905 lines
27 KiB
JavaScript
905 lines
27 KiB
JavaScript
'use strict';
|
||
|
||
const generate = require('@babel/regjsgen').generate;
|
||
const parse = require('regjsparser').parse;
|
||
const regenerate = require('regenerate');
|
||
const unicodeMatchProperty = require('unicode-match-property-ecmascript');
|
||
const unicodeMatchPropertyValue = require('unicode-match-property-value-ecmascript');
|
||
const iuMappings = require('./data/iu-mappings.js');
|
||
const ESCAPE_SETS = require('./data/character-class-escape-sets.js');
|
||
|
||
function flatMap(array, callback) {
|
||
const result = [];
|
||
array.forEach(item => {
|
||
const res = callback(item);
|
||
if (Array.isArray(res)) {
|
||
result.push.apply(result, res);
|
||
} else {
|
||
result.push(res);
|
||
}
|
||
});
|
||
return result;
|
||
}
|
||
|
||
const SPECIAL_CHARS = /([\\^$.*+?()[\]{}|])/g;
|
||
|
||
// Prepare a Regenerate set containing all code points, used for negative
|
||
// character classes (if any).
|
||
const UNICODE_SET = regenerate().addRange(0x0, 0x10FFFF);
|
||
|
||
const ASTRAL_SET = regenerate().addRange(0x10000, 0x10FFFF);
|
||
|
||
const NEWLINE_SET = regenerate().add(
|
||
// `LineTerminator`s (https://mths.be/es6#sec-line-terminators):
|
||
0x000A, // Line Feed <LF>
|
||
0x000D, // Carriage Return <CR>
|
||
0x2028, // Line Separator <LS>
|
||
0x2029 // Paragraph Separator <PS>
|
||
);
|
||
|
||
// Prepare a Regenerate set containing all code points that are supposed to be
|
||
// matched by `/./u`. https://mths.be/es6#sec-atom
|
||
const DOT_SET_UNICODE = UNICODE_SET.clone() // all Unicode code points
|
||
.remove(NEWLINE_SET);
|
||
|
||
const getCharacterClassEscapeSet = (character, unicode, ignoreCase) => {
|
||
if (unicode) {
|
||
if (ignoreCase) {
|
||
return ESCAPE_SETS.UNICODE_IGNORE_CASE.get(character);
|
||
}
|
||
return ESCAPE_SETS.UNICODE.get(character);
|
||
}
|
||
return ESCAPE_SETS.REGULAR.get(character);
|
||
};
|
||
|
||
const getUnicodeDotSet = (dotAll) => {
|
||
return dotAll ? UNICODE_SET : DOT_SET_UNICODE;
|
||
};
|
||
|
||
const getUnicodePropertyValueSet = (property, value) => {
|
||
const path = value ?
|
||
`${ property }/${ value }` :
|
||
`Binary_Property/${ property }`;
|
||
try {
|
||
return require(`regenerate-unicode-properties/${ path }.js`);
|
||
} catch (exception) {
|
||
throw new Error(
|
||
`Failed to recognize value \`${ value }\` for property ` +
|
||
`\`${ property }\`.`
|
||
);
|
||
}
|
||
};
|
||
|
||
const handleLoneUnicodePropertyNameOrValue = (value) => {
|
||
// It could be a `General_Category` value or a binary property.
|
||
// Note: `unicodeMatchPropertyValue` throws on invalid values.
|
||
try {
|
||
const property = 'General_Category';
|
||
const category = unicodeMatchPropertyValue(property, value);
|
||
return getUnicodePropertyValueSet(property, category);
|
||
} catch (exception) {}
|
||
// It’s not a `General_Category` value, so check if it’s a property
|
||
// of strings.
|
||
try {
|
||
return getUnicodePropertyValueSet('Property_of_Strings', value);
|
||
} catch (exception) {}
|
||
// Lastly, check if it’s a binary property of single code points.
|
||
// Note: `unicodeMatchProperty` throws on invalid properties.
|
||
const property = unicodeMatchProperty(value);
|
||
return getUnicodePropertyValueSet(property);
|
||
};
|
||
|
||
const getUnicodePropertyEscapeSet = (value, isNegative) => {
|
||
const parts = value.split('=');
|
||
const firstPart = parts[0];
|
||
let set;
|
||
if (parts.length == 1) {
|
||
set = handleLoneUnicodePropertyNameOrValue(firstPart);
|
||
} else {
|
||
// The pattern consists of two parts, i.e. `Property=Value`.
|
||
const property = unicodeMatchProperty(firstPart);
|
||
const value = unicodeMatchPropertyValue(property, parts[1]);
|
||
set = getUnicodePropertyValueSet(property, value);
|
||
}
|
||
if (isNegative) {
|
||
if (set.strings) {
|
||
throw new Error('Cannot negate Unicode property of strings');
|
||
}
|
||
return {
|
||
characters: UNICODE_SET.clone().remove(set.characters),
|
||
strings: new Set()
|
||
};
|
||
}
|
||
return {
|
||
characters: set.characters.clone(),
|
||
strings: set.strings
|
||
// We need to escape strings like *️⃣ to make sure that they can be safely used in unions.
|
||
? new Set(set.strings.map(str => str.replace(SPECIAL_CHARS, '\\$1')))
|
||
: new Set()
|
||
};
|
||
};
|
||
|
||
const getUnicodePropertyEscapeCharacterClassData = (property, isNegative) => {
|
||
const set = getUnicodePropertyEscapeSet(property, isNegative);
|
||
const data = getCharacterClassEmptyData();
|
||
data.singleChars = set.characters;
|
||
if (set.strings.size > 0) {
|
||
data.longStrings = set.strings;
|
||
data.maybeIncludesStrings = true;
|
||
}
|
||
return data;
|
||
};
|
||
|
||
function configNeedCaseFoldAscii() {
|
||
return !!config.modifiersData.i;
|
||
}
|
||
|
||
function configNeedCaseFoldUnicode() {
|
||
// config.modifiersData.i : undefined | false
|
||
if (config.modifiersData.i === false) return false;
|
||
if (!config.transform.unicodeFlag) return false;
|
||
return Boolean(config.modifiersData.i || config.flags.ignoreCase);
|
||
}
|
||
|
||
// Given a range of code points, add any case-folded code points in that range
|
||
// to a set.
|
||
regenerate.prototype.iuAddRange = function(min, max) {
|
||
const $this = this;
|
||
do {
|
||
const folded = caseFold(min, configNeedCaseFoldAscii(), configNeedCaseFoldUnicode());
|
||
if (folded) {
|
||
$this.add(folded);
|
||
}
|
||
} while (++min <= max);
|
||
return $this;
|
||
};
|
||
regenerate.prototype.iuRemoveRange = function(min, max) {
|
||
const $this = this;
|
||
do {
|
||
const folded = caseFold(min, configNeedCaseFoldAscii(), configNeedCaseFoldUnicode());
|
||
if (folded) {
|
||
$this.remove(folded);
|
||
}
|
||
} while (++min <= max);
|
||
return $this;
|
||
};
|
||
|
||
const update = (item, pattern) => {
|
||
let tree = parse(pattern, config.useUnicodeFlag ? 'u' : '', {
|
||
lookbehind: true,
|
||
namedGroups: true,
|
||
unicodePropertyEscape: true,
|
||
unicodeSet: true,
|
||
modifiers: true,
|
||
});
|
||
switch (tree.type) {
|
||
case 'characterClass':
|
||
case 'group':
|
||
case 'value':
|
||
// No wrapping needed.
|
||
break;
|
||
default:
|
||
// Wrap the pattern in a non-capturing group.
|
||
tree = wrap(tree, pattern);
|
||
}
|
||
Object.assign(item, tree);
|
||
};
|
||
|
||
const wrap = (tree, pattern) => {
|
||
// Wrap the pattern in a non-capturing group.
|
||
return {
|
||
'type': 'group',
|
||
'behavior': 'ignore',
|
||
'body': [tree],
|
||
'raw': `(?:${ pattern })`
|
||
};
|
||
};
|
||
|
||
const caseFold = (codePoint, includeAscii, includeUnicode) => {
|
||
let folded = (includeUnicode ? iuMappings.get(codePoint) : undefined) || [];
|
||
if (typeof folded === 'number') folded = [folded];
|
||
if (includeAscii) {
|
||
if (codePoint >= 0x41 && codePoint <= 0x5A) {
|
||
folded.push(codePoint + 0x20);
|
||
} else if (codePoint >= 0x61 && codePoint <= 0x7A) {
|
||
folded.push(codePoint - 0x20);
|
||
}
|
||
}
|
||
return folded.length == 0 ? false : folded;
|
||
};
|
||
|
||
const buildHandler = (action) => {
|
||
switch (action) {
|
||
case 'union':
|
||
return {
|
||
single: (data, cp) => {
|
||
data.singleChars.add(cp);
|
||
},
|
||
regSet: (data, set2) => {
|
||
data.singleChars.add(set2);
|
||
},
|
||
range: (data, start, end) => {
|
||
data.singleChars.addRange(start, end);
|
||
},
|
||
iuRange: (data, start, end) => {
|
||
data.singleChars.iuAddRange(start, end);
|
||
},
|
||
nested: (data, nestedData) => {
|
||
data.singleChars.add(nestedData.singleChars);
|
||
for (const str of nestedData.longStrings) data.longStrings.add(str);
|
||
if (nestedData.maybeIncludesStrings) data.maybeIncludesStrings = true;
|
||
}
|
||
};
|
||
case 'union-negative': {
|
||
const regSet = (data, set2) => {
|
||
data.singleChars = UNICODE_SET.clone().remove(set2).add(data.singleChars);
|
||
};
|
||
return {
|
||
single: (data, cp) => {
|
||
const unicode = UNICODE_SET.clone();
|
||
data.singleChars = data.singleChars.contains(cp) ? unicode : unicode.remove(cp);
|
||
},
|
||
regSet: regSet,
|
||
range: (data, start, end) => {
|
||
data.singleChars = UNICODE_SET.clone().removeRange(start, end).add(data.singleChars);
|
||
},
|
||
iuRange: (data, start, end) => {
|
||
data.singleChars = UNICODE_SET.clone().iuRemoveRange(start, end).add(data.singleChars);
|
||
},
|
||
nested: (data, nestedData) => {
|
||
regSet(data, nestedData.singleChars);
|
||
if (nestedData.maybeIncludesStrings) throw new Error('ASSERTION ERROR');
|
||
}
|
||
};
|
||
}
|
||
case 'intersection': {
|
||
const regSet = (data, set2) => {
|
||
if (data.first) data.singleChars = set2;
|
||
else data.singleChars.intersection(set2);
|
||
};
|
||
return {
|
||
single: (data, cp) => {
|
||
data.singleChars = data.first || data.singleChars.contains(cp) ? regenerate(cp) : regenerate();
|
||
data.longStrings.clear();
|
||
data.maybeIncludesStrings = false;
|
||
},
|
||
regSet: (data, set) => {
|
||
regSet(data, set);
|
||
data.longStrings.clear();
|
||
data.maybeIncludesStrings = false;
|
||
},
|
||
range: (data, start, end) => {
|
||
if (data.first) data.singleChars.addRange(start, end);
|
||
else data.singleChars.intersection(regenerate().addRange(start, end));
|
||
data.longStrings.clear();
|
||
data.maybeIncludesStrings = false;
|
||
},
|
||
iuRange: (data, start, end) => {
|
||
if (data.first) data.singleChars.iuAddRange(start, end);
|
||
else data.singleChars.intersection(regenerate().iuAddRange(start, end));
|
||
data.longStrings.clear();
|
||
data.maybeIncludesStrings = false;
|
||
},
|
||
nested: (data, nestedData) => {
|
||
regSet(data, nestedData.singleChars);
|
||
|
||
if (data.first) {
|
||
data.longStrings = nestedData.longStrings;
|
||
data.maybeIncludesStrings = nestedData.maybeIncludesStrings;
|
||
} else {
|
||
for (const str of data.longStrings) {
|
||
if (!nestedData.longStrings.has(str)) data.longStrings.delete(str);
|
||
}
|
||
if (!nestedData.maybeIncludesStrings) data.maybeIncludesStrings = false;
|
||
}
|
||
}
|
||
};
|
||
}
|
||
case 'subtraction': {
|
||
const regSet = (data, set2) => {
|
||
if (data.first) data.singleChars.add(set2);
|
||
else data.singleChars.remove(set2);
|
||
};
|
||
return {
|
||
single: (data, cp) => {
|
||
if (data.first) data.singleChars.add(cp);
|
||
else data.singleChars.remove(cp);
|
||
},
|
||
regSet: regSet,
|
||
range: (data, start, end) => {
|
||
if (data.first) data.singleChars.addRange(start, end);
|
||
else data.singleChars.removeRange(start, end);
|
||
},
|
||
iuRange: (data, start, end) => {
|
||
if (data.first) data.singleChars.iuAddRange(start, end);
|
||
else data.singleChars.iuRemoveRange(start, end);
|
||
},
|
||
nested: (data, nestedData) => {
|
||
regSet(data, nestedData.singleChars);
|
||
|
||
if (data.first) {
|
||
data.longStrings = nestedData.longStrings;
|
||
data.maybeIncludesStrings = nestedData.maybeIncludesStrings;
|
||
} else {
|
||
for (const str of data.longStrings) {
|
||
if (nestedData.longStrings.has(str)) data.longStrings.delete(str);
|
||
}
|
||
}
|
||
}
|
||
};
|
||
}
|
||
// The `default` clause is only here as a safeguard; it should never be
|
||
// reached. Code coverage tools should ignore it.
|
||
/* istanbul ignore next */
|
||
default:
|
||
throw new Error(`Unknown set action: ${ characterClassItem.kind }`);
|
||
}
|
||
};
|
||
|
||
const getCharacterClassEmptyData = () => ({
|
||
transformed: config.transform.unicodeFlag,
|
||
singleChars: regenerate(),
|
||
longStrings: new Set(),
|
||
hasEmptyString: false,
|
||
first: true,
|
||
maybeIncludesStrings: false
|
||
});
|
||
|
||
const maybeFold = (codePoint) => {
|
||
const caseFoldAscii = configNeedCaseFoldAscii();
|
||
const caseFoldUnicode = configNeedCaseFoldUnicode();
|
||
|
||
if (caseFoldAscii || caseFoldUnicode) {
|
||
const folded = caseFold(codePoint, caseFoldAscii, caseFoldUnicode);
|
||
if (folded) {
|
||
return [codePoint, folded];
|
||
}
|
||
}
|
||
return [codePoint];
|
||
};
|
||
|
||
const computeClassStrings = (classStrings, regenerateOptions) => {
|
||
let data = getCharacterClassEmptyData();
|
||
|
||
const caseFoldAscii = configNeedCaseFoldAscii();
|
||
const caseFoldUnicode = configNeedCaseFoldUnicode();
|
||
|
||
for (const string of classStrings.strings) {
|
||
if (string.characters.length === 1) {
|
||
maybeFold(string.characters[0].codePoint).forEach((cp) => {
|
||
data.singleChars.add(cp);
|
||
});
|
||
} else {
|
||
let stringifiedString;
|
||
if (caseFoldUnicode || caseFoldAscii) {
|
||
stringifiedString = '';
|
||
for (const ch of string.characters) {
|
||
let set = regenerate(ch.codePoint);
|
||
const folded = maybeFold(ch.codePoint);
|
||
if (folded) set.add(folded);
|
||
stringifiedString += set.toString(regenerateOptions);
|
||
}
|
||
} else {
|
||
stringifiedString = string.characters.map(ch => generate(ch)).join('')
|
||
}
|
||
|
||
data.longStrings.add(stringifiedString);
|
||
data.maybeIncludesStrings = true;
|
||
}
|
||
}
|
||
|
||
return data;
|
||
}
|
||
|
||
const computeCharacterClass = (characterClassItem, regenerateOptions) => {
|
||
let data = getCharacterClassEmptyData();
|
||
|
||
let handlePositive;
|
||
let handleNegative;
|
||
|
||
switch (characterClassItem.kind) {
|
||
case 'union':
|
||
handlePositive = buildHandler('union');
|
||
handleNegative = buildHandler('union-negative');
|
||
break;
|
||
case 'intersection':
|
||
handlePositive = buildHandler('intersection');
|
||
handleNegative = buildHandler('subtraction');
|
||
if (config.transform.unicodeSetsFlag) data.transformed = true;
|
||
break;
|
||
case 'subtraction':
|
||
handlePositive = buildHandler('subtraction');
|
||
handleNegative = buildHandler('intersection');
|
||
if (config.transform.unicodeSetsFlag) data.transformed = true;
|
||
break;
|
||
// The `default` clause is only here as a safeguard; it should never be
|
||
// reached. Code coverage tools should ignore it.
|
||
/* istanbul ignore next */
|
||
default:
|
||
throw new Error(`Unknown character class kind: ${ characterClassItem.kind }`);
|
||
}
|
||
|
||
const caseFoldAscii = configNeedCaseFoldAscii();
|
||
const caseFoldUnicode = configNeedCaseFoldUnicode();
|
||
|
||
for (const item of characterClassItem.body) {
|
||
switch (item.type) {
|
||
case 'value':
|
||
maybeFold(item.codePoint).forEach((cp) => {
|
||
handlePositive.single(data, cp);
|
||
});
|
||
break;
|
||
case 'characterClassRange':
|
||
const min = item.min.codePoint;
|
||
const max = item.max.codePoint;
|
||
handlePositive.range(data, min, max);
|
||
if (caseFoldAscii || caseFoldUnicode) {
|
||
handlePositive.iuRange(data, min, max);
|
||
data.transformed = true;
|
||
}
|
||
break;
|
||
case 'characterClassEscape':
|
||
handlePositive.regSet(data, getCharacterClassEscapeSet(
|
||
item.value,
|
||
config.flags.unicode,
|
||
config.flags.ignoreCase
|
||
));
|
||
break;
|
||
case 'unicodePropertyEscape':
|
||
const nestedData = getUnicodePropertyEscapeCharacterClassData(item.value, item.negative);
|
||
handlePositive.nested(data, nestedData);
|
||
data.transformed =
|
||
data.transformed ||
|
||
config.transform.unicodePropertyEscapes ||
|
||
(config.transform.unicodeSetsFlag && nestedData.maybeIncludesStrings);
|
||
break;
|
||
case 'characterClass':
|
||
const handler = item.negative ? handleNegative : handlePositive;
|
||
const res = computeCharacterClass(item, regenerateOptions);
|
||
handler.nested(data, res);
|
||
data.transformed = true;
|
||
break;
|
||
case 'classStrings':
|
||
handlePositive.nested(data, computeClassStrings(item, regenerateOptions));
|
||
data.transformed = true;
|
||
break;
|
||
// The `default` clause is only here as a safeguard; it should never be
|
||
// reached. Code coverage tools should ignore it.
|
||
/* istanbul ignore next */
|
||
default:
|
||
throw new Error(`Unknown term type: ${ item.type }`);
|
||
}
|
||
|
||
data.first = false;
|
||
}
|
||
|
||
if (characterClassItem.negative && data.maybeIncludesStrings) {
|
||
throw new SyntaxError('Cannot negate set containing strings');
|
||
}
|
||
|
||
return data;
|
||
}
|
||
|
||
const processCharacterClass = (
|
||
characterClassItem,
|
||
regenerateOptions,
|
||
computed = computeCharacterClass(characterClassItem, regenerateOptions)
|
||
) => {
|
||
const negative = characterClassItem.negative;
|
||
const { singleChars, transformed, longStrings } = computed;
|
||
if (transformed) {
|
||
const setStr = singleChars.toString(regenerateOptions);
|
||
|
||
if (negative) {
|
||
if (config.useUnicodeFlag) {
|
||
update(characterClassItem, `[^${setStr[0] === '[' ? setStr.slice(1, -1) : setStr}]`)
|
||
} else {
|
||
if (config.flags.unicode) {
|
||
if (config.flags.ignoreCase) {
|
||
const astralCharsSet = singleChars.clone().intersection(ASTRAL_SET);
|
||
// Assumption: singleChars do not contain lone surrogates.
|
||
// Regex like /[^\ud800]/u is not supported
|
||
const surrogateOrBMPSetStr = singleChars
|
||
.clone()
|
||
.remove(astralCharsSet)
|
||
.addRange(0xd800, 0xdfff)
|
||
.toString({ bmpOnly: true });
|
||
// Don't generate negative lookahead for astral characters
|
||
// because the case folding is not working anyway as we break
|
||
// code points into surrogate pairs.
|
||
const astralNegativeSetStr = ASTRAL_SET
|
||
.clone()
|
||
.remove(astralCharsSet)
|
||
.toString(regenerateOptions);
|
||
// The transform here does not support lone surrogates.
|
||
update(
|
||
characterClassItem,
|
||
`(?!${surrogateOrBMPSetStr})[\\s\\S]|${astralNegativeSetStr}`
|
||
);
|
||
} else {
|
||
// Generate negative set directly when case folding is not involved.
|
||
update(
|
||
characterClassItem,
|
||
UNICODE_SET.clone().remove(singleChars).toString(regenerateOptions)
|
||
);
|
||
}
|
||
} else {
|
||
update(characterClassItem, `(?!${setStr})[\\s\\S]`);
|
||
}
|
||
}
|
||
} else {
|
||
const hasEmptyString = longStrings.has('');
|
||
const pieces = Array.from(longStrings).sort((a, b) => b.length - a.length);
|
||
|
||
if (setStr !== '[]' || longStrings.size === 0) {
|
||
pieces.splice(pieces.length - (hasEmptyString ? 1 : 0), 0, setStr);
|
||
}
|
||
|
||
update(characterClassItem, pieces.join('|'));
|
||
}
|
||
}
|
||
return characterClassItem;
|
||
};
|
||
|
||
const assertNoUnmatchedReferences = (groups) => {
|
||
const unmatchedReferencesNames = Object.keys(groups.unmatchedReferences);
|
||
if (unmatchedReferencesNames.length > 0) {
|
||
throw new Error(`Unknown group names: ${unmatchedReferencesNames}`);
|
||
}
|
||
};
|
||
|
||
const processModifiers = (item, regenerateOptions, groups) => {
|
||
const enabling = item.modifierFlags.enabling;
|
||
const disabling = item.modifierFlags.disabling;
|
||
|
||
delete item.modifierFlags;
|
||
item.behavior = 'ignore';
|
||
|
||
const oldData = Object.assign({}, config.modifiersData);
|
||
|
||
enabling.split('').forEach(flag => {
|
||
config.modifiersData[flag] = true;
|
||
});
|
||
disabling.split('').forEach(flag => {
|
||
config.modifiersData[flag] = false;
|
||
});
|
||
|
||
item.body = item.body.map(term => {
|
||
return processTerm(term, regenerateOptions, groups);
|
||
});
|
||
|
||
config.modifiersData = oldData;
|
||
|
||
return item;
|
||
}
|
||
|
||
const processTerm = (item, regenerateOptions, groups) => {
|
||
switch (item.type) {
|
||
case 'dot':
|
||
if (config.transform.unicodeFlag) {
|
||
update(
|
||
item,
|
||
getUnicodeDotSet(config.flags.dotAll || config.modifiersData.s).toString(regenerateOptions)
|
||
);
|
||
} else if (config.transform.dotAllFlag || config.modifiersData.s) {
|
||
// TODO: consider changing this at the regenerate level.
|
||
update(item, '[\\s\\S]');
|
||
}
|
||
break;
|
||
case 'characterClass':
|
||
item = processCharacterClass(item, regenerateOptions);
|
||
break;
|
||
case 'unicodePropertyEscape':
|
||
const data = getUnicodePropertyEscapeCharacterClassData(item.value, item.negative);
|
||
if (data.maybeIncludesStrings) {
|
||
if (!config.flags.unicodeSets) {
|
||
throw new Error(
|
||
'Properties of strings are only supported when using the unicodeSets (v) flag.'
|
||
);
|
||
}
|
||
if (config.transform.unicodeSetsFlag) {
|
||
data.transformed = true;
|
||
item = processCharacterClass(item, regenerateOptions, data);
|
||
}
|
||
} else if (config.transform.unicodePropertyEscapes) {
|
||
update(
|
||
item,
|
||
data.singleChars.toString(regenerateOptions)
|
||
);
|
||
}
|
||
break;
|
||
case 'characterClassEscape':
|
||
if (config.transform.unicodeFlag) {
|
||
update(
|
||
item,
|
||
getCharacterClassEscapeSet(
|
||
item.value,
|
||
/* config.transform.unicodeFlag implies config.flags.unicode */ true,
|
||
config.flags.ignoreCase
|
||
).toString(regenerateOptions)
|
||
);
|
||
}
|
||
break;
|
||
case 'group':
|
||
if (item.behavior == 'normal') {
|
||
groups.lastIndex++;
|
||
}
|
||
if (item.name) {
|
||
const name = item.name.value;
|
||
|
||
if (groups.namesConflicts[name]) {
|
||
throw new Error(
|
||
`Group '${ name }' has already been defined in this context.`
|
||
);
|
||
}
|
||
groups.namesConflicts[name] = true;
|
||
|
||
if (config.transform.namedGroups) {
|
||
delete item.name;
|
||
}
|
||
|
||
const index = groups.lastIndex;
|
||
if (!groups.names[name]) {
|
||
groups.names[name] = [];
|
||
}
|
||
groups.names[name].push(index);
|
||
|
||
if (groups.onNamedGroup) {
|
||
groups.onNamedGroup.call(null, name, index);
|
||
}
|
||
|
||
if (groups.unmatchedReferences[name]) {
|
||
delete groups.unmatchedReferences[name];
|
||
}
|
||
}
|
||
if (item.modifierFlags && config.transform.modifiers) {
|
||
return processModifiers(item, regenerateOptions, groups);
|
||
}
|
||
/* falls through */
|
||
case 'quantifier':
|
||
item.body = item.body.map(term => {
|
||
return processTerm(term, regenerateOptions, groups);
|
||
});
|
||
break;
|
||
case 'disjunction':
|
||
const outerNamesConflicts = groups.namesConflicts;
|
||
item.body = item.body.map(term => {
|
||
groups.namesConflicts = Object.create(outerNamesConflicts);
|
||
return processTerm(term, regenerateOptions, groups);
|
||
});
|
||
break;
|
||
case 'alternative':
|
||
item.body = flatMap(item.body, term => {
|
||
const res = processTerm(term, regenerateOptions, groups);
|
||
// Alternatives cannot contain alternatives; flatten them.
|
||
return res.type === 'alternative' ? res.body : res;
|
||
});
|
||
break;
|
||
case 'value':
|
||
const codePoint = item.codePoint;
|
||
const set = regenerate(codePoint);
|
||
const folded = maybeFold(codePoint);
|
||
set.add(folded);
|
||
update(item, set.toString(regenerateOptions));
|
||
break;
|
||
case 'reference':
|
||
if (item.name) {
|
||
const name = item.name.value;
|
||
const indexes = groups.names[name];
|
||
if (!indexes) {
|
||
groups.unmatchedReferences[name] = true;
|
||
}
|
||
|
||
if (config.transform.namedGroups) {
|
||
if (indexes) {
|
||
const body = indexes.map(index => ({
|
||
'type': 'reference',
|
||
'matchIndex': index,
|
||
'raw': '\\' + index,
|
||
}));
|
||
if (body.length === 1) {
|
||
return body[0];
|
||
}
|
||
return {
|
||
'type': 'alternative',
|
||
'body': body,
|
||
'raw': body.map(term => term.raw).join(''),
|
||
};
|
||
}
|
||
|
||
// This named reference comes before the group where it’s defined,
|
||
// so it’s always an empty match.
|
||
return {
|
||
'type': 'group',
|
||
'behavior': 'ignore',
|
||
'body': [],
|
||
'raw': '(?:)',
|
||
};
|
||
}
|
||
}
|
||
break;
|
||
case 'anchor':
|
||
if (config.modifiersData.m) {
|
||
if (item.kind == 'start') {
|
||
update(item, `(?:^|(?<=${NEWLINE_SET.toString()}))`);
|
||
} else if (item.kind == 'end') {
|
||
update(item, `(?:$|(?=${NEWLINE_SET.toString()}))`);
|
||
}
|
||
}
|
||
case 'empty':
|
||
// Nothing to do here.
|
||
break;
|
||
// The `default` clause is only here as a safeguard; it should never be
|
||
// reached. Code coverage tools should ignore it.
|
||
/* istanbul ignore next */
|
||
default:
|
||
throw new Error(`Unknown term type: ${ item.type }`);
|
||
}
|
||
return item;
|
||
};
|
||
|
||
const config = {
|
||
'flags': {
|
||
'ignoreCase': false,
|
||
'unicode': false,
|
||
'unicodeSets': false,
|
||
'dotAll': false,
|
||
'multiline': false,
|
||
},
|
||
'transform': {
|
||
'dotAllFlag': false,
|
||
'unicodeFlag': false,
|
||
'unicodeSetsFlag': false,
|
||
'unicodePropertyEscapes': false,
|
||
'namedGroups': false,
|
||
'modifiers': false,
|
||
},
|
||
'modifiersData': {
|
||
'i': undefined,
|
||
's': undefined,
|
||
'm': undefined,
|
||
},
|
||
get useUnicodeFlag() {
|
||
return (this.flags.unicode || this.flags.unicodeSets) && !this.transform.unicodeFlag;
|
||
}
|
||
};
|
||
|
||
const validateOptions = (options) => {
|
||
if (!options) return;
|
||
|
||
for (const key of Object.keys(options)) {
|
||
const value = options[key];
|
||
switch (key) {
|
||
case 'dotAllFlag':
|
||
case 'unicodeFlag':
|
||
case 'unicodePropertyEscapes':
|
||
case 'namedGroups':
|
||
if (value != null && value !== false && value !== 'transform') {
|
||
throw new Error(`.${key} must be false (default) or 'transform'.`);
|
||
}
|
||
break;
|
||
case 'modifiers':
|
||
case 'unicodeSetsFlag':
|
||
if (value != null && value !== false && value !== 'parse' && value !== 'transform') {
|
||
throw new Error(`.${key} must be false (default), 'parse' or 'transform'.`);
|
||
}
|
||
break;
|
||
case 'onNamedGroup':
|
||
case 'onNewFlags':
|
||
if (value != null && typeof value !== 'function') {
|
||
throw new Error(`.${key} must be a function.`);
|
||
}
|
||
break;
|
||
default:
|
||
throw new Error(`.${key} is not a valid regexpu-core option.`);
|
||
}
|
||
}
|
||
};
|
||
|
||
const hasFlag = (flags, flag) => flags ? flags.includes(flag) : false;
|
||
const transform = (options, name) => options ? options[name] === 'transform' : false;
|
||
|
||
const rewritePattern = (pattern, flags, options) => {
|
||
validateOptions(options);
|
||
|
||
config.flags.unicode = hasFlag(flags, 'u');
|
||
config.flags.unicodeSets = hasFlag(flags, 'v');
|
||
config.flags.ignoreCase = hasFlag(flags, 'i');
|
||
config.flags.dotAll = hasFlag(flags, 's');
|
||
config.flags.multiline = hasFlag(flags, 'm');
|
||
|
||
config.transform.dotAllFlag = config.flags.dotAll && transform(options, 'dotAllFlag');
|
||
config.transform.unicodeFlag = (config.flags.unicode || config.flags.unicodeSets) && transform(options, 'unicodeFlag');
|
||
config.transform.unicodeSetsFlag = config.flags.unicodeSets && transform(options, 'unicodeSetsFlag');
|
||
|
||
// unicodeFlag: 'transform' implies unicodePropertyEscapes: 'transform'
|
||
config.transform.unicodePropertyEscapes = config.flags.unicode && (
|
||
transform(options, 'unicodeFlag') || transform(options, 'unicodePropertyEscapes')
|
||
);
|
||
config.transform.namedGroups = transform(options, 'namedGroups');
|
||
config.transform.modifiers = transform(options, 'modifiers');
|
||
|
||
config.modifiersData.i = undefined;
|
||
config.modifiersData.s = undefined;
|
||
config.modifiersData.m = undefined;
|
||
|
||
const regjsparserFeatures = {
|
||
'unicodeSet': Boolean(options && options.unicodeSetsFlag),
|
||
'modifiers': Boolean(options && options.modifiers),
|
||
|
||
// Enable every stable RegExp feature by default
|
||
'unicodePropertyEscape': true,
|
||
'namedGroups': true,
|
||
'lookbehind': true,
|
||
};
|
||
|
||
const regenerateOptions = {
|
||
'hasUnicodeFlag': config.useUnicodeFlag,
|
||
'bmpOnly': !config.flags.unicode
|
||
};
|
||
|
||
const groups = {
|
||
'onNamedGroup': options && options.onNamedGroup,
|
||
'lastIndex': 0,
|
||
'names': Object.create(null), // { [name]: Array<index> }
|
||
'namesConflicts': Object.create(null), // { [name]: true }
|
||
'unmatchedReferences': Object.create(null) // { [name]: true }
|
||
};
|
||
|
||
const tree = parse(pattern, flags, regjsparserFeatures);
|
||
|
||
if (config.transform.modifiers) {
|
||
if (/\(\?[a-z]*-[a-z]+:/.test(pattern)) {
|
||
// the pattern _likely_ contain inline disabled modifiers
|
||
// we need to traverse to make sure that they are actually modifiers and to collect them
|
||
const allDisabledModifiers = Object.create(null)
|
||
const itemStack = [tree];
|
||
let node;
|
||
while (node = itemStack.pop(), node != undefined) {
|
||
if (Array.isArray(node)) {
|
||
Array.prototype.push.apply(itemStack, node);
|
||
} else if (typeof node == 'object' && node != null) {
|
||
for (const key of Object.keys(node)) {
|
||
const value = node[key];
|
||
if (key == 'modifierFlags') {
|
||
if (value.disabling.length > 0){
|
||
value.disabling.split('').forEach((flag)=>{
|
||
allDisabledModifiers[flag] = true
|
||
});
|
||
}
|
||
} else if (typeof value == 'object' && value != null) {
|
||
itemStack.push(value);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
for (const flag of Object.keys(allDisabledModifiers)) {
|
||
config.modifiersData[flag] = true;
|
||
}
|
||
}
|
||
}
|
||
|
||
// Note: `processTerm` mutates `tree` and `groups`.
|
||
processTerm(tree, regenerateOptions, groups);
|
||
assertNoUnmatchedReferences(groups);
|
||
|
||
const onNewFlags = options && options.onNewFlags;
|
||
if (onNewFlags) {
|
||
let newFlags = flags.split('').filter((flag) => !config.modifiersData[flag]).join('');
|
||
if (config.transform.unicodeSetsFlag) {
|
||
newFlags = newFlags.replace('v', 'u');
|
||
}
|
||
if (config.transform.unicodeFlag) {
|
||
newFlags = newFlags.replace('u', '');
|
||
}
|
||
if (config.transform.dotAllFlag === 'transform') {
|
||
newFlags = newFlags.replace('s', '');
|
||
}
|
||
onNewFlags(newFlags);
|
||
}
|
||
|
||
return generate(tree);
|
||
};
|
||
|
||
module.exports = rewritePattern;
|