feat(fts): add BM25F cross-field search - #7905

Open

sbrunk wants to merge 25 commits into

lance-format:mainfrom

sbrunk:combined-fields-bm25f

Commits on Jul 30, 2026

feat(fts): add combined_fields (BM25F) cross-field search
sbrunk
committed
perf(fts): order inverted-index docs by row_id at build time
sbrunk
committed
perf(fts): MAXSCORE + block-skip read pruning for combined_fields
sbrunk
committed
fix(fts): offload write_new_partition doc reorder off the async runtime
sbrunk
committed
fix(fts): reject duplicate combined_fields columns; review fixups
sbrunk
committed
fix(fts): invalidate row_ids_ascending cache on mutation; harden tests
sbrunk
committed
feat(java): add combined_fields (BM25F) full-text query
sbrunk
committed
fix(java): address review feedback on combined_fields query
sbrunk
committed
fix(fts): report the real schema from JsonTextStream
sbrunk
committed
fix(fts): score unindexed and partially indexed fragments in combined_fields
sbrunk
committed
perf(fts): blend combined_fields flat rows during tokenization
sbrunk
committed
test(fts): cover combined_fields flat path gaps
sbrunk
committed
fix(bench): never delete a caller-supplied WORK directory
sbrunk
committed
fix(fts): make the combined_fields column/weight pairing unbreakable
sbrunk
committed
fix(fts): round the combined_fields MAXSCORE ceiling conservatively
sbrunk
committed
perf(fts): offload the combined_fields indexed scan off the async runtime
sbrunk
committed
fix(fts): score legacy list indexes by row in combined_fields
sbrunk
committed
refactor(fts): port combined_fields onto the document identity layer
sbrunk
committed
fix(fts): drop stale segments from combined_fields when a data overlay exists
sbrunk
committed
fix(fts): decide the ascending-address gate on liveness, not on live_docs
sbrunk
committed
test(fts): drive the combined_fields identity oracle through both doc sources
sbrunk
committed
fix(fts): report index cache lookups from the combined_fields scorer build
sbrunk
committed
style(fts): reword two comments the typos hook rejects
sbrunk
committed
test(fts): cover combined_fields nested in a compound query
sbrunk
committed
refactor(fts): share the FTS score sort keys and tie comparator
sbrunk
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(fts): add BM25F cross-field search - #7905

feat(fts): add BM25F cross-field search#7905
sbrunk wants to merge 25 commits into
lance-format:mainfrom
sbrunk:combined-fields-bm25f

Commits on Jul 30, 2026

feat(fts): add combined_fields (BM25F) cross-field search

perf(fts): order inverted-index docs by row_id at build time

perf(fts): MAXSCORE + block-skip read pruning for combined_fields

fix(fts): offload write_new_partition doc reorder off the async runtime

fix(fts): reject duplicate combined_fields columns; review fixups

fix(fts): invalidate row_ids_ascending cache on mutation; harden tests

feat(java): add combined_fields (BM25F) full-text query

fix(java): address review feedback on combined_fields query

fix(fts): report the real schema from JsonTextStream

fix(fts): score unindexed and partially indexed fragments in combined_fields

perf(fts): blend combined_fields flat rows during tokenization

test(fts): cover combined_fields flat path gaps

fix(bench): never delete a caller-supplied WORK directory

fix(fts): make the combined_fields column/weight pairing unbreakable

fix(fts): round the combined_fields MAXSCORE ceiling conservatively

perf(fts): offload the combined_fields indexed scan off the async runtime

fix(fts): score legacy list indexes by row in combined_fields

refactor(fts): port combined_fields onto the document identity layer

fix(fts): drop stale segments from combined_fields when a data overlay exists

fix(fts): decide the ascending-address gate on liveness, not on live_docs

test(fts): drive the combined_fields identity oracle through both doc sources

fix(fts): report index cache lookups from the combined_fields scorer build

style(fts): reword two comments the typos hook rejects

test(fts): cover combined_fields nested in a compound query

refactor(fts): share the FTS score sort keys and tie comparator