Eventual-Inc/Daft - CodSpeed

Daft

Blog Docs Changelog

Performance History

Latest Results

fix(core): compare nested float keys with canonicalized zeros Join, groupby, and distinct equality for lists and structs goes through Arrow's row format, which uses IEEE totalOrder and keeps -0.0 distinct from 0.0. Flat float keys already compare as equal, but a list key [-0.0] never matched [0.0], and distinct/groupby split identical keys. Canonicalize negative zeros inside list and struct arrays before building the row comparator, reusing the float kernel's recursive helper. Arrays without a negative zero keep their original allocation.

hello-peter-tang:fix/7656-negative-zero

6 hours ago

test(iceberg): cover is_null predicates over initial defaults

hello-peter-tang:fix/7659-iceberg-initial-default

6 hours ago

fix(functions): accept null operands in try arithmetic try_add/try_subtract/try_multiply rejected a null literal or all-null column because the supertype check ran before inference. Let a null side take its type from the numeric operand, matching ordinary arithmetic, and error only when both inputs are null. Also correct the docstring: mixed int64/uint64 promotes to float64, while narrower mixed integers widen to a checked int64. Add tests for null literals, all-null columns, and the both-null error.

hello-peter-tang:feat/7232-try-arithmetic

6 hours ago

test: cover Parallel classification batching

devin-gupta:pg/parallel-decisions-classifier

8 hours ago

docs: remove stale Parallel provider link

devin-gupta:pg/parallel-decisions-classifier

8 hours ago

feat(functions): add overflow-checked try_add, try_subtract and try_multiply Integer arithmetic in Daft silently wraps on overflow, so large counters can produce wrong results with no signal. Add opt-in try_* variants that promote integer inputs to 64 bits and return NULL where checked_add/checked_sub/ checked_mul would overflow, including negative results for unsigned subtraction. Floats keep IEEE semantics and are never nulled.

hello-peter-tang:feat/7232-try-arithmetic

10 hours ago

fix(iceberg): budget live rows when limiting scans with positional deletes The file-level limit budget subtracted each data file's physical record count, so a file whose rows are largely removed by positional deletes falsely used up the budget and later files were never scheduled. Subtract an estimate of the rows surviving deletes instead. Over-subtracting can only schedule extra files, never stop before enough live rows are planned. Closes #7559

hello-peter-tang:fix/7559-iceberg-limit-positional-deletes

11 hours ago

docs: trim Parallel provider reference

devin-gupta:pg/parallel-decisions-classifier

11 hours ago

Latest Branches

0%

fix(core): treat -0.0 and 0.0 as equal#7661

6 hours ago

655892d

hello-peter-tang:fix/7656-negative-zero

0%

fix(iceberg): read Iceberg v3 initial-defaults instead of NULL#7674

6 hours ago

95a274d

hello-peter-tang:fix/7659-iceberg-initial-default

0%

feat(functions): add overflow-checked try_add, try_subtract and try_multiply#7680

6 hours ago

d539d12

hello-peter-tang:feat/7232-try-arithmetic

© 2026 CodSpeed Technology

Home Terms Privacy Docs