joplin/packages/lib/import-enex.ts

import uuid from './uuid';
import BaseModel from './BaseModel';
import Note from './models/Note';
import Tag from './models/Tag';
import Resource from './models/Resource';
import Setting from './models/Setting';
import time from './time';
import shim from './shim';
import { NoteEntity } from './services/database/types';
import { enexXmlToMd } from './import-enex-md-gen';
import { MarkupToHtml } from '@joplin/renderer';
const moment = require('moment');
const { wrapError } = require('./errorUtils');
const { enexXmlToHtml } = require('./import-enex-html-gen.js');
const Levenshtein = require('levenshtein');
const md5 = require('md5');
const { Base64Decode } = require('base64-stream');
const md5File = require('md5-file');
const { mime } = require('./mime-utils');

// const Promise = require('promise');
const fs = require('fs-extra');

function dateToTimestamp(s: string, defaultValue: number = null): number {
	// Most dates seem to be in this format
	let m = moment(s, 'YYYYMMDDTHHmmssZ');

	// But sometimes they might be in this format eg. 20180306T91108 AMZ
	// https://github.com/laurent22/joplin/issues/557
	if (!m.isValid()) m = moment(s, 'YYYYMMDDThmmss AZ');

	if (!m.isValid()) {
		if (defaultValue !== null) return defaultValue;
		throw new Error(`Invalid date: ${s}`);
	}

	return m.toDate().getTime();
}

function extractRecognitionObjId(recognitionXml: string) {
	const r = recognitionXml.match(/objID="(.*?)"/);
	return r && r.length >= 2 ? r[1] : null;
}

async function decodeBase64File(sourceFilePath: string, destFilePath: string) {
	// When something goes wrong with streams you can get an error "EBADF, Bad file descriptor"
	// with no strack trace to tell where the error happened.

	// Also note that this code is not great because there's a source and a destination stream
	// and while one stream might end, the other might throw an error or vice-versa. However
	// we can only throw one error from a promise. So before one stream
	// could end with resolve(), then another stream would get an error and call reject(), which
	// would be ignored. I don't think it's happening anymore, but something to keep in mind
	// anyway.

	return new Promise(function(resolve, reject) {
		// Note: we manually handle closing the file so that we can
		// force flusing it before close. This is needed because
		// "end" might be called before the file has been flushed
		// to disk, thus resulting in the calling code to find a
		// file with size 0.

		const destFile = fs.openSync(destFilePath, 'w');
		const sourceStream = fs.createReadStream(sourceFilePath);
		const destStream = fs.createWriteStream(destFile, {
			fd: destFile,
			autoClose: false,
		});
		sourceStream.pipe(new Base64Decode()).pipe(destStream);

		// We wait for the destination stream "finish" event, not the source stream "end" event
		// because even if the source has finished sending data, the destination might not have
		// finished receiving it and writing it to disk.
		destStream.on('finish', () => {
			fs.fdatasyncSync(destFile);
			fs.closeSync(destFile);
			resolve(null);
		});

		sourceStream.on('error', (error: any) => reject(error));
		destStream.on('error', (error: any) => reject(error));
	});
}

async function md5FileAsync(filePath: string): Promise<string> {
	return new Promise((resolve, reject) => {
		md5File(filePath, (error: any, hash: string) => {
			if (error) {
				reject(error);
				return;
			}

			resolve(hash);
		});
	});
}

function removeUndefinedProperties(note: NoteEntity) {
	const output: any = {};
	for (const n in note) {
		if (!note.hasOwnProperty(n)) continue;
		const v = (note as any)[n];
		if (v === undefined || v === null) continue;
		output[n] = v;
	}
	return output;
}

function levenshteinPercent(s1: string, s2: string) {
	const l = new Levenshtein(s1, s2);
	if (!s1.length || !s2.length) return 1;
	return Math.abs(l.distance / s1.length);
}

async function fuzzyMatch(note: ExtractedNote) {
	if (note.created_time < time.unixMs() - 1000 * 60 * 60 * 24 * 360) {
		const notes = await Note.modelSelectAll('SELECT * FROM notes WHERE is_conflict = 0 AND created_time = ? AND title = ?', [note.created_time, note.title]);
		return notes.length !== 1 ? null : notes[0];
	}

	const notes = await Note.modelSelectAll('SELECT * FROM notes WHERE is_conflict = 0 AND created_time = ?', [note.created_time]);
	if (notes.length === 0) return null;
	if (notes.length === 1) return notes[0];

	let lowestL = 1;
	let lowestN = null;
	for (let i = 0; i < notes.length; i++) {
		const n = notes[i];
		const l = levenshteinPercent(note.title, n.title);
		if (l < lowestL) {
			lowestL = l;
			lowestN = n;
		}
	}

	if (lowestN && lowestL < 0.2) return lowestN;

	return null;
}

interface ExtractedResource {
	hasData?: boolean;
	id?: string;
	size?: number;
	dataFilePath?: string;
	dataEncoding?: string;
	data?: string;
	filename?: string;
	sourceUrl?: string;
	mime?: string;
	title?: string;
}

interface ExtractedNote extends NoteEntity {
	resources?: ExtractedResource[];
	tags?: string[];
	title?: string;
	bodyXml?: string;
	// is_todo?: boolean;
}

// At this point we have the resource has it's been parsed from the XML, but additional
// processing needs to be done to get the final resource file, its size, MD5, etc.
async function processNoteResource(resource: ExtractedResource) {
	if (!resource.hasData) {
		// Some resources have no data, go figure, so we need a special case for this.
		resource.id = md5(Date.now() + Math.random());
		resource.size = 0;
		resource.dataFilePath = `${Setting.value('tempDir')}/${resource.id}.empty`;
		await fs.writeFile(resource.dataFilePath, '');
	} else {
		if (resource.dataEncoding == 'base64') {
			const decodedFilePath = `${resource.dataFilePath}.decoded`;
			await decodeBase64File(resource.dataFilePath, decodedFilePath);
			resource.dataFilePath = decodedFilePath;
		} else if (resource.dataEncoding) {
			throw new Error(`Cannot decode resource with encoding: ${resource.dataEncoding}`);
		}

		const stats = fs.statSync(resource.dataFilePath);
		resource.size = stats.size;

		if (!resource.id) {
			// If no resource ID is present, the resource ID is actually the MD5 of the data.
			// This ID will match the "hash" attribute of the corresponding <en-media> tag.
			// resourceId = md5(decodedData);
			resource.id = await md5FileAsync(resource.dataFilePath);
		}

		if (!resource.id || !resource.size) {
			const debugTemp = Object.assign({}, resource);
			debugTemp.data = debugTemp.data ? `${debugTemp.data.substr(0, 32)}...` : debugTemp.data;
			throw new Error(`This resource was not added because it has no ID or no content: ${JSON.stringify(debugTemp)}`);
		}
	}

	return resource;
}

async function saveNoteResources(note: ExtractedNote) {
	let resourcesCreated = 0;
	for (let i = 0; i < note.resources.length; i++) {
		const resource = note.resources[i];

		const toSave = Object.assign({}, resource);
		delete toSave.dataFilePath;
		delete toSave.dataEncoding;
		delete toSave.hasData;

		// The same resource sometimes appear twice in the same enex (exact same ID and file).
		// In that case, just skip it - it means two different notes might be linked to the
		// same resource.
		const existingResource = await Resource.load(toSave.id);
		if (existingResource) continue;

		await fs.move(resource.dataFilePath, Resource.fullPath(toSave), { overwrite: true });
		await Resource.save(toSave, { isNew: true });
		resourcesCreated++;
	}
	return resourcesCreated;
}

async function saveNoteTags(note: ExtractedNote) {
	let notesTagged = 0;
	for (let i = 0; i < note.tags.length; i++) {
		const tagTitle = note.tags[i];

		let tag = await Tag.loadByTitle(tagTitle);
		if (!tag) tag = await Tag.save({ title: tagTitle });

		await Tag.addNote(tag.id, note.id);

		notesTagged++;
	}
	return notesTagged;
}

interface ImportOptions {
	fuzzyMatching?: boolean;
	onProgress?: Function;
	onError?: Function;
	outputFormat?: string;
}

async function saveNoteToStorage(note: ExtractedNote, importOptions: ImportOptions) {
	importOptions = Object.assign({}, {
		fuzzyMatching: false,
	}, importOptions);

	note = Note.filter(note as any);

	const existingNote = importOptions.fuzzyMatching ? await fuzzyMatch(note) : null;

	const result = {
		noteCreated: false,
		noteUpdated: false,
		noteSkipped: false,
		resourcesCreated: 0,
		notesTagged: 0,
	};

	const resourcesCreated = await saveNoteResources(note);
	result.resourcesCreated += resourcesCreated;

	const notesTagged = await saveNoteTags(note);
	result.notesTagged += notesTagged;

	if (existingNote) {
		const diff = BaseModel.diffObjects(existingNote, note);
		delete diff.tags;
		delete diff.resources;
		delete diff.id;

		if (!Object.getOwnPropertyNames(diff).length) {
			result.noteSkipped = true;
			return result;
		}

		diff.id = existingNote.id;
		diff.type_ = existingNote.type_;
		await Note.save(diff, { autoTimestamp: false });
		result.noteUpdated = true;
	} else {
		await Note.save(note, {
			isNew: true,
			autoTimestamp: false,
		});
		result.noteCreated = true;
	}

	return result;
}

interface Node {
	name: string;
	attributes: Record<string, any>;
}

interface NoteResourceRecognition {
	objID?: string;
}

const preProcessFile = async (filePath: string): Promise<string> => {
	const content: string = await shim.fsDriver().readFile(filePath, 'utf8');

	// The note content in an ENEX file is wrapped in a CDATA block so it means
	// that any "]]>" inside the note must be somehow escaped, or else the CDATA
	// block would be closed at the wrong point.
	//
	// The problem is that Evernote appears to encode "]]>" as "]]<![CDATA[>]]>"
	// instead of the more sensible "]]&gt;", or perhaps they have nothing in
	// place to properly escape data imported from their web clipper. In any
	// case it results in invalid XML that Evernote cannot even import back.
	//
	// Handling that invalid XML with SAX would also be very tricky, so instead
	// we add a pre-processing step that converts this tags to just "&gt;". It
	// should be safe to do so because such content can only be within the body
	// of a note - and ">" or "&gt;" is equivalent.
	//
	// Ref: https://discourse.joplinapp.org/t/20470/4
	const newContent = content.replace(/<!\[CDATA\[>\]\]>/g, '&gt;');
	if (content === newContent) return filePath;
	const newFilePath = `${Setting.value('tempDir')}/${md5(Date.now() + Math.random())}.enex`;
	await shim.fsDriver().writeFile(newFilePath, newContent, 'utf8');
	return newFilePath;
};

export default async function importEnex(parentFolderId: string, filePath: string, importOptions: ImportOptions = null) {
	if (!importOptions) importOptions = {};
	if (!('fuzzyMatching' in importOptions)) importOptions.fuzzyMatching = false;
	if (!('onProgress' in importOptions)) importOptions.onProgress = function() {};
	if (!('onError' in importOptions)) importOptions.onError = function() {};

	function handleSaxStreamEvent(fn: Function) {
		return function(...args: any[]) {
			// Pass the parser to the wrapped function for debugging purposes
			if (this._parser) (fn as any)._parser = this._parser;

			try {
				fn.call(this, ...args);
			} catch (error) {
				if (importOptions.onError) {
					importOptions.onError(error);
				} else {
					console.error(error);
				}
			}
		};
	}

	const fileToProcess = await preProcessFile(filePath);
	const needToDeleteFileToProcess = fileToProcess !== filePath;

	return new Promise((resolve) => {
		const progressState = {
			loaded: 0,
			created: 0,
			updated: 0,
			skipped: 0,
			resourcesCreated: 0,
			notesTagged: 0,
		};

		const stream = fs.createReadStream(fileToProcess);

		const options = {};
		const strict = true;
		const saxStream = require('@joplin/fork-sax').createStream(strict, options);

		const nodes: Node[] = []; // LIFO list of nodes so that we know in which node we are in the onText event
		let note: ExtractedNote = null;
		let noteAttributes: Record<string, any> = null;
		let noteResource: ExtractedResource = null;
		let noteResourceAttributes: Record<string, any> = null;
		let noteResourceRecognition: NoteResourceRecognition = null;
		const notes: ExtractedNote[] = [];
		let processingNotes = false;

		const createErrorWithNoteTitle = (fnThis: any, error: any) => {
			const line = [];

			const parser = fnThis ? fnThis._parser : null;
			if (parser) {
				line.push(`Line ${parser.line}:${parser.column}`);
			}

			if (note && note.title) {
				line.push(`"${note.title}"`);
			}

			line.push(error.message);

			error.message = line.join(': ');

			return error;
		};

		stream.on('error', function(error: any) {
			importOptions.onError(createErrorWithNoteTitle(this, error));
		});

		function currentNodeName() {
			if (!nodes.length) return null;
			return nodes[nodes.length - 1].name;
		}

		function currentNodeAttributes() {
			if (!nodes.length) return {};
			return nodes[nodes.length - 1].attributes;
		}

		async function processNotes() {
			if (processingNotes) return false;

			processingNotes = true;
			stream.pause();

			while (notes.length) {
				const note = notes.shift();

				try {
					for (let i = 0; i < note.resources.length; i++) {
						let resource = note.resources[i];

						try {
							resource = await processNoteResource(resource);
						} catch (error) {
							importOptions.onError(createErrorWithNoteTitle(null, error));
							continue;
						}

						note.resources[i] = resource;
					}

					const body = importOptions.outputFormat === 'html' ?
						await enexXmlToHtml(note.bodyXml, note.resources) :
						await enexXmlToMd(note.bodyXml, note.resources);
					delete note.bodyXml;

					note.markup_language = importOptions.outputFormat === 'html' ?
						MarkupToHtml.MARKUP_LANGUAGE_HTML :
						MarkupToHtml.MARKUP_LANGUAGE_MARKDOWN;

					// console.info('*************************************************************************');
					// console.info(body);
					// console.info('*************************************************************************');

					note.id = uuid.create();
					note.parent_id = parentFolderId;
					note.body = body;

					// If the created timestamp was invalid, it would be
					// set to zero, so set it to the current date here
					if (!note.created_time) note.created_time = Date.now();

					// Notes in enex files always have a created timestamp
					// but not always an updated timestamp (it the note has
					// never been modified). For sync we require an
					// updated_time property, so set it to create_time in
					// that case
					if (!note.updated_time) note.updated_time = note.created_time;

					const result = await saveNoteToStorage(note, importOptions);

					if (result.noteUpdated) {
						progressState.updated++;
					} else if (result.noteCreated) {
						progressState.created++;
					} else if (result.noteSkipped) {
						progressState.skipped++;
					}
					progressState.resourcesCreated += result.resourcesCreated;
					progressState.notesTagged += result.notesTagged;
					importOptions.onProgress(progressState);
				} catch (error) {
					const newError = wrapError(`Error on note "${note.title}"`, error);
					importOptions.onError(createErrorWithNoteTitle(null, newError));
				}
			}

			stream.resume();
			processingNotes = false;
			return true;
		}

		saxStream.on('error', function(error: any) {
			importOptions.onError(createErrorWithNoteTitle(this, error));
		});

		saxStream.on('text', handleSaxStreamEvent(function(text: string) {
			const n = currentNodeName();

			if (noteAttributes) {
				noteAttributes[n] = text;
			} else if (noteResourceAttributes) {
				noteResourceAttributes[n] = text;
			} else if (noteResource) {
				if (n == 'data') {
					if (!noteResource.dataEncoding) {
						const attr = currentNodeAttributes();
						noteResource.dataEncoding = attr.encoding;
					}

					if (!noteResource.dataFilePath) {
						noteResource.dataFilePath = `${Setting.value('tempDir')}/${md5(Date.now() + Math.random())}.base64`;
					}

					noteResource.hasData = true;

					fs.appendFileSync(noteResource.dataFilePath, text);
				} else {
					if (!(n in noteResource)) (noteResource as any)[n] = '';
					(noteResource as any)[n] += text;
				}
			} else if (note) {
				if (n == 'title') {
					note.title = text;
				} else if (n == 'created') {
					note.created_time = dateToTimestamp(text, 0);
				} else if (n == 'updated') {
					note.updated_time = dateToTimestamp(text, 0);
				} else if (n == 'tag') {
					note.tags.push(text);
				} else if (n == 'note') {
					// Ignore - white space between the opening tag <note> and the first sub-tag
				} else if (n == 'content') {
					// Ignore - white space between the opening tag <content> and the <![CDATA[< block where the content actually is
				} else {
					console.warn(createErrorWithNoteTitle(this, new Error(`Unsupported note tag: ${n}`)));
				}
			}
		}));

		saxStream.on('opentag', handleSaxStreamEvent(function(node: Node) {
			const n = node.name.toLowerCase();
			nodes.push(node);

			if (n == 'note') {
				note = {
					resources: [],
					tags: [],
					bodyXml: '',
				};
			} else if (n == 'resource-attributes') {
				noteResourceAttributes = {};
			} else if (n == 'recognition') {
				if (noteResource) noteResourceRecognition = {};
			} else if (n == 'note-attributes') {
				noteAttributes = {};
			} else if (n == 'resource') {
				noteResource = {
					hasData: false,
				};
			}
		}));

		saxStream.on('cdata', handleSaxStreamEvent(function(data: any) {
			const n = currentNodeName();

			if (noteResourceRecognition) {
				noteResourceRecognition.objID = extractRecognitionObjId(data);
			} else if (note) {
				if (n == 'content') {
					note.bodyXml += data;
				}
			}
		}));

		saxStream.on('closetag', handleSaxStreamEvent(function(n: string) {
			nodes.pop();

			if (n == 'note') {
				note = removeUndefinedProperties(note);

				progressState.loaded++;
				importOptions.onProgress(progressState);

				notes.push(note);

				if (notes.length >= 10) {
					processNotes().catch(error => {
						importOptions.onError(createErrorWithNoteTitle(this, error));
					});
				}
				note = null;
			} else if (n == 'recognition' && noteResource) {
				noteResource.id = noteResourceRecognition.objID;
				noteResourceRecognition = null;
			} else if (n == 'resource-attributes') {
				noteResource.filename = noteResourceAttributes['file-name'];
				if (noteResourceAttributes['source-url']) noteResource.sourceUrl = noteResourceAttributes['source-url'];
				noteResourceAttributes = null;
			} else if (n == 'note-attributes') {
				note.latitude = noteAttributes.latitude;
				note.longitude = noteAttributes.longitude;
				note.altitude = noteAttributes.altitude;
				note.author = noteAttributes.author ? noteAttributes.author.trim() : '';
				note.is_todo = noteAttributes['reminder-order'] !== '0' && !!noteAttributes['reminder-order'] as any;
				note.todo_due = dateToTimestamp(noteAttributes['reminder-time'], 0);
				note.todo_completed = dateToTimestamp(noteAttributes['reminder-done-time'], 0);
				note.order = dateToTimestamp(noteAttributes['reminder-order'], 0);
				note.source = noteAttributes.source ? `evernote.${noteAttributes.source.trim()}` : 'evernote';
				note.source_url = noteAttributes['source-url'] ? noteAttributes['source-url'].trim() : '';

				noteAttributes = null;
			} else if (n == 'resource') {
				let mimeType = noteResource.mime ? noteResource.mime.trim() : '';

				// Evernote sometimes gives an invalid or generic
				// "application/octet-stream" mime type for files that could
				// have a valid mime type, based on the extension. So in
				// general, we trust the filename more than the provided mime
				// type.
				// https://discourse.joplinapp.org/t/importing-a-note-with-a-zip-file/12123
				if (noteResource.filename) {
					const mimeTypeFromFile = mime.fromFilename(noteResource.filename);
					if (mimeTypeFromFile && mimeTypeFromFile !== mimeType) {
						// Don't print statement by default because it would show up in test units
						// console.info(`Invalid mime type "${mimeType}" for resource "${noteResource.filename}". Using "${mimeTypeFromFile}" instead.`);
						mimeType = mimeTypeFromFile;
					}
				}

				note.resources.push({
					id: noteResource.id,
					dataFilePath: noteResource.dataFilePath,
					dataEncoding: noteResource.dataEncoding,
					mime: mimeType,
					title: noteResource.filename ? noteResource.filename.trim() : '',
					filename: noteResource.filename ? noteResource.filename.trim() : '',
					hasData: noteResource.hasData,
				});

				noteResource = null;
			}
		}));

		saxStream.on('end', handleSaxStreamEvent(function() {
			// Wait till there is no more notes to process.
			const iid = shim.setInterval(() => {
				void processNotes().then(allDone => {
					if (allDone) {
						shim.clearTimeout(iid);
						if (needToDeleteFileToProcess) void shim.fsDriver().remove(fileToProcess);
						resolve(null);
					}
				});
			}, 500);
		}));

		stream.pipe(saxStream);
	});
}