vortex-data/vortex - CodSpeed

vortex

Blog Docs Changelog

Performance History

Latest Results

Allow caching LayoutReader in VortexFile (#8244) Not-reinitializing LayoutReader saves us some time. As a related change, make root() static throughout the program. Before ScanBuilder::new() created a new root() which didn't compare with the old one pointer-eq-wise which was an issue Signed-off-by: Mikhail Kot <mikhail@spiraldb.com>

develop

38 minutes ago

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

refactor/parent-ref-stack-allocated

12 hours ago

Unify Matcher on a single ParentRef-based entry point Replace the dual Matcher entry points (matches/try_match over ParentRef plus matches_ref/try_match_ref over ArrayRef) with one generic pair over a new sealed AsParent trait, implemented by ArrayRef and ParentRef. Each matcher now has a single Match associated type and one implementation. - Merge CanonicalView/ParentCanonicalView and ColumnarView/ ParentColumnarView into single ParentView-backed enums; drop the ScalarFnArrayView inner-view generic and ParentScalarFnArrayView alias. - Remove the unused ParentRef::optimize; ArrayParts::optimize remains the construction-side entry point. - Add ParentView::materialize_view and Array::as_parent_view to bridge call sites that need heap-only ArrayView APIs (free for heap-backed parents). - Heap-only compute kernels downcast via as_typed to keep ArrayView ergonomics; matcher-driven call sites use explicit materialization. Signed-off-by: Robert Kruszewski <github@robertk.io>

refactor/parent-ref-stack-allocated

12 hours ago

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

rk/sparkstats

12 hours ago

Optimize `fsst_compress_offsets_overflow_i32` by 2X locally (#8324) ## Summary Optimize the infamous `fsst_compress_offsets_overflow_i32` test by dropping the symbol table - instead compressing with an empty symbol table to get constant escapes. Locally this change drops the test's runtime from 136s to 57s (in normal test profile), which should have a real impact on our CI runtimes. This change is provided by the new fable-5 model. Signed-off-by: Adam Gutglick <adam@spiraldb.com>

develop

12 hours ago

Fix generation of stat pruning expression for unsupported dtypes Signed-off-by: Robert Kruszewski <github@robertk.io>

rk/statfixes

12 hours ago

Allow running reduce_parent operations on stack allocated parents Add indirection logic to support stack allocated array parents in reduce rules. This lets us avoid heap allocation for commonly used operations that are often immediately optimised away. Signed-off-by: Robert Kruszewski <github@robertk.io>

refactor/parent-ref-stack-allocated

12 hours ago

ArrowImportTable impl Signed-off-by: Andrew Duffy <andrew@a10y.dev>

aduffy/df-geo

12 hours ago

Latest Branches

-21%

Allow running reduce_parent operations on stack allocated parents#7751

12 hours ago

5b8d3e9

refactor/parent-ref-stack-allocated

+29%

Support reporting statistics in spark datasource#8057

12 hours ago

deb1dc0

rk/sparkstats

+20%

Fix generation of stat pruning expression for unsupported dtypes#8326

12 hours ago

0c6b621

rk/statfixes

© 2026 CodSpeed Technology

Home Terms Privacy Docs