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.
560 lines
24 KiB
560 lines
24 KiB
<?php declare(strict_types=1);
|
|
|
|
namespace PhpParser;
|
|
|
|
use PhpParser\Parser\Tokens;
|
|
|
|
class Lexer
|
|
{
|
|
protected $code;
|
|
protected $tokens;
|
|
protected $pos;
|
|
protected $line;
|
|
protected $filePos;
|
|
protected $prevCloseTagHasNewline;
|
|
|
|
protected $tokenMap;
|
|
protected $dropTokens;
|
|
protected $identifierTokens;
|
|
|
|
private $attributeStartLineUsed;
|
|
private $attributeEndLineUsed;
|
|
private $attributeStartTokenPosUsed;
|
|
private $attributeEndTokenPosUsed;
|
|
private $attributeStartFilePosUsed;
|
|
private $attributeEndFilePosUsed;
|
|
private $attributeCommentsUsed;
|
|
|
|
/**
|
|
* Creates a Lexer.
|
|
*
|
|
* @param array $options Options array. Currently only the 'usedAttributes' option is supported,
|
|
* which is an array of attributes to add to the AST nodes. Possible
|
|
* attributes are: 'comments', 'startLine', 'endLine', 'startTokenPos',
|
|
* 'endTokenPos', 'startFilePos', 'endFilePos'. The option defaults to the
|
|
* first three. For more info see getNextToken() docs.
|
|
*/
|
|
public function __construct(array $options = []) {
|
|
// Create Map from internal tokens to PhpParser tokens.
|
|
$this->defineCompatibilityTokens();
|
|
$this->tokenMap = $this->createTokenMap();
|
|
$this->identifierTokens = $this->createIdentifierTokenMap();
|
|
|
|
// map of tokens to drop while lexing (the map is only used for isset lookup,
|
|
// that's why the value is simply set to 1; the value is never actually used.)
|
|
$this->dropTokens = array_fill_keys(
|
|
[\T_WHITESPACE, \T_OPEN_TAG, \T_COMMENT, \T_DOC_COMMENT, \T_BAD_CHARACTER], 1
|
|
);
|
|
|
|
$defaultAttributes = ['comments', 'startLine', 'endLine'];
|
|
$usedAttributes = array_fill_keys($options['usedAttributes'] ?? $defaultAttributes, true);
|
|
|
|
// Create individual boolean properties to make these checks faster.
|
|
$this->attributeStartLineUsed = isset($usedAttributes['startLine']);
|
|
$this->attributeEndLineUsed = isset($usedAttributes['endLine']);
|
|
$this->attributeStartTokenPosUsed = isset($usedAttributes['startTokenPos']);
|
|
$this->attributeEndTokenPosUsed = isset($usedAttributes['endTokenPos']);
|
|
$this->attributeStartFilePosUsed = isset($usedAttributes['startFilePos']);
|
|
$this->attributeEndFilePosUsed = isset($usedAttributes['endFilePos']);
|
|
$this->attributeCommentsUsed = isset($usedAttributes['comments']);
|
|
}
|
|
|
|
/**
|
|
* Initializes the lexer for lexing the provided source code.
|
|
*
|
|
* This function does not throw if lexing errors occur. Instead, errors may be retrieved using
|
|
* the getErrors() method.
|
|
*
|
|
* @param string $code The source code to lex
|
|
* @param ErrorHandler|null $errorHandler Error handler to use for lexing errors. Defaults to
|
|
* ErrorHandler\Throwing
|
|
*/
|
|
public function startLexing(string $code, ErrorHandler $errorHandler = null) {
|
|
if (null === $errorHandler) {
|
|
$errorHandler = new ErrorHandler\Throwing();
|
|
}
|
|
|
|
$this->code = $code; // keep the code around for __halt_compiler() handling
|
|
$this->pos = -1;
|
|
$this->line = 1;
|
|
$this->filePos = 0;
|
|
|
|
// If inline HTML occurs without preceding code, treat it as if it had a leading newline.
|
|
// This ensures proper composability, because having a newline is the "safe" assumption.
|
|
$this->prevCloseTagHasNewline = true;
|
|
|
|
$scream = ini_set('xdebug.scream', '0');
|
|
|
|
$this->tokens = @token_get_all($code);
|
|
$this->postprocessTokens($errorHandler);
|
|
|
|
if (false !== $scream) {
|
|
ini_set('xdebug.scream', $scream);
|
|
}
|
|
}
|
|
|
|
private function handleInvalidCharacterRange($start, $end, $line, ErrorHandler $errorHandler) {
|
|
$tokens = [];
|
|
for ($i = $start; $i < $end; $i++) {
|
|
$chr = $this->code[$i];
|
|
if ($chr === "\0") {
|
|
// PHP cuts error message after null byte, so need special case
|
|
$errorMsg = 'Unexpected null byte';
|
|
} else {
|
|
$errorMsg = sprintf(
|
|
'Unexpected character "%s" (ASCII %d)', $chr, ord($chr)
|
|
);
|
|
}
|
|
|
|
$tokens[] = [\T_BAD_CHARACTER, $chr, $line];
|
|
$errorHandler->handleError(new Error($errorMsg, [
|
|
'startLine' => $line,
|
|
'endLine' => $line,
|
|
'startFilePos' => $i,
|
|
'endFilePos' => $i,
|
|
]));
|
|
}
|
|
return $tokens;
|
|
}
|
|
|
|
/**
|
|
* Check whether comment token is unterminated.
|
|
*
|
|
* @return bool
|
|
*/
|
|
private function isUnterminatedComment($token) : bool {
|
|
return ($token[0] === \T_COMMENT || $token[0] === \T_DOC_COMMENT)
|
|
&& substr($token[1], 0, 2) === '/*'
|
|
&& substr($token[1], -2) !== '*/';
|
|
}
|
|
|
|
protected function postprocessTokens(ErrorHandler $errorHandler) {
|
|
// PHP's error handling for token_get_all() is rather bad, so if we want detailed
|
|
// error information we need to compute it ourselves. Invalid character errors are
|
|
// detected by finding "gaps" in the token array. Unterminated comments are detected
|
|
// by checking if a trailing comment has a "*/" at the end.
|
|
//
|
|
// Additionally, we perform a number of canonicalizations here:
|
|
// * Use the PHP 8.0 comment format, which does not include trailing whitespace anymore.
|
|
// * Use PHP 8.0 T_NAME_* tokens.
|
|
// * Use PHP 8.1 T_AMPERSAND_NOT_FOLLOWED_BY_VAR_OR_VARARG and
|
|
// T_AMPERSAND_FOLLOWED_BY_VAR_OR_VARARG tokens used to disambiguate intersection types.
|
|
|
|
$filePos = 0;
|
|
$line = 1;
|
|
$numTokens = \count($this->tokens);
|
|
for ($i = 0; $i < $numTokens; $i++) {
|
|
$token = $this->tokens[$i];
|
|
|
|
// Since PHP 7.4 invalid characters are represented by a T_BAD_CHARACTER token.
|
|
// In this case we only need to emit an error.
|
|
if ($token[0] === \T_BAD_CHARACTER) {
|
|
$this->handleInvalidCharacterRange($filePos, $filePos + 1, $line, $errorHandler);
|
|
}
|
|
|
|
if ($token[0] === \T_COMMENT && substr($token[1], 0, 2) !== '/*'
|
|
&& preg_match('/(\r\n|\n|\r)$/D', $token[1], $matches)) {
|
|
$trailingNewline = $matches[0];
|
|
$token[1] = substr($token[1], 0, -strlen($trailingNewline));
|
|
$this->tokens[$i] = $token;
|
|
if (isset($this->tokens[$i + 1]) && $this->tokens[$i + 1][0] === \T_WHITESPACE) {
|
|
// Move trailing newline into following T_WHITESPACE token, if it already exists.
|
|
$this->tokens[$i + 1][1] = $trailingNewline . $this->tokens[$i + 1][1];
|
|
$this->tokens[$i + 1][2]--;
|
|
} else {
|
|
// Otherwise, we need to create a new T_WHITESPACE token.
|
|
array_splice($this->tokens, $i + 1, 0, [
|
|
[\T_WHITESPACE, $trailingNewline, $line],
|
|
]);
|
|
$numTokens++;
|
|
}
|
|
}
|
|
|
|
// Emulate PHP 8 T_NAME_* tokens, by combining sequences of T_NS_SEPARATOR and T_STRING
|
|
// into a single token.
|
|
if (\is_array($token)
|
|
&& ($token[0] === \T_NS_SEPARATOR || isset($this->identifierTokens[$token[0]]))) {
|
|
$lastWasSeparator = $token[0] === \T_NS_SEPARATOR;
|
|
$text = $token[1];
|
|
for ($j = $i + 1; isset($this->tokens[$j]); $j++) {
|
|
if ($lastWasSeparator) {
|
|
if (!isset($this->identifierTokens[$this->tokens[$j][0]])) {
|
|
break;
|
|
}
|
|
$lastWasSeparator = false;
|
|
} else {
|
|
if ($this->tokens[$j][0] !== \T_NS_SEPARATOR) {
|
|
break;
|
|
}
|
|
$lastWasSeparator = true;
|
|
}
|
|
$text .= $this->tokens[$j][1];
|
|
}
|
|
if ($lastWasSeparator) {
|
|
// Trailing separator is not part of the name.
|
|
$j--;
|
|
$text = substr($text, 0, -1);
|
|
}
|
|
if ($j > $i + 1) {
|
|
if ($token[0] === \T_NS_SEPARATOR) {
|
|
$type = \T_NAME_FULLY_QUALIFIED;
|
|
} else if ($token[0] === \T_NAMESPACE) {
|
|
$type = \T_NAME_RELATIVE;
|
|
} else {
|
|
$type = \T_NAME_QUALIFIED;
|
|
}
|
|
$token = [$type, $text, $line];
|
|
array_splice($this->tokens, $i, $j - $i, [$token]);
|
|
$numTokens -= $j - $i - 1;
|
|
}
|
|
}
|
|
|
|
if ($token === '&') {
|
|
$next = $i + 1;
|
|
while (isset($this->tokens[$next]) && $this->tokens[$next][0] === \T_WHITESPACE) {
|
|
$next++;
|
|
}
|
|
$followedByVarOrVarArg = isset($this->tokens[$next]) &&
|
|
($this->tokens[$next][0] === \T_VARIABLE || $this->tokens[$next][0] === \T_ELLIPSIS);
|
|
$this->tokens[$i] = $token = [
|
|
$followedByVarOrVarArg
|
|
? \T_AMPERSAND_FOLLOWED_BY_VAR_OR_VARARG
|
|
: \T_AMPERSAND_NOT_FOLLOWED_BY_VAR_OR_VARARG,
|
|
'&',
|
|
$line,
|
|
];
|
|
}
|
|
|
|
$tokenValue = \is_string($token) ? $token : $token[1];
|
|
$tokenLen = \strlen($tokenValue);
|
|
|
|
if (substr($this->code, $filePos, $tokenLen) !== $tokenValue) {
|
|
// Something is missing, must be an invalid character
|
|
$nextFilePos = strpos($this->code, $tokenValue, $filePos);
|
|
$badCharTokens = $this->handleInvalidCharacterRange(
|
|
$filePos, $nextFilePos, $line, $errorHandler);
|
|
$filePos = (int) $nextFilePos;
|
|
|
|
array_splice($this->tokens, $i, 0, $badCharTokens);
|
|
$numTokens += \count($badCharTokens);
|
|
$i += \count($badCharTokens);
|
|
}
|
|
|
|
$filePos += $tokenLen;
|
|
$line += substr_count($tokenValue, "\n");
|
|
}
|
|
|
|
if ($filePos !== \strlen($this->code)) {
|
|
if (substr($this->code, $filePos, 2) === '/*') {
|
|
// Unlike PHP, HHVM will drop unterminated comments entirely
|
|
$comment = substr($this->code, $filePos);
|
|
$errorHandler->handleError(new Error('Unterminated comment', [
|
|
'startLine' => $line,
|
|
'endLine' => $line + substr_count($comment, "\n"),
|
|
'startFilePos' => $filePos,
|
|
'endFilePos' => $filePos + \strlen($comment),
|
|
]));
|
|
|
|
// Emulate the PHP behavior
|
|
$isDocComment = isset($comment[3]) && $comment[3] === '*';
|
|
$this->tokens[] = [$isDocComment ? \T_DOC_COMMENT : \T_COMMENT, $comment, $line];
|
|
} else {
|
|
// Invalid characters at the end of the input
|
|
$badCharTokens = $this->handleInvalidCharacterRange(
|
|
$filePos, \strlen($this->code), $line, $errorHandler);
|
|
$this->tokens = array_merge($this->tokens, $badCharTokens);
|
|
}
|
|
return;
|
|
}
|
|
|
|
if (count($this->tokens) > 0) {
|
|
// Check for unterminated comment
|
|
$lastToken = $this->tokens[count($this->tokens) - 1];
|
|
if ($this->isUnterminatedComment($lastToken)) {
|
|
$errorHandler->handleError(new Error('Unterminated comment', [
|
|
'startLine' => $line - substr_count($lastToken[1], "\n"),
|
|
'endLine' => $line,
|
|
'startFilePos' => $filePos - \strlen($lastToken[1]),
|
|
'endFilePos' => $filePos,
|
|
]));
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Fetches the next token.
|
|
*
|
|
* The available attributes are determined by the 'usedAttributes' option, which can
|
|
* be specified in the constructor. The following attributes are supported:
|
|
*
|
|
* * 'comments' => Array of PhpParser\Comment or PhpParser\Comment\Doc instances,
|
|
* representing all comments that occurred between the previous
|
|
* non-discarded token and the current one.
|
|
* * 'startLine' => Line in which the node starts.
|
|
* * 'endLine' => Line in which the node ends.
|
|
* * 'startTokenPos' => Offset into the token array of the first token in the node.
|
|
* * 'endTokenPos' => Offset into the token array of the last token in the node.
|
|
* * 'startFilePos' => Offset into the code string of the first character that is part of the node.
|
|
* * 'endFilePos' => Offset into the code string of the last character that is part of the node.
|
|
*
|
|
* @param mixed $value Variable to store token content in
|
|
* @param mixed $startAttributes Variable to store start attributes in
|
|
* @param mixed $endAttributes Variable to store end attributes in
|
|
*
|
|
* @return int Token id
|
|
*/
|
|
public function getNextToken(&$value = null, &$startAttributes = null, &$endAttributes = null) : int {
|
|
$startAttributes = [];
|
|
$endAttributes = [];
|
|
|
|
while (1) {
|
|
if (isset($this->tokens[++$this->pos])) {
|
|
$token = $this->tokens[$this->pos];
|
|
} else {
|
|
// EOF token with ID 0
|
|
$token = "\0";
|
|
}
|
|
|
|
if ($this->attributeStartLineUsed) {
|
|
$startAttributes['startLine'] = $this->line;
|
|
}
|
|
if ($this->attributeStartTokenPosUsed) {
|
|
$startAttributes['startTokenPos'] = $this->pos;
|
|
}
|
|
if ($this->attributeStartFilePosUsed) {
|
|
$startAttributes['startFilePos'] = $this->filePos;
|
|
}
|
|
|
|
if (\is_string($token)) {
|
|
$value = $token;
|
|
if (isset($token[1])) {
|
|
// bug in token_get_all
|
|
$this->filePos += 2;
|
|
$id = ord('"');
|
|
} else {
|
|
$this->filePos += 1;
|
|
$id = ord($token);
|
|
}
|
|
} elseif (!isset($this->dropTokens[$token[0]])) {
|
|
$value = $token[1];
|
|
$id = $this->tokenMap[$token[0]];
|
|
if (\T_CLOSE_TAG === $token[0]) {
|
|
$this->prevCloseTagHasNewline = false !== strpos($token[1], "\n")
|
|
|| false !== strpos($token[1], "\r");
|
|
} elseif (\T_INLINE_HTML === $token[0]) {
|
|
$startAttributes['hasLeadingNewline'] = $this->prevCloseTagHasNewline;
|
|
}
|
|
|
|
$this->line += substr_count($value, "\n");
|
|
$this->filePos += \strlen($value);
|
|
} else {
|
|
$origLine = $this->line;
|
|
$origFilePos = $this->filePos;
|
|
$this->line += substr_count($token[1], "\n");
|
|
$this->filePos += \strlen($token[1]);
|
|
|
|
if (\T_COMMENT === $token[0] || \T_DOC_COMMENT === $token[0]) {
|
|
if ($this->attributeCommentsUsed) {
|
|
$comment = \T_DOC_COMMENT === $token[0]
|
|
? new Comment\Doc($token[1],
|
|
$origLine, $origFilePos, $this->pos,
|
|
$this->line, $this->filePos - 1, $this->pos)
|
|
: new Comment($token[1],
|
|
$origLine, $origFilePos, $this->pos,
|
|
$this->line, $this->filePos - 1, $this->pos);
|
|
$startAttributes['comments'][] = $comment;
|
|
}
|
|
}
|
|
continue;
|
|
}
|
|
|
|
if ($this->attributeEndLineUsed) {
|
|
$endAttributes['endLine'] = $this->line;
|
|
}
|
|
if ($this->attributeEndTokenPosUsed) {
|
|
$endAttributes['endTokenPos'] = $this->pos;
|
|
}
|
|
if ($this->attributeEndFilePosUsed) {
|
|
$endAttributes['endFilePos'] = $this->filePos - 1;
|
|
}
|
|
|
|
return $id;
|
|
}
|
|
|
|
throw new \RuntimeException('Reached end of lexer loop');
|
|
}
|
|
|
|
/**
|
|
* Returns the token array for current code.
|
|
*
|
|
* The token array is in the same format as provided by the
|
|
* token_get_all() function and does not discard tokens (i.e.
|
|
* whitespace and comments are included). The token position
|
|
* attributes are against this token array.
|
|
*
|
|
* @return array Array of tokens in token_get_all() format
|
|
*/
|
|
public function getTokens() : array {
|
|
return $this->tokens;
|
|
}
|
|
|
|
/**
|
|
* Handles __halt_compiler() by returning the text after it.
|
|
*
|
|
* @return string Remaining text
|
|
*/
|
|
public function handleHaltCompiler() : string {
|
|
// text after T_HALT_COMPILER, still including ();
|
|
$textAfter = substr($this->code, $this->filePos);
|
|
|
|
// ensure that it is followed by ();
|
|
// this simplifies the situation, by not allowing any comments
|
|
// in between of the tokens.
|
|
if (!preg_match('~^\s*\(\s*\)\s*(?:;|\?>\r?\n?)~', $textAfter, $matches)) {
|
|
throw new Error('__HALT_COMPILER must be followed by "();"');
|
|
}
|
|
|
|
// prevent the lexer from returning any further tokens
|
|
$this->pos = count($this->tokens);
|
|
|
|
// return with (); removed
|
|
return substr($textAfter, strlen($matches[0]));
|
|
}
|
|
|
|
private function defineCompatibilityTokens() {
|
|
static $compatTokensDefined = false;
|
|
if ($compatTokensDefined) {
|
|
return;
|
|
}
|
|
|
|
$compatTokens = [
|
|
// PHP 7.4
|
|
'T_BAD_CHARACTER',
|
|
'T_FN',
|
|
'T_COALESCE_EQUAL',
|
|
// PHP 8.0
|
|
'T_NAME_QUALIFIED',
|
|
'T_NAME_FULLY_QUALIFIED',
|
|
'T_NAME_RELATIVE',
|
|
'T_MATCH',
|
|
'T_NULLSAFE_OBJECT_OPERATOR',
|
|
'T_ATTRIBUTE',
|
|
// PHP 8.1
|
|
'T_ENUM',
|
|
'T_AMPERSAND_NOT_FOLLOWED_BY_VAR_OR_VARARG',
|
|
'T_AMPERSAND_FOLLOWED_BY_VAR_OR_VARARG',
|
|
'T_READONLY',
|
|
];
|
|
|
|
// PHP-Parser might be used together with another library that also emulates some or all
|
|
// of these tokens. Perform a sanity-check that all already defined tokens have been
|
|
// assigned a unique ID.
|
|
$usedTokenIds = [];
|
|
foreach ($compatTokens as $token) {
|
|
if (\defined($token)) {
|
|
$tokenId = \constant($token);
|
|
$clashingToken = $usedTokenIds[$tokenId] ?? null;
|
|
if ($clashingToken !== null) {
|
|
throw new \Error(sprintf(
|
|
'Token %s has same ID as token %s, ' .
|
|
'you may be using a library with broken token emulation',
|
|
$token, $clashingToken
|
|
));
|
|
}
|
|
$usedTokenIds[$tokenId] = $token;
|
|
}
|
|
}
|
|
|
|
// Now define any tokens that have not yet been emulated. Try to assign IDs from -1
|
|
// downwards, but skip any IDs that may already be in use.
|
|
$newTokenId = -1;
|
|
foreach ($compatTokens as $token) {
|
|
if (!\defined($token)) {
|
|
while (isset($usedTokenIds[$newTokenId])) {
|
|
$newTokenId--;
|
|
}
|
|
\define($token, $newTokenId);
|
|
$newTokenId--;
|
|
}
|
|
}
|
|
|
|
$compatTokensDefined = true;
|
|
}
|
|
|
|
/**
|
|
* Creates the token map.
|
|
*
|
|
* The token map maps the PHP internal token identifiers
|
|
* to the identifiers used by the Parser. Additionally it
|
|
* maps T_OPEN_TAG_WITH_ECHO to T_ECHO and T_CLOSE_TAG to ';'.
|
|
*
|
|
* @return array The token map
|
|
*/
|
|
protected function createTokenMap() : array {
|
|
$tokenMap = [];
|
|
|
|
// 256 is the minimum possible token number, as everything below
|
|
// it is an ASCII value
|
|
for ($i = 256; $i < 1000; ++$i) {
|
|
if (\T_DOUBLE_COLON === $i) {
|
|
// T_DOUBLE_COLON is equivalent to T_PAAMAYIM_NEKUDOTAYIM
|
|
$tokenMap[$i] = Tokens::T_PAAMAYIM_NEKUDOTAYIM;
|
|
} elseif(\T_OPEN_TAG_WITH_ECHO === $i) {
|
|
// T_OPEN_TAG_WITH_ECHO with dropped T_OPEN_TAG results in T_ECHO
|
|
$tokenMap[$i] = Tokens::T_ECHO;
|
|
} elseif(\T_CLOSE_TAG === $i) {
|
|
// T_CLOSE_TAG is equivalent to ';'
|
|
$tokenMap[$i] = ord(';');
|
|
} elseif ('UNKNOWN' !== $name = token_name($i)) {
|
|
if ('T_HASHBANG' === $name) {
|
|
// HHVM uses a special token for #! hashbang lines
|
|
$tokenMap[$i] = Tokens::T_INLINE_HTML;
|
|
} elseif (defined($name = Tokens::class . '::' . $name)) {
|
|
// Other tokens can be mapped directly
|
|
$tokenMap[$i] = constant($name);
|
|
}
|
|
}
|
|
}
|
|
|
|
// HHVM uses a special token for numbers that overflow to double
|
|
if (defined('T_ONUMBER')) {
|
|
$tokenMap[\T_ONUMBER] = Tokens::T_DNUMBER;
|
|
}
|
|
// HHVM also has a separate token for the __COMPILER_HALT_OFFSET__ constant
|
|
if (defined('T_COMPILER_HALT_OFFSET')) {
|
|
$tokenMap[\T_COMPILER_HALT_OFFSET] = Tokens::T_STRING;
|
|
}
|
|
|
|
// Assign tokens for which we define compatibility constants, as token_name() does not know them.
|
|
$tokenMap[\T_FN] = Tokens::T_FN;
|
|
$tokenMap[\T_COALESCE_EQUAL] = Tokens::T_COALESCE_EQUAL;
|
|
$tokenMap[\T_NAME_QUALIFIED] = Tokens::T_NAME_QUALIFIED;
|
|
$tokenMap[\T_NAME_FULLY_QUALIFIED] = Tokens::T_NAME_FULLY_QUALIFIED;
|
|
$tokenMap[\T_NAME_RELATIVE] = Tokens::T_NAME_RELATIVE;
|
|
$tokenMap[\T_MATCH] = Tokens::T_MATCH;
|
|
$tokenMap[\T_NULLSAFE_OBJECT_OPERATOR] = Tokens::T_NULLSAFE_OBJECT_OPERATOR;
|
|
$tokenMap[\T_ATTRIBUTE] = Tokens::T_ATTRIBUTE;
|
|
$tokenMap[\T_AMPERSAND_NOT_FOLLOWED_BY_VAR_OR_VARARG] = Tokens::T_AMPERSAND_NOT_FOLLOWED_BY_VAR_OR_VARARG;
|
|
$tokenMap[\T_AMPERSAND_FOLLOWED_BY_VAR_OR_VARARG] = Tokens::T_AMPERSAND_FOLLOWED_BY_VAR_OR_VARARG;
|
|
$tokenMap[\T_ENUM] = Tokens::T_ENUM;
|
|
$tokenMap[\T_READONLY] = Tokens::T_READONLY;
|
|
|
|
return $tokenMap;
|
|
}
|
|
|
|
private function createIdentifierTokenMap(): array {
|
|
// Based on semi_reserved production.
|
|
return array_fill_keys([
|
|
\T_STRING,
|
|
\T_STATIC, \T_ABSTRACT, \T_FINAL, \T_PRIVATE, \T_PROTECTED, \T_PUBLIC, \T_READONLY,
|
|
\T_INCLUDE, \T_INCLUDE_ONCE, \T_EVAL, \T_REQUIRE, \T_REQUIRE_ONCE, \T_LOGICAL_OR, \T_LOGICAL_XOR, \T_LOGICAL_AND,
|
|
\T_INSTANCEOF, \T_NEW, \T_CLONE, \T_EXIT, \T_IF, \T_ELSEIF, \T_ELSE, \T_ENDIF, \T_ECHO, \T_DO, \T_WHILE,
|
|
\T_ENDWHILE, \T_FOR, \T_ENDFOR, \T_FOREACH, \T_ENDFOREACH, \T_DECLARE, \T_ENDDECLARE, \T_AS, \T_TRY, \T_CATCH,
|
|
\T_FINALLY, \T_THROW, \T_USE, \T_INSTEADOF, \T_GLOBAL, \T_VAR, \T_UNSET, \T_ISSET, \T_EMPTY, \T_CONTINUE, \T_GOTO,
|
|
\T_FUNCTION, \T_CONST, \T_RETURN, \T_PRINT, \T_YIELD, \T_LIST, \T_SWITCH, \T_ENDSWITCH, \T_CASE, \T_DEFAULT,
|
|
\T_BREAK, \T_ARRAY, \T_CALLABLE, \T_EXTENDS, \T_IMPLEMENTS, \T_NAMESPACE, \T_TRAIT, \T_INTERFACE, \T_CLASS,
|
|
\T_CLASS_C, \T_TRAIT_C, \T_FUNC_C, \T_METHOD_C, \T_LINE, \T_FILE, \T_DIR, \T_NS_C, \T_HALT_COMPILER, \T_FN,
|
|
\T_MATCH,
|
|
], true);
|
|
}
|
|
}
|