed properties, counting-sort writer, delta-packed ids, row-window lookups, parallel decode1d20681d2e21bench: README carries the ASF header (license-eye: 46 valid, 0 invalid)40f7bc5e6e42ci: licence check skips bench/README.mdcd86afc393c4bench: C++ CSR read collects every offset (AdjListOffsetArrowChunkReader::GetChunk returns only the first Arrow chunk of the offset column)eb6aead9ee34tests: ORC cases only with the orc feature, so a default cargo test passes (62/0; 64/0 with --all-features); CI runs both12485bf8e1cfGREEN: CSR reads int32 offsets and index columns (widened to int64); bench conform command (64/0)641c19702af9TEST: CSR reads int32 offsets, as in the Spark-written neo4j and nebula samples (RED: chunk has no int64 column _graphArOffset)e2a34b553455bench: C++ writers get explicit ZSTD Parquet options (their default without options is uncompressed); reads use a retried reference writeda5cf77dcbd1bench: native Rust vs GraphAr C++ vs graphar-rs harnesses, run.sh and summarize.py1fd49a1141cdGREEN: CSV and JSON property chunks are read with the declared property types (io::read_chunk_typed; 63/0)e5f4e4f9a8eaTEST: CSV property chunks are read with the declared property types (RED: inferred, a string creationDate comes back as a timestamp)0f77c3788c99GREEN: read_csr reads offset chunks padded to chunk_size + 1 rows (62/0)7d14a95596c5TEST: read_csr reads offset chunks padded to chunk_size + 1 rows, as the GraphAr C++ table writer writes them (RED: rejected)e71c040f324bGREEN: info files indent block sequences like the upstream files (metadata::yaml emitter; GraphAr C++ loads info files written here; 61/0)8a9cfcdfa4d4TEST: info files indent block sequences like the upstream files (RED: serde_yaml writes them flush, which the GraphAr C++ YAML reader rejects)80a940a163ffGREEN: an empty graph prefix is not written100f986f9990TEST: an empty graph prefix is not written, so GraphAr C++ loads the info file (RED: prefix: '' rejected by the C++ YAML reader)976f02819e59GREEN: Parquet chunks default to ZSTD like the GraphAr C++ writer (io::write_chunk and ChunkStore share writer_properties; 59/0)a4a31fb6eeb3TEST: Parquet chunks default to ZSTD like the GraphAr C++ writer (RED: written uncompressed)6915204c7862Standalone crate prepared for an Apache GraphAr contributiondbbf254ccae9GREEN: graphar writes and reads the upstream gar/v1 on-disk layout (extensionless chunk<N>, _graphAr* columns, vertex_count/edge_count, write_vertex_table/write_edge_table, EdgeReader::read_csr; ldbc_sample parquet/csv/orc read and validate; 58/0 with --features orc)811f7b71484cTEST: graphar writes and reads the upstream gar/v1 on-disk layout (RED: extensionless chunk<N>, _graphAr* columns, vertex_count/edge_count files, table writers, CSR read, upstream testing data)8af306be07c2publish prep: names out of the shipped files, dev notes out of the packagesd447ba1df162graphar: retire orc-rust-arrow59 fork, use upstream orc-rust 0.941d747246115vendor orc-rust on arrow 59, the same way the spark-connect fork works88658e34f714arrow 59, and ORC behind a feature because orc-rust has no 593b44ada05764Commit the formatting rustfmt.toml already declared, so the tree stops re-dirtying88b61c4b44fbchore: remove two empty test-module stubs in graphar (imports-only, no #[test])ac56ea511eectest(graphar): chunk codec roundtrip coverage (all 5 FileTypes) + light timing baseline (ArrowIpc ~17x cheaper than Parquet)2aa53c3c1cd8wire: bencher + emit markers for GraphAr graph-load read pathf73016c3e2f2knut: add repository metadata to graphar/knut member crates (crates.io publish gate)2bba0b6f9fb1deps(edda): point moved crates at published versions; drop cross-repo pathfd91a49aff93graphar: direct tests for merge_property_groups error branches and validate_edge_info2f2456f82e5ffeat(introspection): close the remaining functional_status emit holes (#8)7ce2f397ebacUpdate Cargo.lockfe7e31c78aecdocs: relocate per-crate cards into each crate's own .nornir/312f8dd1d267Merge branch 'agent/knut-tests'38bf84f54d24Merge branch 'agent/knut-unimplemented'3d6d7d9d7e39Merge branch 'agent/knut-bugs'25b170e72790tests: close coverage gaps + fix the bugs they expose