// 1. Import transformers.js
import { pipeline, env } from './transformers.min.js';

// 2. Constants
env.allowLocalModels = true;
const MODEL_NAME = 'nomic-ai/nomic-embed-text-v1.5';
const EMBEDDING_DIM = 256;
const INDEX_PATH = 'data/quotes_index.bin';
const APPROX_MODEL_SIZE_MB = 180; // Actual observed size of nomic-embed-text-v1.5 model

const DB_NAME = 'QuoteSearchDB';
const DB_VERSION = 1;
const STORE_NAME = 'quoteIndex';

let model;
let indexData;
let isReady = false;

// IndexedDB Helper Functions
function openDB() {
    return new Promise((resolve, reject) => {
        const request = indexedDB.open(DB_NAME, DB_VERSION);

        request.onupgradeneeded = (event) => {
            const db = event.target.result;
            db.createObjectStore(STORE_NAME, { keyPath: 'id' });
        };

        request.onsuccess = (event) => {
            resolve(event.target.result);
        };

        request.onerror = (event) => {
            reject('IndexedDB error: ' + event.target.errorCode);
        };
    });
}

async function getFromDB(key) {
    const db = await openDB();
    return new Promise((resolve, reject) => {
        const transaction = db.transaction([STORE_NAME], 'readonly');
        const store = transaction.objectStore(STORE_NAME);
        const request = store.get(key);

        request.onsuccess = () => {
            resolve(request.result ? request.result.value : null);
        };

        request.onerror = () => {
            reject('Error getting data from DB');
        };
    });
}

async function putInDB(key, value) {
    const db = await openDB();
    return new Promise((resolve, reject) => {
        const transaction = db.transaction([STORE_NAME], 'readwrite');
        const store = transaction.objectStore(STORE_NAME);
        const request = store.put({ id: key, value: value });

        request.onsuccess = () => {
            resolve();
        };

        request.onerror = () => {
            reject('Error putting data in DB');
        };
    });
}

async function deleteFromDB(key) {
    const db = await openDB();
    return new Promise((resolve, reject) => {
        const transaction = db.transaction([STORE_NAME], 'readwrite');
        const store = transaction.objectStore(STORE_NAME);
        const request = store.delete(key);

        request.onsuccess = () => {
            resolve();
        };

        request.onerror = () => {
            reject('Error deleting data from DB');
        };
    });
}

// 3. Load Model and Index
async function loadModel() {
    try {
        // Inform the UI that the model will be downloaded/loaded
        self.postMessage({ type: 'loading', payload: 'Downloading model (this may take a while)...' });

        // Load the model with a progress callback
        model = await pipeline('feature-extraction', MODEL_NAME, {
            progress_callback: (progress) => {
                // Make it explicit that this progress refers to model download or model file operations
                const detailMessage = `Downloading model ${progress.status}: ${progress.file || ''} ${Math.floor(progress.progress || 0)}%`;
                    self.postMessage({ type: 'progress', payload: { ...progress, detail: detailMessage } });
            }
        });
    } catch (error) {
        console.error('Error loading model:', error);
        self.postMessage({ type: 'error', payload: error.message });
        throw error; // Re-throw to prevent further execution if model fails to load
    }
}

async function loadIndex() {
    try {
    self.postMessage({ type: 'loading', payload: 'Checking for cached index file...' });
        const cachedIndex = await getFromDB('quoteIndexData');

        if (cachedIndex) {
            indexData = cachedIndex;
            self.postMessage({ type: 'loading', payload: 'Index loaded from cache.' });
            } else {
            // Inform UI that the index file will be downloaded
            self.postMessage({ type: 'loading', payload: 'Downloading index file (this may take a while)...' });

            // Fetch and parse the index file with progress reporting
            const response = await fetch(INDEX_PATH);
            const contentLength = response.headers.get('Content-Length');
            const total = parseInt(contentLength, 10);
            let loaded = 0;

            const reader = response.body.getReader();
            const chunks = [];

            while (true) {
                const { done, value } = await reader.read();
                if (done) {
                    break;
                }
                chunks.push(value);
                loaded += value.length;

                const progress = {
                    status: 'Downloading Index',
                    progress: (loaded / total) * 100,
                    file: INDEX_PATH,
                    detail: `Downloading index file: ${Math.floor((loaded / total) * 100)}% (${(loaded / (1024 * 1024)).toFixed(2)}MB / ${(total / (1024 * 1024)).toFixed(2)}MB)`
                };
                self.postMessage({ type: 'progress', payload: progress });
            }

            const buffer = await new Response(new Blob(chunks)).arrayBuffer();

            // Parse the binary file
            let offset = 0;
            const numQuotes = new Uint32Array(buffer.slice(offset, offset + 4))[0];
            offset += 4;
            const embeddingDim = new Uint16Array(buffer.slice(offset, offset + 2))[0];
            offset += 2;
            const scale = new Float32Array(buffer.slice(offset, offset + 4))[0];
            offset += 4;
            const metadataSize = new Uint32Array(buffer.slice(offset, offset + 4))[0];
            offset += 4;

            // Read metadata format flag (1 byte) first, then metadata bytes
            let metadata_format = 0; // 0 = uncompressed JSON (legacy), 1 = gzip-compressed JSON
            if (offset + 1 <= buffer.byteLength) {
                metadata_format = new Uint8Array(buffer.slice(offset, offset + 1))[0];
                offset += 1;
            }

            let metadataBytes = buffer.slice(offset, offset + metadataSize);
            offset += metadataSize;

            async function decodeMetadata(bytes, format) {
                const decoder = new TextDecoder('utf-8');
                if (format === 0) {
                    return JSON.parse(decoder.decode(bytes));
                } else if (format === 1) {
                    // Try using native DecompressionStream if available
                    if (typeof DecompressionStream !== 'undefined') {
                        const ds = new DecompressionStream('gzip');
                        const decompressed = await new Response(new Blob([bytes]).stream().pipeThrough(ds)).arrayBuffer();
                        return JSON.parse(decoder.decode(decompressed));
                    } else {
                        // Fallback: DecompressionStream is not available. To support gzip in
                        // older browsers, add a browser-ready pako build to the repo and
                        // implement the decompression here. For now, fail with a clear error.
                        throw new Error('Gzip decompression not available: add DecompressionStream support or include pako in the worker.');
                    }
                } else {
                    throw new Error('Unknown metadata format: ' + format);
                }
            }

            const metadata = await decodeMetadata(metadataBytes, metadata_format);

            const quantizedEmbeddings = new Int8Array(buffer.slice(offset));

            // De-quantize embeddings (processing step) with progress reporting
            const embeddings = new Float32Array(quantizedEmbeddings.length);
            const totalEmbeddings = quantizedEmbeddings.length;
            const updateInterval = Math.floor(totalEmbeddings / 100); // Update every 1%

            for (let i = 0; i < totalEmbeddings; i++) {
                embeddings[i] = quantizedEmbeddings[i] / scale;
                if (updateInterval > 0 && i % updateInterval === 0) {
                    // This is processing (de-quantization), not download
                    const progress = { 
                        status: 'Processing index (de-quantizing)',
                        progress: (i / totalEmbeddings) * 100,
                        file: INDEX_PATH,
                        detail: `Processing index: ${Math.floor((i / totalEmbeddings) * 100)}%`
                    };
                    self.postMessage({ type: 'progress', payload: progress });
                }
            }
            indexData = {
                metadata,
                embeddings: reshape(embeddings, [numQuotes, embeddingDim]),
                embeddingsByteLength: quantizedEmbeddings.byteLength // Store byteLength
            };
            await putInDB('quoteIndexData', indexData); // Store in IndexedDB
        }
    } catch (error) {
        console.error('Error loading index:', error);
        self.postMessage({ type: 'error', payload: error.message });
        throw error;
    }
}

async function load() {
    try {
        await loadIndex();
        isReady = true;
        self.postMessage({ type: 'ready' });
    } catch (error) {
        console.error('Error in worker load function:', error);
        self.postMessage({ type: 'error', payload: error.message });
    }
}

// 4. Listen for Messages
self.onmessage = async (event) => {
    const { type, payload } = event.data;

    if (type === 'search') {
        if (!indexData) {
            // Be explicit: index may be downloaded before search
            self.postMessage({ type: 'loading', payload: 'Downloading index before running your search...' });
            await loadIndex();
        }
        if (!model) {
            self.postMessage({ type: 'loading', payload: 'Downloading model before running your search...' });
            await loadModel();
        }
        self.postMessage({ type: 'loading', payload: 'Searching...' });
        const results = await search(payload);
        self.postMessage({ type: 'results', payload: results });
    } else if (type === 'deleteData') {
        self.postMessage({ type: 'loading', payload: 'Deleting cached data...' });
        // Run cleanup with a timeout so UI doesn't get stuck if some API call blocks or takes too long
        const cleanup = async () => {
            // 1) Delete the specific object from the DB if present
            try {
                await deleteFromDB('quoteIndexData');
            } catch (e) {
                // ignore individual delete error and proceed to full DB deletion
            }

            // 2) Delete the entire IndexedDB database to ensure all stored artifacts are removed
            await new Promise((resolve, reject) => {
                const deleteRequest = indexedDB.deleteDatabase(DB_NAME);
                deleteRequest.onsuccess = () => resolve();
                deleteRequest.onerror = () => reject(deleteRequest.error || new Error('Failed to delete IndexedDB'));
                deleteRequest.onblocked = () => {
                    // If deletion is blocked, still try to continue with cache cleanup
                    resolve();
                };
            });

            // 3) Clear Cache Storage entries that may contain model or transformers resources
            try {
                if (typeof caches !== 'undefined' && caches.keys) {
                    const cacheNames = await caches.keys();
                    for (const cacheName of cacheNames) {
                        // Delete transformers related caches and any caches that contain the model name
                        if (cacheName.startsWith('transformers-cache') || cacheName.includes(MODEL_NAME.replace(/\//g, '-')) || cacheName.includes('nomic')) {
                            await caches.delete(cacheName);
                        }
                    }
                }
            } catch (e) {
                // Non-fatal
                console.warn('Cache cleanup error', e);
            }

            // 4) Clear localStorage keys commonly used by transformers.js or model caches (best-effort)
            try {
                if (typeof localStorage !== 'undefined') {
                    const keysToClear = [];
                    for (let i = 0; i < localStorage.length; i++) {
                        const key = localStorage.key(i);
                        if (!key) continue;
                        if (key.startsWith('transformers') || key.includes('transformers') || key.includes('nomic') || key.includes('hf_')) {
                            keysToClear.push(key);
                        }
                    }
                    for (const k of keysToClear) localStorage.removeItem(k);
                }
            } catch (e) {
                // localStorage may not be available in worker-like contexts; ignore
            }

            // 5) Clear in-memory references
            indexData = null;
            model = null;
            isReady = false;
        };

        // Timeout in ms
        const TIMEOUT_MS = 8000;

        try {
            await Promise.race([
                cleanup(),
                new Promise((_, reject) => setTimeout(() => reject(new Error('cleanup-timeout')), TIMEOUT_MS))
            ]);
            // If cleanup completed within timeout
            self.postMessage({ type: 'dataDeleted', payload: 'Cached data deleted successfully.' });
        } catch (error) {
            if (error && error.message === 'cleanup-timeout') {
                console.warn('deleteData: cleanup timed out');
                // Post a success-like message so UI doesn't stay stuck; note that some cleanup may still be pending
                self.postMessage({ type: 'dataDeleted', payload: 'Cached data deletion attempted (timed out). Some cleanup may remain.' });
            } else {
                console.error('deleteData error:', error);
                self.postMessage({ type: 'error', payload: 'Failed to delete cached data: ' + (error && error.message ? error.message : String(error)) });
            }
        }
    } else if (type === 'getIndexSize') {
        try {
            let totalSize = 0;
            let indexCached = false;
            let modelCached = false;

            // Check if index is cached
            const cachedIndex = await getFromDB('quoteIndexData');
            if (cachedIndex) {
                totalSize += JSON.stringify(cachedIndex.metadata).length + cachedIndex.embeddingsByteLength;
                indexCached = true;
            } else {
                const response = await fetch(INDEX_PATH, { method: 'HEAD' });
                const contentLength = response.headers.get('Content-Length');
                totalSize += parseInt(contentLength, 10);
            }

            // Add approximate model size
            totalSize += APPROX_MODEL_SIZE_MB * 1024 * 1024; // Convert MB to bytes
            // Heuristic: assume model is cached if `model` object is initialized
            if (model) {
                modelCached = true;
            }

            self.postMessage({ type: 'indexSize', payload: { size: totalSize, indexCached: indexCached, modelCached: modelCached } });
        } catch (error) {
            self.postMessage({ type: 'error', payload: 'Failed to get index size: ' + error.message });
        }
    }
};

// 5. Search Function
async function search(query) {
    if (!model || !indexData) {
        return [];
    }

    // Generate query embedding
    const queryEmbedding = await model("search_query: " + query, { pooling: 'mean', normalize: true });
    const truncatedQueryEmbedding = queryEmbedding.data.slice(0, EMBEDDING_DIM);

    // Calculate cosine similarities
    const similarities = [];
    for (let i = 0; i < indexData.embeddings.length; i++) {
        const similarity = cosineSimilarity(truncatedQueryEmbedding, indexData.embeddings[i]);
        similarities.push({ index: i, similarity });
    }

    // Sort by similarity
    similarities.sort((a, b) => b.similarity - a.similarity);

    // Get top 30 results
    const topResults = similarities.slice(0, 30).map(item => {
        return indexData.metadata[item.index];
    });

    return topResults;
}

// 6. Helper Functions
function cosineSimilarity(vecA, vecB) {
    let dotProduct = 0;
    let normA = 0;
    let normB = 0;
    for (let i = 0; i < vecA.length; i++) {
        dotProduct += vecA[i] * vecB[i];
        normA += vecA[i] * vecA[i];
        normB += vecB[i] * vecB[i];
    }
    return dotProduct / (Math.sqrt(normA) * Math.sqrt(normB));
}

function reshape(array, shape) {
    const reshaped = [];
    let offset = 0;
    for (let i = 0; i < shape[0]; i++) {
        reshaped.push(array.slice(offset, offset + shape[1]));
        offset += shape[1];
    }
    return reshaped;
}