vortex-data/vortex - CodSpeed

vortex

Blog Docs Changelog

Performance History

Latest Results

comments Signed-off-by: Robert Kruszewski <github@robertk.io>

claude/remove-cast-values-crutch-7u8efe

10 hours ago

refactor: give patches correct dtype by construction `Patches::cast_values` was a crutch that papered over patch values not matching the parent array's dtype, and ALPRD/Sparse each carried a construction-time normalization step (`canonicalize_patches`, `normalize_patches_dtype`) to fix patches up after the fact. Remove all three: patch-producing operations now yield the correct dtype directly, and constructors assert rather than normalize. - `Patches::mask` widens the surviving values to nullable. Masking an array always yields a nullable result, so the masked patches now carry the correct dtype and callers (ALP mask) no longer re-cast them. - ALP take/mask: `Patches::take`/`mask` already produce values matching the taken/masked encoded array, so build via `ALP::try_new`, which asserts the patch dtype instead of forcing it. - bitpacking take: `PrimitiveArray::patch` (via `Validity::patch`) already asserts matching nullability, so no cast is needed. - ALPRD: drop `canonicalize_patches`. Exceptions are always the non-nullable left-parts dtype; `validate_parts` now requires exactly that, `take` casts the taken exceptions back to it (folded into the take call), and the other construction paths already satisfy it. Neither `ALPRD::try_new` nor `RDEncoder::encode` needs an `ExecutionCtx` anymore (the RD split does no execution); their callers that only plumbed a ctx to pass it down are cleaned up too. - Sparse: drop `normalize_patches_dtype`. Both `Sparse::try_new` and the patches-based paths now assert that the values/patches already match the fill dtype instead of casting; callers construct matching values. Signed-off-by: Robert Kruszewski <robert@spiraldb.com> Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UDa4BYuhQyRHWCUGTanVzE Signed-off-by: Robert Kruszewski <github@robertk.io>

claude/remove-cast-values-crutch-7u8efe

10 hours ago

chore: address PR review comments - Replace non-ASCII characters in comments with ASCII. - Document why catalog registration needs RunFunctionInTransaction. - Reference the FFI function in register_geo_aliases doc. Signed-off-by: Nemo Yu <zyu379@wisc.edu>

nemo/geo-native-pushdown

15 hours ago

feat: support ST_DWithin pushdown in vortex Signed-off-by: Nemo Yu <zyu379@wisc.edu>

nemo/geo-native-pushdown

17 hours ago

chore: test CUDA Arrow Device capsule exports Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>

ad/pycudf4

17 hours ago

chore: test CUDA Arrow Device capsule exports Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>

ad/pycudf4

17 hours ago

chore: test CUDA Arrow Device capsule exports Signed-off-by: Alexander Droste <alexander.droste@protonmail.com>

ad/pycudf4

17 hours ago

Bound scan read state lifetime Signed-off-by: Nicholas Gates <nick@nickgates.com>

ngates/layout27

17 hours ago

Latest Branches

-15%

Patches have correct dtype by construction instead of normalised during array construction#8626

10 hours ago

9597529

claude/remove-cast-values-crutch-7u8efe

+12%

feat: support `ST_DWithin` pushdown in vortex#8625

15 hours ago

94674e0

nemo/geo-native-pushdown

+4%

chore: test CUDA Arrow Device capsule exports#8624

17 hours ago

8ebaae5

ad/pycudf4

© 2026 CodSpeed Technology

Home Terms Privacy Docs