feat: loose parser mode

.changeset/twelve-hairs-marry.md

@@ -0,0 +1,5 @@
+---
+'svelte': minor
+---
+
+feat: provide loose parser mode

packages/svelte/src/compiler/index.js

@@ -76,7 +76,7 @@ export function compileModule(source, options) {
  *
  * @overload
  * @param {string} source
- * @param {{ filename?: string; modern: true }} options
+ * @param {{ filename?: string; modern: true; loose?: boolean }} options
  * @returns {AST.Root}
  */
 
@@ -88,7 +88,7 @@ export function compileModule(source, options) {
  *
  * @overload
  * @param {string} source
- * @param {{ filename?: string; modern?: false }} [options]
+ * @param {{ filename?: string; modern?: false; loose?: boolean }} [options]
  * @returns {Record<string, any>}
  */
 
@@ -99,15 +99,15 @@ export function compileModule(source, options) {
  * `modern` will become `true` by default in Svelte 6, and the option will be removed in Svelte 7.
  *
  * @param {string} source
- * @param {{ filename?: string; rootDir?: string; modern?: boolean }} [options]
+ * @param {{ filename?: string; rootDir?: string; modern?: boolean; loose?: boolean }} [options]
  * @returns {AST.Root | LegacyRoot}
  */
-export function parse(source, { filename, rootDir, modern } = {}) {
+export function parse(source, { filename, rootDir, modern, loose } = {}) {
 	source = remove_bom(source);
 	state.reset_warning_filter(() => false);
 	state.reset(source, { filename: filename ?? '(unknown)', rootDir });
 
-	const ast = _parse(source);
+	const ast = _parse(source, loose);
 	return to_public_ast(source, ast, modern);
 }
 

packages/svelte/src/compiler/phases/1-parse/index.js

@@ -22,6 +22,19 @@ export class Parser {
 	 */
 	template;
 
+	/**
+	 * @readonly
+	 * @type {string}
+	 */
+	template_untrimmed;
+
+	/**
+	 * Whether or not we're in loose parsing mode, in which
+	 * case we try to continue parsing as much as possible
+	 * @type {boolean}
+	 */
+	loose;
+
 	/** */
 	index = 0;
 
@@ -43,12 +56,17 @@ export class Parser {
 	/** @type {LastAutoClosedTag | undefined} */
 	last_auto_closed_tag;
 
-	/** @param {string} template */
-	constructor(template) {
+	/**
+	 * @param {string} template
+	 * @param {boolean} loose
+	 */
+	constructor(template, loose) {
 		if (typeof template !== 'string') {
 			throw new TypeError('Template must be a string');
 		}
 
+		this.loose = loose;
+		this.template_untrimmed = template;
 		this.template = template.trimEnd();
 
 		let match_lang;
@@ -88,7 +106,9 @@ export class Parser {
 		if (this.stack.length > 1) {
 			const current = this.current();
 
-			if (current.type === 'RegularElement') {
+			if (this.loose) {
+				current.end = this.template.length;
+			} else if (current.type === 'RegularElement') {
 				current.end = current.start + 1;
 				e.element_unclosed(current, current.name);
 			} else {
@@ -151,14 +171,15 @@ export class Parser {
 	/**
 	 * @param {string} str
 	 * @param {boolean} required
+	 * @param {boolean} required_in_loose
 	 */
-	eat(str, required = false) {
+	eat(str, required = false, required_in_loose = true) {
 		if (this.match(str)) {
 			this.index += str.length;
 			return true;
 		}
 
-		if (required) {
+		if (required && (!this.loose || required_in_loose)) {
 			e.expected_token(this.index, str);
 		}
 
@@ -233,6 +254,7 @@ export class Parser {
 	/** @param {RegExp} pattern */
 	read_until(pattern) {
 		if (this.index >= this.template.length) {
+			if (this.loose) return '';
 			e.unexpected_eof(this.template.length);
 		}
 
@@ -274,10 +296,11 @@ export class Parser {
 
 /**
  * @param {string} template
+ * @param {boolean} [loose]
  * @returns {AST.Root}
  */
-export function parse(template) {
-	const parser = new Parser(template);
+export function parse(template, loose = false) {
+	const parser = new Parser(template, loose);
 	return parser.root;
 }
 

packages/svelte/src/compiler/phases/1-parse/read/expression.js

@@ -3,6 +3,7 @@
 import { parse_expression_at } from '../acorn.js';
 import { regex_whitespace } from '../../patterns.js';
 import * as e from '../../../errors.js';
+import { find_matching_bracket } from '../utils/bracket.js';
 
 /**
  * @param {Parser} parser
@@ -39,6 +40,22 @@ export default function read_expression(parser) {
 
 		return /** @type {Expression} */ (node);
 	} catch (err) {
+		if (parser.loose) {
+			// Find the next } and treat it as the end of the expression
+			const end = find_matching_bracket(parser.template, parser.index, '{');
+			if (end) {
+				const start = parser.index;
+				parser.index = end;
+				// We don't know what the expression is and signal this by returning an empty identifier
+				return {
+					type: 'Identifier',
+					start,
+					end,
+					name: ''
+				};
+			}
+		}
+
 		parser.acorn_error(err);
 	}
 }

packages/svelte/src/compiler/phases/1-parse/state/element.js

@@ -9,10 +9,11 @@ import { decode_character_references } from '../utils/html.js';
 import * as e from '../../../errors.js';
 import * as w from '../../../warnings.js';
 import { create_fragment } from '../utils/create.js';
-import { create_attribute, create_expression_metadata } from '../../nodes.js';
+import { create_attribute, create_expression_metadata, is_element_node } from '../../nodes.js';
 import { get_attribute_expression, is_expression_attribute } from '../../../utils/ast.js';
 import { closing_tag_omitted } from '../../../../html-tree-validation.js';
 import { list } from '../../../utils/string.js';
+import { regex_whitespace } from '../../patterns.js';
 
 const regex_invalid_unquoted_attribute_value = /^(\/>|[\s"'=<>`])/;
 const regex_closing_textarea_tag = /^<\/textarea(\s[^>]*)?>/i;
@@ -80,7 +81,19 @@ export default function element(parser) {
 
 		// close any elements that don't have their own closing tags, e.g. <div><p></div>
 		while (/** @type {AST.RegularElement} */ (parent).name !== name) {
-			if (parent.type !== 'RegularElement') {
+			if (parser.loose) {
+				// If the previous element did interpret the next opening tag as an attribute, backtrack
+				if (is_element_node(parent)) {
+					const last = parent.attributes.at(-1);
+					if (last?.type === 'Attribute' && last.name === `<${name}`) {
+						parser.index = last.start;
+						parent.attributes.pop();
+						break;
+					}
+				}
+			}
+
+			if (parent.type !== 'RegularElement' && !parser.loose) {
 				if (parser.last_auto_closed_tag && parser.last_auto_closed_tag.tag === name) {
 					e.element_invalid_closing_tag_autoclosed(start, name, parser.last_auto_closed_tag.reason);
 				} else {
@@ -319,10 +332,41 @@ export default function element(parser) {
 	parser.append(element);
 
 	const self_closing = parser.eat('/') || is_void(name);
+	const closed = parser.eat('>', true, false);
+
+	// Loose parsing mode
+	if (!closed) {
+		// We may have eaten an opening `<` of the next element and treated is as an attribute...
+		const last = element.attributes.at(-1);
+		if (last?.type === 'Attribute' && last.name === '<') {
+			parser.index = last.start;
+			element.attributes.pop();
+		} else {
+			// ... or we may have eaten part of a following block ...
+			const prev_1 = parser.template[parser.index - 1];
+			const prev_2 = parser.template[parser.index - 2];
+			const current = parser.template[parser.index];
+			if (prev_2 === '{' && prev_1 === '/') {
+				parser.index -= 2;
+			} else if (prev_1 === '{' && (current === '#' || current === '@' || current === ':')) {
+				parser.index -= 1;
+			} else {
+				// ... or we're followed by whitespace, for example near the end of the template,
+				// which we want to take in so that language tools has more room to work with
+				parser.allow_whitespace();
+				if (parser.index === parser.template.length) {
+					while (
+						parser.index < parser.template_untrimmed.length &&
+						regex_whitespace.test(parser.template_untrimmed[parser.index])
+					) {
+						parser.index++;
+					}
+				}
+			}
+		}
+	}
 
-	parser.eat('>', true);
-
-	if (self_closing) {
+	if (self_closing || !closed) {
 		// don't push self-closing elements onto the stack
 		element.end = parser.index;
 	} else if (name === 'textarea') {
@@ -461,10 +505,22 @@ function read_attribute(parser) {
 			return spread;
 		} else {
 			const value_start = parser.index;
-			const name = parser.read_identifier();
+			let name = parser.read_identifier();
 
 			if (name === null) {
-				e.attribute_empty_shorthand(start);
+				if (
+					parser.loose &&
+					(parser.match('#') || parser.match('/') || parser.match('@') || parser.match(':'))
+				) {
+					// We're likely in an unclosed opening tag and did read part of a block.
+					// Return null to not crash the parser so it can continue with closing the tag.
+					return null;
+				} else if (parser.loose && parser.match('}')) {
+					// Likely in the middle of typing, just created the shorthand
+					name = '';
+				} else {
+					e.attribute_empty_shorthand(start);
+				}
 			}
 
 			parser.allow_whitespace();
@@ -740,5 +796,9 @@ function read_sequence(parser, done, location) {
 		}
 	}
 
-	e.unexpected_eof(parser.template.length);
+	if (parser.loose) {
+		return chunks;
+	} else {
+		e.unexpected_eof(parser.template.length);
+	}
 }