AnointedAutomation.Search 1.0.0

AnointedAutomation.Search

Jesus is King ✝️

In-process full-text search for .NET with no dependencies: Apache Lucene (Java) and tantivy (Rust) translated to C#. Every source file names the Lucene and/or tantivy file(s) it was ported from. Nothing else is in here: no product-specific ranking rules, just the search building blocks. See NOTICE.md for the attribution and the Lucene, KStem and tantivy license texts.

Installation

dotnet add package AnointedAutomation.Search

• Target framework: net10.0
• Dependencies: none

What is inside

Quick start

using AnointedAutomation.Search;
using AnointedAutomation.Search.Analysis;
using AnointedAutomation.Search.Index;
using AnointedAutomation.Search.Query;
using AnointedAutomation.Search.Scoring;

public sealed class Book
{
    public string Title { get; init; }
    public string Body { get; init; }
    public decimal? Price { get; init; }
}

TextAnalyzer analyzer = new TextAnalyzer(
    new StandardTokenizer(StandardTokenizer.DefaultMaxTokenLength),
    new ITokenFilter[] { new LowerCaseFilter(), new AsciiFoldingFilter(preserveOriginal: false), new KStemFilter() });

SearchSchema<Book> schema = new SearchSchema<Book>(analyzer);
schema.AddText("title", b => b.Title, new TextFieldOptions { Boost = 2m, Positions = true, K1 = Bm25.DefaultK1, B = Bm25.DefaultB });
schema.AddText("body", b => b.Body, new TextFieldOptions { Boost = 1m, Positions = true, K1 = Bm25.DefaultK1, B = Bm25.DefaultB });
schema.AddDecimal("price", b => b.Price);

SearchIndex<Book> index = new SearchIndex<Book>(schema);
index.Rebuild(books); // call again whenever the documents change; searches never block

SearchRequest<Book> request = new SearchRequest<Book>(limit: 20, offset: 0);
request.Clauses.Add(new TermClause("quick brown", Occur.Must, null));
request.Clauses.Add(new FuzzyClause("foxx", maxEdits: 1, transpositions: true, prefixLength: 1, Occur.Should, null));
request.Filters.Add(FastFieldFilter.Range("price", null, 50m));
SearchResult<Book> result = index.Search(request);

How a query is scored

• TermClause: every analyzed word must match; each word scores BM25 in its best field times the field boost (Lucene DisjunctionMaxQuery, tie breaker 0); the clause score is the sum over its words.
• PhraseClause: the words in order within the slop, in one positions field.
• FuzzyClause: every dictionary term within the edit budget, boosted 1 - edits / shorter length (Lucene FuzzyTermsEnum); a document keeps its best expansion.
• Occur.Must clauses are required and add up, Should clauses add where they match (at least one is required when there is no Must), MustNot excludes. Filters exclude without scoring.
• Results order by score, or by Sort fields then score; ties fall to document order.

Rules this code follows

• decimal for all scoring math; double appears only where .NET has no decimal logarithm.
• No fallback defaults: options are required, a missing snapshot throws, invalid input throws.
• Enums reserve 0 for Unknown.

License

Apache License 2.0 for the portions translated from Apache Lucene, MIT for the portions translated from tantivy; see NOTICE.md.