Module search 

Exact best-first top-k search over a Trie with weighted edit costs.

Each queued trie node carries one banded row of the edit-distance matrix. A row cell k at trie depth j stands for query prefix length i = j + k - W, where W = budget / c_indel_min is the band half-width. A node is queued with a lower bound on the cost of every term below it, so the first terminal popped is always the best remaining one.

The budget, the band and the candidates always use the weighted costs of CostModel. The order of the results is set by Ranking: by default (Ranking::Coarse) terms are ranked by the weighted cost rounded up to whole units of 16 (whole_units), then by higher weight, then by lower term id, so the weight decides between terms whose costs round to the same number of units. This is not a count of edits: the x1.5 factor on the first byte makes an ordinary (non-neighbouring-key) edit there cost 24, i.e. two units, while a neighbouring-key substitution there costs 12 (one unit) and a transposition 18 (two units); two cheap edits (8 + 8) count as one. Ranking::Exact ranks by the exact weighted cost instead, then weight, then id. In both modes Hit::cost is the exact weighted cost.

Searcher::search_prefix is the autocomplete variant: a term costs what its best prefix costs. See docs/design/prefix-mode.md.

Queries and terms are compared per Unicode scalar value (a symbol), not per UTF-8 byte; for ASCII the two are the same. See docs/design/unicode.md.

Structs

Hit

One result.

Output

Hits in ranking order plus work counters.

SearchConfig

Search parameters.

Searcher

A reusable search context. It keeps its priority queue and scratch buffers across queries; the returned Output is still allocated per query.

Stats

Work counters for one search.

Enums

Ranking

How results are ordered.

SearchError

Why a search was rejected.

Constants

MAX_QUERY_LEN

Longest accepted query, in symbols (code points; bytes for ASCII).