Repository navigation
fix(score): sum scores across indexed columns - #18
Merged
Merged
Conversation
Scoring bound only the first searched column that had a tin index, so a search over several indexed columns scored one of them and the result depended on the order of the predicates. Group the searches by the tin index each binds to, score every group, and sum the groups that match the row, as tin does. max_score takes the highest summed score among documents that match every group the quals require. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
tin cannot scan a disjunction in which a search without a tin index is an alternative to an indexed one, so it refuses to score that query and names the unindexed search. Do the same, require a group for max_score whenever the quals imply a match of it, and keep an empty sum at +0. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
Pin the scores tin returns for the order-swapped searches, the summed scores and max_score of rows matching several columns, and the shapes with a partial index that the quals do not imply. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
Scoring and highlighting treated any operator named ==> with two arguments as a search, so a ==> from another schema or for other types bound to them. A ==>(text, int) put an int query into the text[] of searches and crashed the backend. Compare the operator OID with Lead's pg_catalog.==>(text, text) instead, looked up on each use. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
eeeebbbbrrrr
added this pull request to stack #20
October 6, 2026 18:25
A row that only a non-search qual admits, such as the id = 8 of body ==> 'gems' OR id = 8, scored 0. tin returns NULL for it, so return NULL when none of the row's indexed columns matches its searches. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
tin scores a relation with one policy. max_score takes the policy of the relation's tin.score calls, sharing their dense_ratio, term_add, and term_replace, and otherwise reports the full_score maximum, including when it is the only scoring call. Under the tin.score policy tin computes no maximum for several indexes or a disjunction with other quals, and returns NULL. It refuses tin.score beside tin.full_score on one relation, and tin.score calls whose scan arguments differ. Do the same. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012mKGtrg7DEhVPgqYrJtHGg
eeeebbbbrrrr
approved these changes
Oct 6, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Before: when a query searched several tin-indexed columns,
tin.score,tin.full_score, andtin.max_scorescored only the first searched column.WHERE title ==> 'ruby' OR body ==> 'ruby'gave body-only matches a zero score, and swapping the predicates moved the zero to the title-only matches.ANDhad the same order dependence.After: each indexed column's searches score on their own, against that index's corpus, and a row's score is the sum over the columns whose searches it matches, as in tin. The issue's fixture scores 0.87138504 and 0.6931472 in either order.
max_scoreis the highest summed score among matching documents. A row that only a non-search qual admits scores NULL, as in tin.max_score's policy, and combinedscore/full_scorecalls, follow tin too.How this fixes #15
score_supportused to bind the first==>operand that had a tin index. It now groups the searches by index, andscore_boundtakes every group in one call. It sums the scores of the groups that match the row exactly, so the result does not depend on group order. Inmax_scoremode it makes one corpus pass over the summed scores. Partial-index eligibility is checked per column.Stacked on #17.
Fixes #15