/**
 * Code chunker — turns Methods/Types into embedding-ready text.
 *
 * Strategy:
 *   Tier A: Signature
 *   Tier B: Doc-Comment (if any)
 *   Tier C: Weighted identifier bag (top-100 by frequency, repeated as weight)
 *           plus alphanumeric string literals >5 chars from the body.
 *
 * The weighting embeds the heavy hitters (errorCount used 3x → appears 3x in
 * the text) so the embedding model picks them up as the dominant theme.
 */
export interface CodeChunkInput {
    name: string;
    signature: string;
    docComment?: string | null;
    bodyText?: string | null;
    /** Identifiers used in the method/type body, with usage counts. */
    identifiers?: Map<string, number>;
}
export interface ChunkOutput {
    /**
     * Backwards-compatible field: equals `embeddingText`.
     * Callers that don't care about the display/embedding split can keep using this.
     */
    text: string;
    /** Text fed to the embedding model (with weighted identifier repetition). */
    embeddingText: string;
    /** Human-readable text stored in source_text and shown in the viewer. */
    displayText: string;
    /** Hash of the embedding-relevant text — used for skip-on-no-change. */
    contentHash: string;
}
/** Public for testing. */
export declare function extractStringLiterals(body: string): string[];
/** Public for testing — returns the weighted, repetition-based identifier list (for the model). */
export declare function buildIdentifierBag(identifiers: Map<string, number>): string[];
/** Public for testing — returns "name (count), name (count), ..." for display. */
export declare function buildIdentifierBagDisplay(identifiers: Map<string, number>): string;
export declare function chunkCode(input: CodeChunkInput): ChunkOutput;
//# sourceMappingURL=chunker.d.ts.map