feat(scanner): plan every read through the logical planner - #8561

Closed

wjones127 wants to merge 28 commits into

lance-format:mainfrom

wjones127:feat/logical-scan-planner-default

Commits on Aug 16, 2026

fix: stamp schema metadata onto FilteredReadExec output batches

wjones127
and
claude
committed
2ca66d9
fix: carry input ordering through FilteredReadExec take mode

wjones127
and
claude
committed
8200cf5
spike: logical plan read path for the scanner

wjones127
and
claude
committed
a6f5d24
spike: full-text search through the logical plan path
wjones127
committed
7ad5c0e
spike: ordering, refine, fast_search, and the flat-KNN order contract

wjones127
and
claude
committed
5e70fb6
feat(spike): unify index-coverage splits, including overlays

wjones127
and
claude
committed
4e9fc55
fix(spike): replay the row mask from a memory source

wjones127
and
claude
committed
c4a4c2f
refactor(spike): make mandatory rewrites analyzer rules

wjones127
and
claude
committed
4dce434
refactor(spike): give the scan leaf's index query a place in the plan

wjones127
and
claude
committed
4bbbf3b
spike: finish logical scan planner (guards, oracle, benchmark)

wjones127
and
claude
committed
f73e363
spike: cache the query-independent half of planning state

wjones127
and
claude
committed
6978db1
refactor: split the FTS planning module by node type

wjones127
and
claude
committed
4ab875a
refactor: split the planner's nodes, rules, and lowering by owner

wjones127
and
claude
committed
0bd671c
test: split the planner's tests by what they exercise

wjones127
and
claude
committed
4f332e7
feat: plan scans on legacy (v1) storage
wjones127
committed
2f48af5
feat: support include_deleted_rows and strict_batch_size
wjones127
committed
431c1b4
feat: support with_index_segments, and adopt the index's metric
wjones127
committed
3041f31
perf: drop the restated search ordering with EnforceSorting
wjones127
committed
f4fb229
feat: express Lance search as DataFrame operators
wjones127
committed
107d125
feat: plan aggregates as stock DataFusion nodes
wjones127
committed
4f0b7cc
fix: reject multivector search instead of mis-planning it
wjones127
committed
a8ede4a
feat: plan batch vector search
wjones127
committed
4b43eb4
feat: plan multivector search, row-id takes and _rowoffset
wjones127
committed
590aa57
feat: plan late materialization in the scan leaf
wjones127
committed
98fb916
fix: keep a filtered column out of the late projection
wjones127
committed
ef3abbe
perf(scanner): fold stacked projections into one

wjones127
and
claude
committed
44acc8e
fix(scanner): avoid a capacity underflow in row_ids_for_mask

wjones127
and
claude
committed
b8721ae
feat(scanner): plan every read through the logical planner

wjones127
and
claude
committed
79004a9

Navigation Menu

Uh oh!

feat(scanner): plan every read through the logical planner - #8561

feat(scanner): plan every read through the logical planner#8561
wjones127 wants to merge 28 commits into
lance-format:mainfrom
wjones127:feat/logical-scan-planner-default

Commits on Aug 16, 2026

fix: stamp schema metadata onto FilteredReadExec output batches

fix: carry input ordering through FilteredReadExec take mode

spike: logical plan read path for the scanner

spike: full-text search through the logical plan path

spike: ordering, refine, fast_search, and the flat-KNN order contract

feat(spike): unify index-coverage splits, including overlays

fix(spike): replay the row mask from a memory source

refactor(spike): make mandatory rewrites analyzer rules

refactor(spike): give the scan leaf's index query a place in the plan

spike: finish logical scan planner (guards, oracle, benchmark)

spike: cache the query-independent half of planning state

refactor: split the FTS planning module by node type

refactor: split the planner's nodes, rules, and lowering by owner

test: split the planner's tests by what they exercise

feat: plan scans on legacy (v1) storage

feat: support include_deleted_rows and strict_batch_size

feat: support with_index_segments, and adopt the index's metric

perf: drop the restated search ordering with EnforceSorting

feat: express Lance search as DataFrame operators

feat: plan aggregates as stock DataFusion nodes

fix: reject multivector search instead of mis-planning it

feat: plan batch vector search

feat: plan multivector search, row-id takes and _rowoffset

feat: plan late materialization in the scan leaf

fix: keep a filtered column out of the late projection

perf(scanner): fold stacked projections into one

fix(scanner): avoid a capacity underflow in row_ids_for_mask

feat(scanner): plan every read through the logical planner