0.86.0 (Draft)

Draft — GitHub Release

Changes

⚠️ Breaks

• vortex-file: open_cached uses an explicit cache key (#9733) @myrrc

• feat: require Send on VortexWrite and its futures (#9648) @claude[bot]

• Add physical parameters to OutputSink (#9585) @connortsui20

• Let a row dispatch declare a runtime output dtype (#9581) @connortsui20

✨ Features

• cache DTypes parsed from flatbuffers (#9765) @myrrc

• Support runend canonicalization for lists (#9708) @danking

• Bloom filter implementation (#9398) @joacoc

• refactor(array): use buffer allocator in execution context (#9669) @gatesn

• feat(buffer): add allocator-backed storage (#9668) @gatesn

• Add Utf8 types for RowFn (#9715) @connortsui20

• Add mechamism to lock edition definition and how to evolve objects in editions (#9660) @robert3005

• duckdb: propagate per-file “partition” stats (#9639) @myrrc

• feat(duckdb): report per-file column statistics from the vortex COPY writer (#9471) @moshap-firebolt

• feat(ffi): add vx_scalar_new_extension for extension-typed literals (#9691) @balicat

• Support run-end canonicalize for decimals (#9707) @danking

• Support canonicalising decimal RunEnd arrays (#9665) @onursatici

• Add filter-and-scatter RowFn execution (#9521) @connortsui20

• feat(cuda): add a FastLanes delta decode kernel (#9535) @joseph-isaacs

• duckdb: support hive partitioning and ‘filename’ column (#9540) @myrrc

🚀 Performance

• Push get_item through wrapper arrays (#9719) @danking

• Avoid materializing constant structs for select (#9718) @danking

• Avoid row index canonicalization for trivial masks (#9717) @danking

• perf(fastlanes): borrow unpack inputs (#9731) @gatesn

• perf(buffer): store alignment exponents (#9730) @gatesn

• Avoid materializing constant structs for get_item (#9716) @danking

• Trim constant-operand cases from scalar function benchmarks (#9696) @connortsui20

• Decode RowFn constants directly (#9680) @connortsui20

• Pack Boolean RowFn output directly (#9626) @connortsui20

• Conditionally slice elements to selected index bounds before filtering wide list arrays (#9679) @mhk197

• Avoid iterative drop for shallow expression trees (#9657) @robert3005

• Better fixed-width buffer filtering (#8723) @robert3005

• Use indexed sources for fallible constant rows (#9629) @connortsui20

• Map constant RowFn inputs directly (#9628) @connortsui20

• Build packed boolean output (#9620) @connortsui20

• Avoid single-chunk grouped accumulator states (#9597) @mhk197

• refactor(vortex-spatial): execute distance with RowFn (#9616) @HarukiMoriarty

• Measure numeric arithmetic and comparison benchmarks per CPU feature (#9599) @connortsui20

• Benchmark primitive comparison across CPU features (#9598) @connortsui20

• duckdb: MultiFileReader interface (#9383) @myrrc

• Use dense retry for nullable row functions (#9517) @connortsui20

🐛 Bug Fixes

11 changes

• Support unsigned sequence in the DuckDB exporter (#9720) @danking

• fix(dict): cap max_dict_bytes at the u32 range BinaryView addresses (#9690) @mnshdw

• fix(dict): clear encoder state in DictEncoder::reset (#9693) @mnshdw

• bench: run the full compress suite under --gpu-decompress (#9399) @joseph-isaacs

• duckdb: forbid aggregations and filter pushdowns over virtual columns (#9238) @myrrc

• fix: update the delta snapshot (#9608) @rapour

• feat: sequence cast compute (#8403) @siddarth2810

• fix: price delta’s bases and tail padding in its compression estimate (#9602) @rapour

• fix: pad delta remainder with last value (#9557) @rapour

• fix: stabilize conjunct selectivity (#9543) @joseph-isaacs

• Carry partial RowFn validity with MaskValues (#9534) @connortsui20

📖 Documentation

7 changes

• docs(vortex-layout): mark the bloom filter aggregate as unstable (#9753) @connortsui20

• Clarify Boolean collection code generation (#9676) @connortsui20

• docs(vortex-spatial): add crate overview (#9683) @HarukiMoriarty

• Clarify RowFn infallibility documentation (#9618) @connortsui20

• docs(java): replace the Java API example that no longer compiles (#9635) @jackylee-ch

• docs: filter pushdown is connected, not future work (#9637) @jackylee-ch

• duckdb: add release docs (#9575) @myrrc

🧰 Maintenance

20 changes

• Move benchmark machines to c8gd.metal-24xl (#9761) @robert3005

• Return the lazy filtered row index array directly (#9760) @robert3005

• refactor(decimal): dispatch numeric kernels on one storage width (#9756) @mhk197

• chore: deny clippy::inline_always at the workspace level (#9726) @joseph-isaacs

• nit: go directly from alignment to usize (#9666) @danking

• docs(spark): refresh the user guide against the shipped connector (#9607) @jackylee-ch

• docs(vx): state which inspection commands support –json (#9638) @jackylee-ch

• Add list filter benchmark (#9686) @mhk197

• Use SumV2 in list_sum (#9621) @mhk197

• fix(policy): update policy to match the new claude and codex bot ids (#9650) @robert3005

• refactor(vortex-spatial): execute geo kernels with RowFn (#9623) @HarukiMoriarty

• Upgrade fastlanes to 0.7.0 (#9641) @robert3005

• Bump Datafusion to 55 (#9334) @AdamGS

• refactor(runend): clarify filter implementation (#9633) @connortsui20

• refactor(dict): clarify compute invariants (#9632) @connortsui20

• refactor: simplify array optimizer (#9634) @connortsui20

• Add sum_v2 with SQL empty-input semantics (#9590) @mhk197

• Update anthropics/claude-code-action digest to 16b3b31 (#9559) @renovate[bot]

• Extend MaskValuesRef to filter execution (#9542) @connortsui20

• Upgrade toolchain to 1.98.0 (#9539) @robert3005