You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1785 lines
43 KiB
1785 lines
43 KiB
4 years ago
|
'use strict';
|
||
|
|
||
|
Object.defineProperty(exports, '__esModule', { value: true });
|
||
|
|
||
|
const JAVASCRIPT_OUTPUT_NAME = 'javascript';
|
||
|
const CSS_OUTPUT_NAME = 'css';
|
||
|
const TEMPLATE_OUTPUT_NAME = 'template';
|
||
|
|
||
|
// Tag names
|
||
|
const JAVASCRIPT_TAG = 'script';
|
||
|
const STYLE_TAG = 'style';
|
||
|
const TEXTAREA_TAG = 'textarea';
|
||
|
|
||
|
// Boolean attributes
|
||
|
const IS_RAW = 'isRaw';
|
||
|
const IS_SELF_CLOSING = 'isSelfClosing';
|
||
|
const IS_VOID = 'isVoid';
|
||
|
const IS_BOOLEAN = 'isBoolean';
|
||
|
const IS_CUSTOM = 'isCustom';
|
||
|
const IS_SPREAD = 'isSpread';
|
||
|
|
||
|
var c = /*#__PURE__*/Object.freeze({
|
||
|
__proto__: null,
|
||
|
JAVASCRIPT_OUTPUT_NAME: JAVASCRIPT_OUTPUT_NAME,
|
||
|
CSS_OUTPUT_NAME: CSS_OUTPUT_NAME,
|
||
|
TEMPLATE_OUTPUT_NAME: TEMPLATE_OUTPUT_NAME,
|
||
|
JAVASCRIPT_TAG: JAVASCRIPT_TAG,
|
||
|
STYLE_TAG: STYLE_TAG,
|
||
|
TEXTAREA_TAG: TEXTAREA_TAG,
|
||
|
IS_RAW: IS_RAW,
|
||
|
IS_SELF_CLOSING: IS_SELF_CLOSING,
|
||
|
IS_VOID: IS_VOID,
|
||
|
IS_BOOLEAN: IS_BOOLEAN,
|
||
|
IS_CUSTOM: IS_CUSTOM,
|
||
|
IS_SPREAD: IS_SPREAD
|
||
|
});
|
||
|
|
||
|
/**
|
||
|
* Not all the types are handled in this module.
|
||
|
*
|
||
|
* @enum {number}
|
||
|
* @readonly
|
||
|
*/
|
||
|
const TAG = 1; /* TAG */
|
||
|
const ATTR = 2; /* ATTR */
|
||
|
const TEXT = 3; /* TEXT */
|
||
|
const CDATA = 4; /* CDATA */
|
||
|
const COMMENT = 8; /* COMMENT */
|
||
|
const DOCUMENT = 9; /* DOCUMENT */
|
||
|
const DOCTYPE = 10; /* DOCTYPE */
|
||
|
const DOCUMENT_FRAGMENT = 11; /* DOCUMENT_FRAGMENT */
|
||
|
|
||
|
var types = /*#__PURE__*/Object.freeze({
|
||
|
__proto__: null,
|
||
|
TAG: TAG,
|
||
|
ATTR: ATTR,
|
||
|
TEXT: TEXT,
|
||
|
CDATA: CDATA,
|
||
|
COMMENT: COMMENT,
|
||
|
DOCUMENT: DOCUMENT,
|
||
|
DOCTYPE: DOCTYPE,
|
||
|
DOCUMENT_FRAGMENT: DOCUMENT_FRAGMENT
|
||
|
});
|
||
|
|
||
|
const rootTagNotFound = 'Root tag not found.';
|
||
|
const unclosedTemplateLiteral = 'Unclosed ES6 template literal.';
|
||
|
const unexpectedEndOfFile = 'Unexpected end of file.';
|
||
|
const unclosedComment = 'Unclosed comment.';
|
||
|
const unclosedNamedBlock = 'Unclosed "%1" block.';
|
||
|
const duplicatedNamedTag = 'Duplicate tag "<%1>".';
|
||
|
const unexpectedCharInExpression = 'Unexpected character %1.';
|
||
|
const unclosedExpression = 'Unclosed expression.';
|
||
|
|
||
|
/**
|
||
|
* Matches the start of valid tags names; used with the first 2 chars after the `'<'`.
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const TAG_2C = /^(?:\/[a-zA-Z]|[a-zA-Z][^\s>/]?)/;
|
||
|
/**
|
||
|
* Matches valid tags names AFTER the validation with `TAG_2C`.
|
||
|
* $1: tag name including any `'/'`, $2: non self-closing brace (`>`) w/o attributes.
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const TAG_NAME = /(\/?[^\s>/]+)\s*(>)?/g;
|
||
|
/**
|
||
|
* Matches an attribute name-value pair (both can be empty).
|
||
|
* $1: attribute name, $2: value including any quotes.
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const ATTR_START = /(\S[^>/=\s]*)(?:\s*=\s*([^>/])?)?/g;
|
||
|
|
||
|
/**
|
||
|
* Matches the spread operator
|
||
|
* it will be used for the spread attributes
|
||
|
* @type {RegExp}
|
||
|
*/
|
||
|
const SPREAD_OPERATOR = /\.\.\./;
|
||
|
/**
|
||
|
* Matches the closing tag of a `script` and `style` block.
|
||
|
* Used by parseText fo find the end of the block.
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const RE_SCRYLE = {
|
||
|
script: /<\/script\s*>/gi,
|
||
|
style: /<\/style\s*>/gi,
|
||
|
textarea: /<\/textarea\s*>/gi
|
||
|
};
|
||
|
|
||
|
// Do not touch text content inside this tags
|
||
|
const RAW_TAGS = /^\/?(?:pre|textarea)$/;
|
||
|
|
||
|
/**
|
||
|
* Add an item into a collection, if the collection is not an array
|
||
|
* we create one and add the item to it
|
||
|
* @param {Array} collection - target collection
|
||
|
* @param {*} item - item to add to the collection
|
||
|
* @returns {Array} array containing the new item added to it
|
||
|
*/
|
||
|
function addToCollection(collection = [], item) {
|
||
|
collection.push(item);
|
||
|
return collection
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Run RegExp.exec starting from a specific position
|
||
|
* @param {RegExp} re - regex
|
||
|
* @param {number} pos - last index position
|
||
|
* @param {string} string - regex target
|
||
|
* @returns {Array} regex result
|
||
|
*/
|
||
|
function execFromPos(re, pos, string) {
|
||
|
re.lastIndex = pos;
|
||
|
return re.exec(string)
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Escape special characters in a given string, in preparation to create a regex.
|
||
|
*
|
||
|
* @param {string} str - Raw string
|
||
|
* @returns {string} Escaped string.
|
||
|
*/
|
||
|
var escapeStr = (str) => str.replace(/(?=[-[\](){^*+?.$|\\])/g, '\\');
|
||
|
|
||
|
function formatError(data, message, pos) {
|
||
|
if (!pos) {
|
||
|
pos = data.length;
|
||
|
}
|
||
|
// count unix/mac/win eols
|
||
|
const line = (data.slice(0, pos).match(/\r\n?|\n/g) || '').length + 1;
|
||
|
let col = 0;
|
||
|
while (--pos >= 0 && !/[\r\n]/.test(data[pos])) {
|
||
|
++col;
|
||
|
}
|
||
|
return `[${line},${col}]: ${message}`
|
||
|
}
|
||
|
|
||
|
const $_ES6_BQ = '`';
|
||
|
|
||
|
/**
|
||
|
* Searches the next backquote that signals the end of the ES6 Template Literal
|
||
|
* or the "${" sequence that starts a JS expression, skipping any escaped
|
||
|
* character.
|
||
|
*
|
||
|
* @param {string} code - Whole code
|
||
|
* @param {number} pos - The start position of the template
|
||
|
* @param {string[]} stack - To save nested ES6 TL count
|
||
|
* @returns {number} The end of the string (-1 if not found)
|
||
|
*/
|
||
|
function skipES6TL(code, pos, stack) {
|
||
|
// we are in the char following the backquote (`),
|
||
|
// find the next unescaped backquote or the sequence "${"
|
||
|
const re = /[`$\\]/g;
|
||
|
let c;
|
||
|
while (re.lastIndex = pos, re.exec(code)) {
|
||
|
pos = re.lastIndex;
|
||
|
c = code[pos - 1];
|
||
|
if (c === '`') {
|
||
|
return pos
|
||
|
}
|
||
|
if (c === '$' && code[pos++] === '{') {
|
||
|
stack.push($_ES6_BQ, '}');
|
||
|
return pos
|
||
|
}
|
||
|
// else this is an escaped char
|
||
|
}
|
||
|
throw formatError(code, unclosedTemplateLiteral, pos)
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Custom error handler can be implemented replacing this method.
|
||
|
* The `state` object includes the buffer (`data`)
|
||
|
* The error position (`loc`) contains line (base 1) and col (base 0).
|
||
|
* @param {string} data - string containing the error
|
||
|
* @param {string} msg - Error message
|
||
|
* @param {number} pos - Position of the error
|
||
|
* @returns {undefined} throw an exception error
|
||
|
*/
|
||
|
function panic(data, msg, pos) {
|
||
|
const message = formatError(data, msg, pos);
|
||
|
throw new Error(message)
|
||
|
}
|
||
|
|
||
|
// forked from https://github.com/aMarCruz/skip-regex
|
||
|
|
||
|
// safe characters to precced a regex (including `=>`, `**`, and `...`)
|
||
|
const beforeReChars = '[{(,;:?=|&!^~>%*/';
|
||
|
const beforeReSign = `${beforeReChars}+-`;
|
||
|
|
||
|
// keyword that can preceed a regex (`in` is handled as special case)
|
||
|
const beforeReWords = [
|
||
|
'case',
|
||
|
'default',
|
||
|
'do',
|
||
|
'else',
|
||
|
'in',
|
||
|
'instanceof',
|
||
|
'prefix',
|
||
|
'return',
|
||
|
'typeof',
|
||
|
'void',
|
||
|
'yield'
|
||
|
];
|
||
|
|
||
|
// Last chars of all the beforeReWords elements to speed up the process.
|
||
|
const wordsEndChar = beforeReWords.reduce((s, w) => s + w.slice(-1), '');
|
||
|
|
||
|
// Matches literal regex from the start of the buffer.
|
||
|
// The buffer to search must not include line-endings.
|
||
|
const RE_LIT_REGEX = /^\/(?=[^*>/])[^[/\\]*(?:(?:\\.|\[(?:\\.|[^\]\\]*)*\])[^[\\/]*)*?\/[gimuy]*/;
|
||
|
|
||
|
// Valid characters for JavaScript variable names and literal numbers.
|
||
|
const RE_JS_VCHAR = /[$\w]/;
|
||
|
|
||
|
// Match dot characters that could be part of tricky regex
|
||
|
const RE_DOT_CHAR = /.*/g;
|
||
|
|
||
|
/**
|
||
|
* Searches the position of the previous non-blank character inside `code`,
|
||
|
* starting with `pos - 1`.
|
||
|
*
|
||
|
* @param {string} code - Buffer to search
|
||
|
* @param {number} pos - Starting position
|
||
|
* @returns {number} Position of the first non-blank character to the left.
|
||
|
* @private
|
||
|
*/
|
||
|
function _prev(code, pos) {
|
||
|
while (--pos >= 0 && /\s/.test(code[pos]));
|
||
|
return pos
|
||
|
}
|
||
|
|
||
|
|
||
|
|
||
|
/**
|
||
|
* Check if the character in the `start` position within `code` can be a regex
|
||
|
* and returns the position following this regex or `start+1` if this is not
|
||
|
* one.
|
||
|
*
|
||
|
* NOTE: Ensure `start` points to a slash (this is not checked).
|
||
|
*
|
||
|
* @function skipRegex
|
||
|
* @param {string} code - Buffer to test in
|
||
|
* @param {number} start - Position the first slash inside `code`
|
||
|
* @returns {number} Position of the char following the regex.
|
||
|
*
|
||
|
*/
|
||
|
/* istanbul ignore next */
|
||
|
function skipRegex(code, start) {
|
||
|
let pos = RE_DOT_CHAR.lastIndex = start++;
|
||
|
|
||
|
// `exec()` will extract from the slash to the end of the line
|
||
|
// and the chained `match()` will match the possible regex.
|
||
|
const match = (RE_DOT_CHAR.exec(code) || ' ')[0].match(RE_LIT_REGEX);
|
||
|
|
||
|
if (match) {
|
||
|
const next = pos + match[0].length; // result comes from `re.match`
|
||
|
|
||
|
pos = _prev(code, pos);
|
||
|
let c = code[pos];
|
||
|
|
||
|
// start of buffer or safe prefix?
|
||
|
if (pos < 0 || beforeReChars.includes(c)) {
|
||
|
return next
|
||
|
}
|
||
|
|
||
|
// from here, `pos` is >= 0 and `c` is code[pos]
|
||
|
if (c === '.') {
|
||
|
// can be `...` or something silly like 5./2
|
||
|
if (code[pos - 1] === '.') {
|
||
|
start = next;
|
||
|
}
|
||
|
|
||
|
} else {
|
||
|
|
||
|
if (c === '+' || c === '-') {
|
||
|
// tricky case
|
||
|
if (code[--pos] !== c || // if have a single operator or
|
||
|
(pos = _prev(code, pos)) < 0 || // ...have `++` and no previous token
|
||
|
beforeReSign.includes(c = code[pos])) {
|
||
|
return next // ...this is a regex
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (wordsEndChar.includes(c)) { // looks like a keyword?
|
||
|
const end = pos + 1;
|
||
|
|
||
|
// get the complete (previous) keyword
|
||
|
while (--pos >= 0 && RE_JS_VCHAR.test(code[pos]));
|
||
|
|
||
|
// it is in the allowed keywords list?
|
||
|
if (beforeReWords.includes(code.slice(pos + 1, end))) {
|
||
|
start = next;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
return start
|
||
|
}
|
||
|
|
||
|
/*
|
||
|
* Mini-parser for expressions.
|
||
|
* The main pourpose of this module is to find the end of an expression
|
||
|
* and return its text without the enclosing brackets.
|
||
|
* Does not works with comments, but supports ES6 template strings.
|
||
|
*/
|
||
|
/**
|
||
|
* @exports exprExtr
|
||
|
*/
|
||
|
const S_SQ_STR = /'[^'\n\r\\]*(?:\\(?:\r\n?|[\S\s])[^'\n\r\\]*)*'/.source;
|
||
|
/**
|
||
|
* Matches double quoted JS strings taking care about nested quotes
|
||
|
* and EOLs (escaped EOLs are Ok).
|
||
|
*
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const S_STRING = `${S_SQ_STR}|${S_SQ_STR.replace(/'/g, '"')}`;
|
||
|
/**
|
||
|
* Regex cache
|
||
|
*
|
||
|
* @type {Object.<string, RegExp>}
|
||
|
* @const
|
||
|
* @private
|
||
|
*/
|
||
|
const reBr = {};
|
||
|
/**
|
||
|
* Makes an optimal regex that matches quoted strings, brackets, backquotes
|
||
|
* and the closing brackets of an expression.
|
||
|
*
|
||
|
* @param {string} b - Closing brackets
|
||
|
* @returns {RegExp} - optimized regex
|
||
|
*/
|
||
|
function _regex(b) {
|
||
|
let re = reBr[b];
|
||
|
if (!re) {
|
||
|
let s = escapeStr(b);
|
||
|
if (b.length > 1) {
|
||
|
s = `${s}|[`;
|
||
|
} else {
|
||
|
s = /[{}[\]()]/.test(b) ? '[' : `[${s}`;
|
||
|
}
|
||
|
reBr[b] = re = new RegExp(`${S_STRING}|${s}\`/\\{}[\\]()]`, 'g');
|
||
|
}
|
||
|
return re
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Update the scopes stack removing or adding closures to it
|
||
|
* @param {Array} stack - array stacking the expression closures
|
||
|
* @param {string} char - current char to add or remove from the stack
|
||
|
* @param {string} idx - matching index
|
||
|
* @param {string} code - expression code
|
||
|
* @returns {Object} result
|
||
|
* @returns {Object} result.char - either the char received or the closing braces
|
||
|
* @returns {Object} result.index - either a new index to skip part of the source code,
|
||
|
* or 0 to keep from parsing from the old position
|
||
|
*/
|
||
|
function updateStack(stack, char, idx, code) {
|
||
|
let index = 0;
|
||
|
|
||
|
switch (char) {
|
||
|
case '[':
|
||
|
case '(':
|
||
|
case '{':
|
||
|
stack.push(char === '[' ? ']' : char === '(' ? ')' : '}');
|
||
|
break
|
||
|
case ')':
|
||
|
case ']':
|
||
|
case '}':
|
||
|
if (char !== stack.pop()) {
|
||
|
panic(code, unexpectedCharInExpression.replace('%1', char), index);
|
||
|
}
|
||
|
|
||
|
if (char === '}' && stack[stack.length - 1] === $_ES6_BQ) {
|
||
|
char = stack.pop();
|
||
|
}
|
||
|
|
||
|
index = idx + 1;
|
||
|
break
|
||
|
case '/':
|
||
|
index = skipRegex(code, idx);
|
||
|
}
|
||
|
|
||
|
return { char, index }
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parses the code string searching the end of the expression.
|
||
|
* It skips braces, quoted strings, regexes, and ES6 template literals.
|
||
|
*
|
||
|
* @function exprExtr
|
||
|
* @param {string} code - Buffer to parse
|
||
|
* @param {number} start - Position of the opening brace
|
||
|
* @param {[string,string]} bp - Brackets pair
|
||
|
* @returns {Object} Expression's end (after the closing brace) or -1
|
||
|
* if it is not an expr.
|
||
|
*/
|
||
|
function exprExtr(code, start, bp) {
|
||
|
const [openingBraces, closingBraces] = bp;
|
||
|
const offset = start + openingBraces.length; // skips the opening brace
|
||
|
const stack = []; // expected closing braces ('`' for ES6 TL)
|
||
|
const re = _regex(closingBraces);
|
||
|
|
||
|
re.lastIndex = offset; // begining of the expression
|
||
|
|
||
|
let end;
|
||
|
let match;
|
||
|
|
||
|
while (match = re.exec(code)) { // eslint-disable-line
|
||
|
const idx = match.index;
|
||
|
const str = match[0];
|
||
|
end = re.lastIndex;
|
||
|
|
||
|
// end the iteration
|
||
|
if (str === closingBraces && !stack.length) {
|
||
|
return {
|
||
|
text: code.slice(offset, idx),
|
||
|
start,
|
||
|
end
|
||
|
}
|
||
|
}
|
||
|
|
||
|
const { char, index } = updateStack(stack, str[0], idx, code);
|
||
|
// update the end value depending on the new index received
|
||
|
end = index || end;
|
||
|
// update the regex last index
|
||
|
re.lastIndex = char === $_ES6_BQ ? skipES6TL(code, end, stack) : end;
|
||
|
}
|
||
|
|
||
|
if (stack.length) {
|
||
|
panic(code, unclosedExpression, end);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Outputs the last parsed node. Can be used with a builder too.
|
||
|
*
|
||
|
* @param {ParserStore} store - Parsing store
|
||
|
* @returns {undefined} void function
|
||
|
* @private
|
||
|
*/
|
||
|
function flush(store) {
|
||
|
const last = store.last;
|
||
|
store.last = null;
|
||
|
if (last && store.root) {
|
||
|
store.builder.push(last);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Get the code chunks from start and end range
|
||
|
* @param {string} source - source code
|
||
|
* @param {number} start - Start position of the chunk we want to extract
|
||
|
* @param {number} end - Ending position of the chunk we need
|
||
|
* @returns {string} chunk of code extracted from the source code received
|
||
|
* @private
|
||
|
*/
|
||
|
function getChunk(source, start, end) {
|
||
|
return source.slice(start, end)
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* states text in the last text node, or creates a new one if needed.
|
||
|
*
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @param {number} start - Start position of the tag
|
||
|
* @param {number} end - Ending position (last char of the tag)
|
||
|
* @param {Object} extra - extra properties to add to the text node
|
||
|
* @param {RawExpr[]} extra.expressions - Found expressions
|
||
|
* @param {string} extra.unescape - Brackets to unescape
|
||
|
* @returns {undefined} - void function
|
||
|
* @private
|
||
|
*/
|
||
|
function pushText(state, start, end, extra = {}) {
|
||
|
const text = getChunk(state.data, start, end);
|
||
|
const expressions = extra.expressions;
|
||
|
const unescape = extra.unescape;
|
||
|
|
||
|
let q = state.last;
|
||
|
state.pos = end;
|
||
|
|
||
|
if (q && q.type === TEXT) {
|
||
|
q.text += text;
|
||
|
q.end = end;
|
||
|
} else {
|
||
|
flush(state);
|
||
|
state.last = q = { type: TEXT, text, start, end };
|
||
|
}
|
||
|
|
||
|
if (expressions && expressions.length) {
|
||
|
q.expressions = (q.expressions || []).concat(expressions);
|
||
|
}
|
||
|
|
||
|
if (unescape) {
|
||
|
q.unescape = unescape;
|
||
|
}
|
||
|
|
||
|
return TEXT
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Find the end of the attribute value or text node
|
||
|
* Extract expressions.
|
||
|
* Detect if value have escaped brackets.
|
||
|
*
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @param {HasExpr} node - Node if attr, info if text
|
||
|
* @param {string} endingChars - Ends the value or text
|
||
|
* @param {number} start - Starting position
|
||
|
* @returns {number} Ending position
|
||
|
* @private
|
||
|
*/
|
||
|
function expr(state, node, endingChars, start) {
|
||
|
const re = b0re(state, endingChars);
|
||
|
|
||
|
re.lastIndex = start; // reset re position
|
||
|
|
||
|
const { unescape, expressions, end } = parseExpressions(state, re);
|
||
|
|
||
|
if (node) {
|
||
|
if (unescape) {
|
||
|
node.unescape = unescape;
|
||
|
}
|
||
|
if (expressions.length) {
|
||
|
node.expressions = expressions;
|
||
|
}
|
||
|
} else {
|
||
|
pushText(state, start, end, {expressions, unescape});
|
||
|
}
|
||
|
|
||
|
return end
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse a text chunk finding all the expressions in it
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @param {RegExp} re - regex to match the expressions contents
|
||
|
* @returns {Object} result containing the expression found, the string to unescape and the end position
|
||
|
*/
|
||
|
function parseExpressions(state, re) {
|
||
|
const { data, options } = state;
|
||
|
const { brackets } = options;
|
||
|
const expressions = [];
|
||
|
let unescape, pos, match;
|
||
|
|
||
|
// Anything captured in $1 (closing quote or character) ends the loop...
|
||
|
while ((match = re.exec(data)) && !match[1]) {
|
||
|
// ...else, we have an opening bracket and maybe an expression.
|
||
|
pos = match.index;
|
||
|
if (data[pos - 1] === '\\') {
|
||
|
unescape = match[0]; // it is an escaped opening brace
|
||
|
} else {
|
||
|
const tmpExpr = exprExtr(data, pos, brackets);
|
||
|
if (tmpExpr) {
|
||
|
expressions.push(tmpExpr);
|
||
|
re.lastIndex = tmpExpr.end;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// Even for text, the parser needs match a closing char
|
||
|
if (!match) {
|
||
|
panic(data, unexpectedEndOfFile, pos);
|
||
|
}
|
||
|
|
||
|
return {
|
||
|
unescape,
|
||
|
expressions,
|
||
|
end: match.index
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
|
||
|
/**
|
||
|
* Creates a regex for the given string and the left bracket.
|
||
|
* The string is captured in $1.
|
||
|
*
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @param {string} str - String to search
|
||
|
* @returns {RegExp} Resulting regex.
|
||
|
* @private
|
||
|
*/
|
||
|
function b0re(state, str) {
|
||
|
const { brackets } = state.options;
|
||
|
const re = state.regexCache[str];
|
||
|
|
||
|
if (re) return re
|
||
|
|
||
|
const b0 = escapeStr(brackets[0]);
|
||
|
// cache the regex extending the regexCache object
|
||
|
Object.assign(state.regexCache, { [str]: new RegExp(`(${str})|${b0}`, 'g') });
|
||
|
|
||
|
return state.regexCache[str]
|
||
|
}
|
||
|
|
||
|
// similar to _.uniq
|
||
|
const uniq = l => l.filter((x, i, a) => a.indexOf(x) === i);
|
||
|
|
||
|
/**
|
||
|
* SVG void elements that cannot be auto-closed and shouldn't contain child nodes.
|
||
|
* @const {Array}
|
||
|
*/
|
||
|
const VOID_SVG_TAGS_LIST = [
|
||
|
'circle',
|
||
|
'ellipse',
|
||
|
'line',
|
||
|
'path',
|
||
|
'polygon',
|
||
|
'polyline',
|
||
|
'rect',
|
||
|
'stop',
|
||
|
'use'
|
||
|
];
|
||
|
|
||
|
/**
|
||
|
* List of html elements where the value attribute is allowed
|
||
|
* @type {Array}
|
||
|
*/
|
||
|
const HTML_ELEMENTS_HAVING_VALUE_ATTRIBUTE_LIST = [
|
||
|
'button',
|
||
|
'data',
|
||
|
'input',
|
||
|
'select',
|
||
|
'li',
|
||
|
'meter',
|
||
|
'option',
|
||
|
'output',
|
||
|
'progress',
|
||
|
'textarea',
|
||
|
'param'
|
||
|
];
|
||
|
|
||
|
/**
|
||
|
* List of all the available svg tags
|
||
|
* @const {Array}
|
||
|
* @see {@link https://github.com/wooorm/svg-tag-names}
|
||
|
*/
|
||
|
const SVG_TAGS_LIST = uniq([
|
||
|
'a',
|
||
|
'altGlyph',
|
||
|
'altGlyphDef',
|
||
|
'altGlyphItem',
|
||
|
'animate',
|
||
|
'animateColor',
|
||
|
'animateMotion',
|
||
|
'animateTransform',
|
||
|
'animation',
|
||
|
'audio',
|
||
|
'canvas',
|
||
|
'clipPath',
|
||
|
'color-profile',
|
||
|
'cursor',
|
||
|
'defs',
|
||
|
'desc',
|
||
|
'discard',
|
||
|
'feBlend',
|
||
|
'feColorMatrix',
|
||
|
'feComponentTransfer',
|
||
|
'feComposite',
|
||
|
'feConvolveMatrix',
|
||
|
'feDiffuseLighting',
|
||
|
'feDisplacementMap',
|
||
|
'feDistantLight',
|
||
|
'feDropShadow',
|
||
|
'feFlood',
|
||
|
'feFuncA',
|
||
|
'feFuncB',
|
||
|
'feFuncG',
|
||
|
'feFuncR',
|
||
|
'feGaussianBlur',
|
||
|
'feImage',
|
||
|
'feMerge',
|
||
|
'feMergeNode',
|
||
|
'feMorphology',
|
||
|
'feOffset',
|
||
|
'fePointLight',
|
||
|
'feSpecularLighting',
|
||
|
'feSpotLight',
|
||
|
'feTile',
|
||
|
'feTurbulence',
|
||
|
'filter',
|
||
|
'font',
|
||
|
'font-face',
|
||
|
'font-face-format',
|
||
|
'font-face-name',
|
||
|
'font-face-src',
|
||
|
'font-face-uri',
|
||
|
'foreignObject',
|
||
|
'g',
|
||
|
'glyph',
|
||
|
'glyphRef',
|
||
|
'handler',
|
||
|
'hatch',
|
||
|
'hatchpath',
|
||
|
'hkern',
|
||
|
'iframe',
|
||
|
'image',
|
||
|
'linearGradient',
|
||
|
'listener',
|
||
|
'marker',
|
||
|
'mask',
|
||
|
'mesh',
|
||
|
'meshgradient',
|
||
|
'meshpatch',
|
||
|
'meshrow',
|
||
|
'metadata',
|
||
|
'missing-glyph',
|
||
|
'mpath',
|
||
|
'pattern',
|
||
|
'prefetch',
|
||
|
'radialGradient',
|
||
|
'script',
|
||
|
'set',
|
||
|
'solidColor',
|
||
|
'solidcolor',
|
||
|
'style',
|
||
|
'svg',
|
||
|
'switch',
|
||
|
'symbol',
|
||
|
'tbreak',
|
||
|
'text',
|
||
|
'textArea',
|
||
|
'textPath',
|
||
|
'title',
|
||
|
'tref',
|
||
|
'tspan',
|
||
|
'unknown',
|
||
|
'video',
|
||
|
'view',
|
||
|
'vkern'
|
||
|
].concat(VOID_SVG_TAGS_LIST)).sort();
|
||
|
|
||
|
/**
|
||
|
* HTML void elements that cannot be auto-closed and shouldn't contain child nodes.
|
||
|
* @type {Array}
|
||
|
* @see {@link http://www.w3.org/TR/html-markup/syntax.html#syntax-elements}
|
||
|
* @see {@link http://www.w3.org/TR/html5/syntax.html#void-elements}
|
||
|
*/
|
||
|
const VOID_HTML_TAGS_LIST = [
|
||
|
'area',
|
||
|
'base',
|
||
|
'br',
|
||
|
'col',
|
||
|
'embed',
|
||
|
'hr',
|
||
|
'img',
|
||
|
'input',
|
||
|
'keygen',
|
||
|
'link',
|
||
|
'menuitem',
|
||
|
'meta',
|
||
|
'param',
|
||
|
'source',
|
||
|
'track',
|
||
|
'wbr'
|
||
|
];
|
||
|
|
||
|
/**
|
||
|
* List of all the html tags
|
||
|
* @const {Array}
|
||
|
* @see {@link https://github.com/sindresorhus/html-tags}
|
||
|
*/
|
||
|
const HTML_TAGS_LIST = uniq([
|
||
|
'a',
|
||
|
'abbr',
|
||
|
'address',
|
||
|
'article',
|
||
|
'aside',
|
||
|
'audio',
|
||
|
'b',
|
||
|
'bdi',
|
||
|
'bdo',
|
||
|
'blockquote',
|
||
|
'body',
|
||
|
'canvas',
|
||
|
'caption',
|
||
|
'cite',
|
||
|
'code',
|
||
|
'colgroup',
|
||
|
'datalist',
|
||
|
'dd',
|
||
|
'del',
|
||
|
'details',
|
||
|
'dfn',
|
||
|
'dialog',
|
||
|
'div',
|
||
|
'dl',
|
||
|
'dt',
|
||
|
'em',
|
||
|
'fieldset',
|
||
|
'figcaption',
|
||
|
'figure',
|
||
|
'footer',
|
||
|
'form',
|
||
|
'h1',
|
||
|
'h2',
|
||
|
'h3',
|
||
|
'h4',
|
||
|
'h5',
|
||
|
'h6',
|
||
|
'head',
|
||
|
'header',
|
||
|
'hgroup',
|
||
|
'html',
|
||
|
'i',
|
||
|
'iframe',
|
||
|
'ins',
|
||
|
'kbd',
|
||
|
'label',
|
||
|
'legend',
|
||
|
'main',
|
||
|
'map',
|
||
|
'mark',
|
||
|
'math',
|
||
|
'menu',
|
||
|
'nav',
|
||
|
'noscript',
|
||
|
'object',
|
||
|
'ol',
|
||
|
'optgroup',
|
||
|
'p',
|
||
|
'picture',
|
||
|
'pre',
|
||
|
'q',
|
||
|
'rb',
|
||
|
'rp',
|
||
|
'rt',
|
||
|
'rtc',
|
||
|
'ruby',
|
||
|
's',
|
||
|
'samp',
|
||
|
'script',
|
||
|
'section',
|
||
|
'select',
|
||
|
'slot',
|
||
|
'small',
|
||
|
'span',
|
||
|
'strong',
|
||
|
'style',
|
||
|
'sub',
|
||
|
'summary',
|
||
|
'sup',
|
||
|
'svg',
|
||
|
'table',
|
||
|
'tbody',
|
||
|
'td',
|
||
|
'template',
|
||
|
'tfoot',
|
||
|
'th',
|
||
|
'thead',
|
||
|
'time',
|
||
|
'title',
|
||
|
'tr',
|
||
|
'u',
|
||
|
'ul',
|
||
|
'var',
|
||
|
'video'
|
||
|
]
|
||
|
.concat(VOID_HTML_TAGS_LIST)
|
||
|
.concat(HTML_ELEMENTS_HAVING_VALUE_ATTRIBUTE_LIST)
|
||
|
).sort();
|
||
|
|
||
|
/**
|
||
|
* List of all boolean HTML attributes
|
||
|
* @const {RegExp}
|
||
|
* @see {@link https://www.w3.org/TR/html5/infrastructure.html#sec-boolean-attributes}
|
||
|
*/
|
||
|
const BOOLEAN_ATTRIBUTES_LIST = [
|
||
|
'disabled',
|
||
|
'visible',
|
||
|
'checked',
|
||
|
'readonly',
|
||
|
'required',
|
||
|
'allowfullscreen',
|
||
|
'autofocus',
|
||
|
'autoplay',
|
||
|
'compact',
|
||
|
'controls',
|
||
|
'default',
|
||
|
'formnovalidate',
|
||
|
'hidden',
|
||
|
'ismap',
|
||
|
'itemscope',
|
||
|
'loop',
|
||
|
'multiple',
|
||
|
'muted',
|
||
|
'noresize',
|
||
|
'noshade',
|
||
|
'novalidate',
|
||
|
'nowrap',
|
||
|
'open',
|
||
|
'reversed',
|
||
|
'seamless',
|
||
|
'selected',
|
||
|
'sortable',
|
||
|
'truespeed',
|
||
|
'typemustmatch'
|
||
|
];
|
||
|
|
||
|
/**
|
||
|
* Join a list of items with the pipe symbol (usefull for regex list concatenation)
|
||
|
* @private
|
||
|
* @param {Array} list - list of strings
|
||
|
* @returns {string} the list received joined with pipes
|
||
|
*/
|
||
|
function joinWithPipe(list) {
|
||
|
return list.join('|')
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Convert list of strings to regex in order to test against it ignoring the cases
|
||
|
* @private
|
||
|
* @param {...Array} lists - array of strings
|
||
|
* @returns {RegExp} regex that will match all the strings in the array received ignoring the cases
|
||
|
*/
|
||
|
function listsToRegex(...lists) {
|
||
|
return new RegExp(`^/?(?:${joinWithPipe(lists.map(joinWithPipe))})$`, 'i')
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the html tags ignoring the cases
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const HTML_TAGS_RE = listsToRegex(HTML_TAGS_LIST);
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the svg tags ignoring the cases
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const SVG_TAGS_RE = listsToRegex(SVG_TAGS_LIST);
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the void html tags ignoring the cases
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const VOID_HTML_TAGS_RE = listsToRegex(VOID_HTML_TAGS_LIST);
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the void svg tags ignoring the cases
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const VOID_SVG_TAGS_RE = listsToRegex(VOID_SVG_TAGS_LIST);
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the html tags where the value tag is allowed
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const HTML_ELEMENTS_HAVING_VALUE_ATTRIBUTE_RE = listsToRegex(HTML_ELEMENTS_HAVING_VALUE_ATTRIBUTE_LIST);
|
||
|
|
||
|
/**
|
||
|
* Regex matching all the boolean attributes
|
||
|
* @const {RegExp}
|
||
|
*/
|
||
|
const BOOLEAN_ATTRIBUTES_RE = listsToRegex(BOOLEAN_ATTRIBUTES_LIST);
|
||
|
|
||
|
/**
|
||
|
* True if it's a self closing tag
|
||
|
* @param {string} tag - test tag
|
||
|
* @returns {boolean} true if void
|
||
|
* @example
|
||
|
* isVoid('meta') // true
|
||
|
* isVoid('circle') // true
|
||
|
* isVoid('IMG') // true
|
||
|
* isVoid('div') // false
|
||
|
* isVoid('mask') // false
|
||
|
*/
|
||
|
function isVoid(tag) {
|
||
|
return [
|
||
|
VOID_HTML_TAGS_RE,
|
||
|
VOID_SVG_TAGS_RE
|
||
|
].some(r => r.test(tag))
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* True if it's not SVG nor a HTML known tag
|
||
|
* @param {string} tag - test tag
|
||
|
* @returns {boolean} true if custom element
|
||
|
* @example
|
||
|
* isCustom('my-component') // true
|
||
|
* isCustom('div') // false
|
||
|
*/
|
||
|
function isCustom(tag) {
|
||
|
return [
|
||
|
HTML_TAGS_RE,
|
||
|
SVG_TAGS_RE
|
||
|
].every(l => !l.test(tag))
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* True if it's a boolean attribute
|
||
|
* @param {string} attribute - test attribute
|
||
|
* @returns {boolean} true if the attribute is a boolean type
|
||
|
* @example
|
||
|
* isBoolAttribute('selected') // true
|
||
|
* isBoolAttribute('class') // false
|
||
|
*/
|
||
|
function isBoolAttribute(attribute) {
|
||
|
return BOOLEAN_ATTRIBUTES_RE.test(attribute)
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Memoization function
|
||
|
* @param {Function} fn - function to memoize
|
||
|
* @returns {*} return of the function to memoize
|
||
|
*/
|
||
|
function memoize(fn) {
|
||
|
const cache = new WeakMap();
|
||
|
|
||
|
return (...args) => {
|
||
|
if (cache.has(args[0])) return cache.get(args[0])
|
||
|
|
||
|
const ret = fn(...args);
|
||
|
|
||
|
cache.set(args[0], ret);
|
||
|
|
||
|
return ret
|
||
|
}
|
||
|
}
|
||
|
|
||
|
const expressionsContentRe = memoize(brackets => RegExp(`(${brackets[0]}[^${brackets[1]}]*?${brackets[1]})`, 'g'));
|
||
|
const isSpreadAttribute = name => SPREAD_OPERATOR.test(name);
|
||
|
const isAttributeExpression = (name, brackets) => name[0] === brackets[0];
|
||
|
const getAttributeEnd = (state, attr) => expr(state, attr, '[>/\\s]', attr.start);
|
||
|
|
||
|
/**
|
||
|
* The more complex parsing is for attributes as it can contain quoted or
|
||
|
* unquoted values or expressions.
|
||
|
*
|
||
|
* @param {ParserStore} state - Parser state
|
||
|
* @returns {number} New parser mode.
|
||
|
* @private
|
||
|
*/
|
||
|
function attr(state) {
|
||
|
const { data, last, pos, root } = state;
|
||
|
const tag = last; // the last (current) tag in the output
|
||
|
const _CH = /\S/g; // matches the first non-space char
|
||
|
const ch = execFromPos(_CH, pos, data);
|
||
|
|
||
|
switch (true) {
|
||
|
case !ch:
|
||
|
state.pos = data.length; // reaching the end of the buffer with
|
||
|
// NodeTypes.ATTR will generate error
|
||
|
break
|
||
|
case ch[0] === '>':
|
||
|
// closing char found. If this is a self-closing tag with the name of the
|
||
|
// Root tag, we need decrement the counter as we are changing mode.
|
||
|
state.pos = tag.end = _CH.lastIndex;
|
||
|
if (tag[IS_SELF_CLOSING]) {
|
||
|
state.scryle = null; // allow selfClosing script/style tags
|
||
|
if (root && root.name === tag.name) {
|
||
|
state.count--; // "pop" root tag
|
||
|
}
|
||
|
}
|
||
|
return TEXT
|
||
|
case ch[0] === '/':
|
||
|
state.pos = _CH.lastIndex; // maybe. delegate the validation
|
||
|
tag[IS_SELF_CLOSING] = true; // the next loop
|
||
|
break
|
||
|
default:
|
||
|
delete tag[IS_SELF_CLOSING]; // ensure unmark as selfclosing tag
|
||
|
setAttribute(state, ch.index, tag);
|
||
|
}
|
||
|
|
||
|
return ATTR
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parses an attribute and its expressions.
|
||
|
*
|
||
|
* @param {ParserStore} state - Parser state
|
||
|
* @param {number} pos - Starting position of the attribute
|
||
|
* @param {Object} tag - Current parent tag
|
||
|
* @returns {undefined} void function
|
||
|
* @private
|
||
|
*/
|
||
|
function setAttribute(state, pos, tag) {
|
||
|
const { data } = state;
|
||
|
const expressionContent = expressionsContentRe(state.options.brackets);
|
||
|
const re = ATTR_START; // (\S[^>/=\s]*)(?:\s*=\s*([^>/])?)? g
|
||
|
const start = re.lastIndex = expressionContent.lastIndex = pos; // first non-whitespace
|
||
|
const attrMatches = re.exec(data);
|
||
|
const isExpressionName = isAttributeExpression(attrMatches[1], state.options.brackets);
|
||
|
const match = isExpressionName ? [null, expressionContent.exec(data)[1], null] : attrMatches;
|
||
|
|
||
|
if (match) {
|
||
|
const end = re.lastIndex;
|
||
|
const attr = parseAttribute(state, match, start, end, isExpressionName);
|
||
|
|
||
|
//assert(q && q.type === Mode.TAG, 'no previous tag for the attr!')
|
||
|
// Pushes the attribute and shifts the `end` position of the tag (`last`).
|
||
|
state.pos = tag.end = attr.end;
|
||
|
tag.attributes = addToCollection(tag.attributes, attr);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
function parseNomalAttribute(state, attr, quote) {
|
||
|
const { data } = state;
|
||
|
let { end } = attr;
|
||
|
|
||
|
if (isBoolAttribute(attr.name)) {
|
||
|
attr[IS_BOOLEAN] = true;
|
||
|
}
|
||
|
|
||
|
// parse the whole value (if any) and get any expressions on it
|
||
|
if (quote) {
|
||
|
// Usually, the value's first char (`quote`) is a quote and the lastIndex
|
||
|
// (`end`) is the start of the value.
|
||
|
let valueStart = end;
|
||
|
// If it not, this is an unquoted value and we need adjust the start.
|
||
|
if (quote !== '"' && quote !== '\'') {
|
||
|
quote = ''; // first char of value is not a quote
|
||
|
valueStart--; // adjust the starting position
|
||
|
}
|
||
|
|
||
|
end = expr(state, attr, quote || '[>/\\s]', valueStart);
|
||
|
|
||
|
// adjust the bounds of the value and save its content
|
||
|
return Object.assign(attr, {
|
||
|
value: getChunk(data, valueStart, end),
|
||
|
valueStart,
|
||
|
end: quote ? ++end : end
|
||
|
})
|
||
|
}
|
||
|
|
||
|
return attr
|
||
|
}
|
||
|
|
||
|
|
||
|
/**
|
||
|
* Parse expression names <a {href}>
|
||
|
* @param {ParserStore} state - Parser state
|
||
|
* @param {Object} attr - attribute object parsed
|
||
|
* @returns {Object} normalized attribute object
|
||
|
*/
|
||
|
function parseSpreadAttribute(state, attr) {
|
||
|
const end = getAttributeEnd(state, attr);
|
||
|
|
||
|
return {
|
||
|
[IS_SPREAD]: true,
|
||
|
start: attr.start,
|
||
|
expressions: attr.expressions.map(expr => Object.assign(expr, {
|
||
|
text: expr.text.replace(SPREAD_OPERATOR, '').trim()
|
||
|
})),
|
||
|
end: end
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse expression names <a {href}>
|
||
|
* @param {ParserStore} state - Parser state
|
||
|
* @param {Object} attr - attribute object parsed
|
||
|
* @returns {Object} normalized attribute object
|
||
|
*/
|
||
|
function parseExpressionNameAttribute(state, attr) {
|
||
|
const end = getAttributeEnd(state, attr);
|
||
|
|
||
|
return {
|
||
|
start: attr.start,
|
||
|
name: attr.expressions[0].text.trim(),
|
||
|
expressions: attr.expressions,
|
||
|
end: end
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse the attribute values normalising the quotes
|
||
|
* @param {ParserStore} state - Parser state
|
||
|
* @param {Array} match - results of the attributes regex
|
||
|
* @param {number} start - attribute start position
|
||
|
* @param {number} end - attribute end position
|
||
|
* @param {boolean} isExpressionName - true if the attribute name is an expression
|
||
|
* @returns {Object} attribute object
|
||
|
*/
|
||
|
function parseAttribute(state, match, start, end, isExpressionName) {
|
||
|
const attr = {
|
||
|
name: match[1],
|
||
|
value: '',
|
||
|
start,
|
||
|
end
|
||
|
};
|
||
|
|
||
|
const quote = match[2]; // first letter of value or nothing
|
||
|
|
||
|
switch (true) {
|
||
|
case isSpreadAttribute(attr.name):
|
||
|
return parseSpreadAttribute(state, attr)
|
||
|
case isExpressionName === true:
|
||
|
return parseExpressionNameAttribute(state, attr)
|
||
|
default:
|
||
|
return parseNomalAttribute(state, attr, quote)
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Function to curry any javascript method
|
||
|
* @param {Function} fn - the target function we want to curry
|
||
|
* @param {...[args]} acc - initial arguments
|
||
|
* @returns {Function|*} it will return a function until the target function
|
||
|
* will receive all of its arguments
|
||
|
*/
|
||
|
function curry(fn, ...acc) {
|
||
|
return (...args) => {
|
||
|
args = [...acc, ...args];
|
||
|
|
||
|
return args.length < fn.length ?
|
||
|
curry(fn, ...args) :
|
||
|
fn(...args)
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parses comments in long or short form
|
||
|
* (any DOCTYPE & CDATA blocks are parsed as comments).
|
||
|
*
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @param {string} data - Buffer to parse
|
||
|
* @param {number} start - Position of the '<!' sequence
|
||
|
* @returns {number} node type id
|
||
|
* @private
|
||
|
*/
|
||
|
function comment(state, data, start) {
|
||
|
const pos = start + 2; // skip '<!'
|
||
|
const isLongComment = data.substr(pos, 2) === '--';
|
||
|
const str = isLongComment ? '-->' : '>';
|
||
|
const end = data.indexOf(str, pos);
|
||
|
|
||
|
if (end < 0) {
|
||
|
panic(data, unclosedComment, start);
|
||
|
}
|
||
|
|
||
|
pushComment(
|
||
|
state,
|
||
|
start,
|
||
|
end + str.length,
|
||
|
data.substring(start, end + str.length)
|
||
|
);
|
||
|
|
||
|
return TEXT
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse a comment.
|
||
|
*
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @param {number} start - Start position of the tag
|
||
|
* @param {number} end - Ending position (last char of the tag)
|
||
|
* @param {string} text - Comment content
|
||
|
* @returns {undefined} void function
|
||
|
* @private
|
||
|
*/
|
||
|
function pushComment(state, start, end, text) {
|
||
|
state.pos = end;
|
||
|
if (state.options.comments === true) {
|
||
|
flush(state);
|
||
|
state.last = {
|
||
|
type: COMMENT,
|
||
|
start,
|
||
|
end,
|
||
|
text
|
||
|
};
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Pushes a new *tag* and set `last` to this, so any attributes
|
||
|
* will be included on this and shifts the `end`.
|
||
|
*
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @param {string} name - Name of the node including any slash
|
||
|
* @param {number} start - Start position of the tag
|
||
|
* @param {number} end - Ending position (last char of the tag + 1)
|
||
|
* @returns {undefined} - void function
|
||
|
* @private
|
||
|
*/
|
||
|
function pushTag(state, name, start, end) {
|
||
|
const root = state.root;
|
||
|
const last = { type: TAG, name, start, end };
|
||
|
|
||
|
if (isCustom(name)) {
|
||
|
last[IS_CUSTOM] = true;
|
||
|
}
|
||
|
|
||
|
if (isVoid(name)) {
|
||
|
last[IS_VOID] = true;
|
||
|
}
|
||
|
|
||
|
state.pos = end;
|
||
|
|
||
|
if (root) {
|
||
|
if (name === root.name) {
|
||
|
state.count++;
|
||
|
} else if (name === root.close) {
|
||
|
state.count--;
|
||
|
}
|
||
|
flush(state);
|
||
|
} else {
|
||
|
// start with root (keep ref to output)
|
||
|
state.root = { name: last.name, close: `/${name}` };
|
||
|
state.count = 1;
|
||
|
}
|
||
|
|
||
|
state.last = last;
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse the tag following a '<' character, or delegate to other parser
|
||
|
* if an invalid tag name is found.
|
||
|
*
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @returns {number} New parser mode
|
||
|
* @private
|
||
|
*/
|
||
|
function tag(state) {
|
||
|
const { pos, data } = state; // pos of the char following '<'
|
||
|
const start = pos - 1; // pos of '<'
|
||
|
const str = data.substr(pos, 2); // first two chars following '<'
|
||
|
|
||
|
switch (true) {
|
||
|
case str[0] === '!':
|
||
|
return comment(state, data, start)
|
||
|
case TAG_2C.test(str):
|
||
|
return parseTag(state, start)
|
||
|
default:
|
||
|
return pushText(state, start, pos) // pushes the '<' as text
|
||
|
}
|
||
|
}
|
||
|
|
||
|
function parseTag(state, start) {
|
||
|
const { data, pos } = state;
|
||
|
const re = TAG_NAME; // (\/?[^\s>/]+)\s*(>)? g
|
||
|
const match = execFromPos(re, pos, data);
|
||
|
const end = re.lastIndex;
|
||
|
const name = match[1].toLowerCase(); // $1: tag name including any '/'
|
||
|
// script/style block is parsed as another tag to extract attributes
|
||
|
if (name in RE_SCRYLE) {
|
||
|
state.scryle = name; // used by parseText
|
||
|
}
|
||
|
|
||
|
pushTag(state, name, start, end);
|
||
|
// only '>' can ends the tag here, the '/' is handled in parseAttribute
|
||
|
if (!match[2]) {
|
||
|
return ATTR
|
||
|
}
|
||
|
|
||
|
return TEXT
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parses regular text and script/style blocks ...scryle for short :-)
|
||
|
* (the content of script and style is text as well)
|
||
|
*
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @returns {number} New parser mode.
|
||
|
* @private
|
||
|
*/
|
||
|
function text(state) {
|
||
|
const { pos, data, scryle } = state;
|
||
|
|
||
|
switch (true) {
|
||
|
case typeof scryle === 'string': {
|
||
|
const name = scryle;
|
||
|
const re = RE_SCRYLE[name];
|
||
|
const match = execFromPos(re, pos, data);
|
||
|
|
||
|
if (!match) {
|
||
|
panic(data, unclosedNamedBlock.replace('%1', name), pos - 1);
|
||
|
}
|
||
|
|
||
|
const start = match.index;
|
||
|
const end = re.lastIndex;
|
||
|
state.scryle = null; // reset the script/style flag now
|
||
|
// write the tag content, if any
|
||
|
if (start > pos) {
|
||
|
parseSpecialTagsContent(state, name, match);
|
||
|
}
|
||
|
// now the closing tag, either </script> or </style>
|
||
|
pushTag(state, `/${name}`, start, end);
|
||
|
break
|
||
|
}
|
||
|
case data[pos] === '<':
|
||
|
state.pos++;
|
||
|
return TAG
|
||
|
default:
|
||
|
expr(state, null, '<', pos);
|
||
|
}
|
||
|
|
||
|
return TEXT
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parse the text content depending on the name
|
||
|
* @param {ParserState} state - Parser state
|
||
|
* @param {string} name - one of the tags matched by the RE_SCRYLE regex
|
||
|
* @param {Array} match - result of the regex matching the content of the parsed tag
|
||
|
* @returns {undefined} void function
|
||
|
*/
|
||
|
function parseSpecialTagsContent(state, name, match) {
|
||
|
const { pos } = state;
|
||
|
const start = match.index;
|
||
|
|
||
|
if (name === TEXTAREA_TAG) {
|
||
|
expr(state, null, match[0], pos);
|
||
|
} else {
|
||
|
pushText(state, pos, start);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/*---------------------------------------------------------------------
|
||
|
* Tree builder for the riot tag parser.
|
||
|
*
|
||
|
* The output has a root property and separate arrays for `html`, `css`,
|
||
|
* and `js` tags.
|
||
|
*
|
||
|
* The root tag is included as first element in the `html` array.
|
||
|
* Script tags marked with "defer" are included in `html` instead `js`.
|
||
|
*
|
||
|
* - Mark SVG tags
|
||
|
* - Mark raw tags
|
||
|
* - Mark void tags
|
||
|
* - Split prefixes from expressions
|
||
|
* - Unescape escaped brackets and escape EOLs and backslashes
|
||
|
* - Compact whitespace (option `compact`) for non-raw tags
|
||
|
* - Create an array `parts` for text nodes and attributes
|
||
|
*
|
||
|
* Throws on unclosed tags or closing tags without start tag.
|
||
|
* Selfclosing and void tags has no nodes[] property.
|
||
|
*/
|
||
|
|
||
|
/**
|
||
|
* Escape the carriage return and the line feed from a string
|
||
|
* @param {string} string - input string
|
||
|
* @returns {string} output string escaped
|
||
|
*/
|
||
|
function escapeReturn(string) {
|
||
|
return string
|
||
|
.replace(/\r/g, '\\r')
|
||
|
.replace(/\n/g, '\\n')
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Escape double slashes in a string
|
||
|
* @param {string} string - input string
|
||
|
* @returns {string} output string escaped
|
||
|
*/
|
||
|
function escapeSlashes(string) {
|
||
|
return string.replace(/\\/g, '\\\\')
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Replace the multiple spaces with only one
|
||
|
* @param {string} string - input string
|
||
|
* @returns {string} string without trailing spaces
|
||
|
*/
|
||
|
function cleanSpaces(string) {
|
||
|
return string.replace(/\s+/g, ' ')
|
||
|
}
|
||
|
|
||
|
const TREE_BUILDER_STRUCT = Object.seal({
|
||
|
get() {
|
||
|
const store = this.store;
|
||
|
// The real root tag is in store.root.nodes[0]
|
||
|
return {
|
||
|
[TEMPLATE_OUTPUT_NAME]: store.root.nodes[0],
|
||
|
[CSS_OUTPUT_NAME]: store[STYLE_TAG],
|
||
|
[JAVASCRIPT_OUTPUT_NAME]: store[JAVASCRIPT_TAG]
|
||
|
}
|
||
|
},
|
||
|
|
||
|
/**
|
||
|
* Process the current tag or text.
|
||
|
* @param {Object} node - Raw pseudo-node from the parser
|
||
|
* @returns {undefined} void function
|
||
|
*/
|
||
|
push(node) {
|
||
|
const store = this.store;
|
||
|
|
||
|
switch (node.type) {
|
||
|
case COMMENT:
|
||
|
this.pushComment(store, node);
|
||
|
break
|
||
|
case TEXT:
|
||
|
this.pushText(store, node);
|
||
|
break
|
||
|
case TAG: {
|
||
|
const name = node.name;
|
||
|
const closingTagChar = '/';
|
||
|
const [firstChar] = name;
|
||
|
|
||
|
if (firstChar === closingTagChar && !node.isVoid) {
|
||
|
this.closeTag(store, node, name);
|
||
|
} else if (firstChar !== closingTagChar) {
|
||
|
this.openTag(store, node);
|
||
|
}
|
||
|
break
|
||
|
}
|
||
|
}
|
||
|
},
|
||
|
pushComment(store, node) {
|
||
|
const parent = store.last;
|
||
|
|
||
|
parent.nodes.push(node);
|
||
|
},
|
||
|
closeTag(store, node) {
|
||
|
const last = store.scryle || store.last;
|
||
|
|
||
|
last.end = node.end;
|
||
|
|
||
|
// update always the root node end position
|
||
|
if (store.root.nodes[0]) store.root.nodes[0].end = node.end;
|
||
|
|
||
|
if (store.scryle) {
|
||
|
store.scryle = null;
|
||
|
} else {
|
||
|
store.last = store.stack.pop();
|
||
|
}
|
||
|
},
|
||
|
|
||
|
openTag(store, node) {
|
||
|
const name = node.name;
|
||
|
const attrs = node.attributes;
|
||
|
|
||
|
if ([JAVASCRIPT_TAG, STYLE_TAG].includes(name)) {
|
||
|
// Only accept one of each
|
||
|
if (store[name]) {
|
||
|
panic(this.store.data, duplicatedNamedTag.replace('%1', name), node.start);
|
||
|
}
|
||
|
|
||
|
store[name] = node;
|
||
|
store.scryle = store[name];
|
||
|
|
||
|
} else {
|
||
|
// store.last holds the last tag pushed in the stack and this are
|
||
|
// non-void, non-empty tags, so we are sure the `lastTag` here
|
||
|
// have a `nodes` property.
|
||
|
const lastTag = store.last;
|
||
|
const newNode = node;
|
||
|
|
||
|
lastTag.nodes.push(newNode);
|
||
|
|
||
|
if (lastTag[IS_RAW] || RAW_TAGS.test(name)) {
|
||
|
node[IS_RAW] = true;
|
||
|
}
|
||
|
|
||
|
if (!node[IS_SELF_CLOSING] && !node[IS_VOID]) {
|
||
|
store.stack.push(lastTag);
|
||
|
newNode.nodes = [];
|
||
|
store.last = newNode;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
if (attrs) {
|
||
|
this.attrs(attrs);
|
||
|
}
|
||
|
},
|
||
|
attrs(attributes) {
|
||
|
attributes.forEach(attr => {
|
||
|
if (attr.value) {
|
||
|
this.split(attr, attr.value, attr.valueStart, true);
|
||
|
}
|
||
|
});
|
||
|
},
|
||
|
pushText(store, node) {
|
||
|
const text = node.text;
|
||
|
const empty = !/\S/.test(text);
|
||
|
const scryle = store.scryle;
|
||
|
if (!scryle) {
|
||
|
// store.last always have a nodes property
|
||
|
const parent = store.last;
|
||
|
|
||
|
const pack = this.compact && !parent[IS_RAW];
|
||
|
if (pack && empty) {
|
||
|
return
|
||
|
}
|
||
|
this.split(node, text, node.start, pack);
|
||
|
parent.nodes.push(node);
|
||
|
} else if (!empty) {
|
||
|
scryle.text = node;
|
||
|
}
|
||
|
},
|
||
|
split(node, source, start, pack) {
|
||
|
const expressions = node.expressions;
|
||
|
const parts = [];
|
||
|
|
||
|
if (expressions) {
|
||
|
let pos = 0;
|
||
|
|
||
|
expressions.forEach(expr => {
|
||
|
const text = source.slice(pos, expr.start - start);
|
||
|
const code = expr.text;
|
||
|
parts.push(this.sanitise(node, text, pack), escapeReturn(escapeSlashes(code).trim()));
|
||
|
pos = expr.end - start;
|
||
|
});
|
||
|
|
||
|
if (pos < node.end) {
|
||
|
parts.push(this.sanitise(node, source.slice(pos), pack));
|
||
|
}
|
||
|
} else {
|
||
|
parts[0] = this.sanitise(node, source, pack);
|
||
|
}
|
||
|
|
||
|
node.parts = parts.filter(p => p); // remove the empty strings
|
||
|
},
|
||
|
// unescape escaped brackets and split prefixes of expressions
|
||
|
sanitise(node, text, pack) {
|
||
|
let rep = node.unescape;
|
||
|
if (rep) {
|
||
|
let idx = 0;
|
||
|
rep = `\\${rep}`;
|
||
|
while ((idx = text.indexOf(rep, idx)) !== -1) {
|
||
|
text = text.substr(0, idx) + text.substr(idx + 1);
|
||
|
idx++;
|
||
|
}
|
||
|
}
|
||
|
|
||
|
text = escapeSlashes(text);
|
||
|
|
||
|
return pack ? cleanSpaces(text) : escapeReturn(text)
|
||
|
}
|
||
|
});
|
||
|
|
||
|
function createTreeBuilder(data, options) {
|
||
|
const root = {
|
||
|
type: TAG,
|
||
|
name: '',
|
||
|
start: 0,
|
||
|
end: 0,
|
||
|
nodes: []
|
||
|
};
|
||
|
|
||
|
return Object.assign(Object.create(TREE_BUILDER_STRUCT), {
|
||
|
compact: options.compact !== false,
|
||
|
store: {
|
||
|
last: root,
|
||
|
stack: [],
|
||
|
scryle: null,
|
||
|
root,
|
||
|
style: null,
|
||
|
script: null,
|
||
|
data
|
||
|
}
|
||
|
})
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Factory for the Parser class, exposing only the `parse` method.
|
||
|
* The export adds the Parser class as property.
|
||
|
*
|
||
|
* @param {Object} options - User Options
|
||
|
* @param {Function} customBuilder - Tree builder factory
|
||
|
* @returns {Function} Public Parser implementation.
|
||
|
*/
|
||
|
function parser(options, customBuilder) {
|
||
|
const state = curry(createParserState)(options, customBuilder || createTreeBuilder);
|
||
|
return {
|
||
|
parse: (data) => parse(state(data))
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Create a new state object
|
||
|
* @param {Object} userOptions - parser options
|
||
|
* @param {Function} builder - Tree builder factory
|
||
|
* @param {string} data - data to parse
|
||
|
* @returns {ParserState} it represents the current parser state
|
||
|
*/
|
||
|
function createParserState(userOptions, builder, data) {
|
||
|
const options = Object.assign({
|
||
|
brackets: ['{', '}']
|
||
|
}, userOptions);
|
||
|
|
||
|
return {
|
||
|
options,
|
||
|
regexCache: {},
|
||
|
pos: 0,
|
||
|
count: -1,
|
||
|
root: null,
|
||
|
last: null,
|
||
|
scryle: null,
|
||
|
builder: builder(data, options),
|
||
|
data
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* It creates a raw output of pseudo-nodes with one of three different types,
|
||
|
* all of them having a start/end position:
|
||
|
*
|
||
|
* - TAG -- Opening or closing tags
|
||
|
* - TEXT -- Raw text
|
||
|
* - COMMENT -- Comments
|
||
|
*
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @returns {ParserResult} Result, contains data and output properties.
|
||
|
*/
|
||
|
function parse(state) {
|
||
|
const { data } = state;
|
||
|
|
||
|
walk(state);
|
||
|
flush(state);
|
||
|
|
||
|
if (state.count) {
|
||
|
panic(data, state.count > 0 ? unexpectedEndOfFile : rootTagNotFound, state.pos);
|
||
|
}
|
||
|
|
||
|
return {
|
||
|
data,
|
||
|
output: state.builder.get()
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Parser walking recursive function
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @param {string} type - current parsing context
|
||
|
* @returns {undefined} void function
|
||
|
*/
|
||
|
function walk(state, type) {
|
||
|
const { data } = state;
|
||
|
// extend the state adding the tree builder instance and the initial data
|
||
|
const length = data.length;
|
||
|
|
||
|
// The "count" property is set to 1 when the first tag is found.
|
||
|
// This becomes the root and precedent text or comments are discarded.
|
||
|
// So, at the end of the parsing count must be zero.
|
||
|
if (state.pos < length && state.count) {
|
||
|
walk(state, eat(state, type));
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Function to help iterating on the current parser state
|
||
|
* @param {ParserState} state - Current parser state
|
||
|
* @param {string} type - current parsing context
|
||
|
* @returns {string} parsing context
|
||
|
*/
|
||
|
function eat(state, type) {
|
||
|
switch (type) {
|
||
|
case TAG:
|
||
|
return tag(state)
|
||
|
case ATTR:
|
||
|
return attr(state)
|
||
|
default:
|
||
|
return text(state)
|
||
|
}
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Expose the internal constants
|
||
|
*/
|
||
|
const constants = c;
|
||
|
|
||
|
/**
|
||
|
* The nodeTypes definition
|
||
|
*/
|
||
|
const nodeTypes = types;
|
||
|
|
||
|
exports.constants = constants;
|
||
|
exports.default = parser;
|
||
|
exports.nodeTypes = nodeTypes;
|