import { Database } from "jsr:@db/sqlite";
import {
  assert,
  assertEquals,
} from "https://deno.land/std@0.224.0/testing/asserts.ts";

/**
 * Proof of Concept: Multi-Vec Isolation (Gen 2)
 *
 * Demonstrates the concept of preventing semantic bleed by using physically
 * isolated SQLite vector databases instead of dumping all embeddings into a
 * single database. We use actual SQLite databases for this in Gen 2.
 */

// User-defined function for similarity
function cosineSimilarity(vecA: number[], vecB: number[]): number {
  let dotProduct = 0, normA = 0, normB = 0;
  for (let i = 0; i < vecA.length; i++) {
    dotProduct += vecA[i] * vecB[i];
    normA += vecA[i] ** 2;
    normB += vecB[i] ** 2;
  }
  if (normA === 0 || normB === 0) return 0;
  return dotProduct / (Math.sqrt(normA) * Math.sqrt(normB));
}

function initDb(name: string): Database {
  // Use independent named files in /tmp so they aren't the same memory db
  const db = new Database(`/tmp/${name}.db`);
  db.function("vec_distance", (a: string, b: string) => cosineSimilarity(JSON.parse(a), JSON.parse(b)));
  db.exec("CREATE TABLE IF NOT EXISTS embeddings (id TEXT, text TEXT, vector TEXT)");
  db.exec("DELETE FROM embeddings"); // clear from previous runs
  return db;
}

if (import.meta.main) {
  console.log("Running Multi-Vec Isolation PoC (Gen 2) tests...");

  try {
    const docsDb = initDb("docs_graph");
    const telemetryDb = initDb("telemetry_graph");

    const insertDocs = docsDb.prepare("INSERT INTO embeddings VALUES (?, ?, ?)");
    insertDocs.run("docs-1", "High performance server scaling", JSON.stringify([0.9, 0.1, 0.2]));
    insertDocs.finalize();

    const insertTelemetry = telemetryDb.prepare("INSERT INTO embeddings VALUES (?, ?, ?)");
    insertTelemetry.run("telemetry-1", "Memory leak in main process", JSON.stringify([0.1, 0.9, 0.2]));
    insertTelemetry.finalize();

    // The user asks about "Performance and scaling"
    const queryVector = JSON.stringify([0.85, 0.15, 0.1]);

    const docsResults = docsDb.prepare("SELECT id, vec_distance(vector, ?) as score FROM embeddings ORDER BY score DESC LIMIT 1").get(queryVector) as { id: string, score: number };
    const telemetryResults = telemetryDb.prepare("SELECT id, vec_distance(vector, ?) as score FROM embeddings ORDER BY score DESC LIMIT 1").get(queryVector) as { id: string, score: number };

    console.log("Docs graph match:", docsResults?.id, docsResults?.score);
    console.log("Telemetry graph match:", telemetryResults?.id, telemetryResults?.score);

    assert(docsResults.score > 0.9, "Should find a high match in docs");
    assert(telemetryResults.score < docsResults.score, "Telemetry should be less relevant for this query");

    docsDb.close();
    telemetryDb.close();

    console.log("✅ Multi-Vec Isolation PoC (Gen 2) successful: Isolated graphs prevented cross-contamination.");
  } catch (err) {
    console.error("❌ Multi-Vec Isolation PoC (Gen 2) failed:", err);
    Deno.exit(1);
  }
}