youtube-dl/youtube_dl/jsinterp.py

from __future__ import unicode_literals

from functools import update_wrapper
import itertools
import json
import math
import operator
import re

from .utils import (
    error_to_compat_str,
    ExtractorError,
    js_to_json,
    remove_quotes,
    unified_timestamp,
    variadic,
)
from .compat import (
    compat_basestring,
    compat_chr,
    compat_collections_chain_map as ChainMap,
    compat_itertools_zip_longest as zip_longest,
    compat_str,
)


def wraps_op(op):

    def update_and_rename_wrapper(w):
        f = update_wrapper(w, op)
        # fn names are str in both Py 2/3
        f.__name__ = str('JS_') + f.__name__
        return f

    return update_and_rename_wrapper


def _js_bit_op(op):

    def zeroise(x):
        return 0 if x in (None, JS_Undefined) else x

    @wraps_op(op)
    def wrapped(a, b):
        return op(zeroise(a), zeroise(b)) & 0xffffffff

    return wrapped


def _js_arith_op(op):

    @wraps_op(op)
    def wrapped(a, b):
        if JS_Undefined in (a, b):
            return float('nan')
        return op(a or 0, b or 0)

    return wrapped


def _js_div(a, b):
    if JS_Undefined in (a, b) or not (a and b):
        return float('nan')
    return operator.truediv(a or 0, b) if b else float('inf')


def _js_mod(a, b):
    if JS_Undefined in (a, b) or not b:
        return float('nan')
    return (a or 0) % b


def _js_exp(a, b):
    if not b:
        return 1  # even 0 ** 0 !!
    elif JS_Undefined in (a, b):
        return float('nan')
    return (a or 0) ** b


def _js_eq_op(op):

    @wraps_op(op)
    def wrapped(a, b):
        if set((a, b)) <= set((None, JS_Undefined)):
            return op(a, a)
        return op(a, b)

    return wrapped


def _js_comp_op(op):

    @wraps_op(op)
    def wrapped(a, b):
        if JS_Undefined in (a, b):
            return False
        if isinstance(a, compat_basestring):
            b = compat_str(b or 0)
        elif isinstance(b, compat_basestring):
            a = compat_str(a or 0)
        return op(a or 0, b or 0)

    return wrapped


def _js_ternary(cndn, if_true=True, if_false=False):
    """Simulate JS's ternary operator (cndn?if_true:if_false)"""
    if cndn in (False, None, 0, '', JS_Undefined):
        return if_false
    try:
        if math.isnan(cndn):  # NB: NaN cannot be checked by membership
            return if_false
    except TypeError:
        pass
    return if_true


# (op, definition) in order of binding priority, tightest first
# avoid dict to maintain order
# definition None => Defined in JSInterpreter._operator
_OPERATORS = (
    ('>>', _js_bit_op(operator.rshift)),
    ('<<', _js_bit_op(operator.lshift)),
    ('+', _js_arith_op(operator.add)),
    ('-', _js_arith_op(operator.sub)),
    ('*', _js_arith_op(operator.mul)),
    ('%', _js_mod),
    ('/', _js_div),
    ('**', _js_exp),
)

_COMP_OPERATORS = (
    ('===', operator.is_),
    ('!==', operator.is_not),
    ('==', _js_eq_op(operator.eq)),
    ('!=', _js_eq_op(operator.ne)),
    ('<=', _js_comp_op(operator.le)),
    ('>=', _js_comp_op(operator.ge)),
    ('<', _js_comp_op(operator.lt)),
    ('>', _js_comp_op(operator.gt)),
)

_LOG_OPERATORS = (
    ('|', _js_bit_op(operator.or_)),
    ('^', _js_bit_op(operator.xor)),
    ('&', _js_bit_op(operator.and_)),
)

_SC_OPERATORS = (
    ('?', None),
    ('??', None),
    ('||', None),
    ('&&', None),
)

_OPERATOR_RE = '|'.join(map(lambda x: re.escape(x[0]), _OPERATORS + _LOG_OPERATORS))

_NAME_RE = r'[a-zA-Z_$][\w$]*'
_MATCHING_PARENS = dict(zip(*zip('()', '{}', '[]')))
_QUOTES = '\'"/'


class JS_Undefined(object):
    pass


class JS_Break(ExtractorError):
    def __init__(self):
        ExtractorError.__init__(self, 'Invalid break')


class JS_Continue(ExtractorError):
    def __init__(self):
        ExtractorError.__init__(self, 'Invalid continue')


class JS_Throw(ExtractorError):
    def __init__(self, e):
        self.error = e
        ExtractorError.__init__(self, 'Uncaught exception ' + error_to_compat_str(e))


class LocalNameSpace(ChainMap):
    def __getitem__(self, key):
        try:
            return super(LocalNameSpace, self).__getitem__(key)
        except KeyError:
            return JS_Undefined

    def __setitem__(self, key, value):
        for scope in self.maps:
            if key in scope:
                scope[key] = value
                return
        self.maps[0][key] = value

    def __delitem__(self, key):
        raise NotImplementedError('Deleting is not supported')

    def __repr__(self):
        return 'LocalNameSpace%s' % (self.maps, )


class JSInterpreter(object):
    __named_object_counter = 0

    _OBJ_NAME = '__youtube_dl_jsinterp_obj'

    OP_CHARS = None

    def __init__(self, code, objects=None):
        self.code, self._functions = code, {}
        self._objects = {} if objects is None else objects
        if type(self).OP_CHARS is None:
            type(self).OP_CHARS = self.OP_CHARS = self.__op_chars()

    class Exception(ExtractorError):
        def __init__(self, msg, *args, **kwargs):
            expr = kwargs.pop('expr', None)
            if expr is not None:
                msg = '{0} in: {1!r:.100}'.format(msg.rstrip(), expr)
            super(JSInterpreter.Exception, self).__init__(msg, *args, **kwargs)

    class JS_RegExp(object):
        RE_FLAGS = {
            # special knowledge: Python's re flags are bitmask values, current max 128
            # invent new bitmask values well above that for literal parsing
            # TODO: execute matches with these flags (remaining: d, y)
            'd': 1024,  # Generate indices for substring matches
            'g': 2048,  # Global search
            'i': re.I,  # Case-insensitive search
            'm': re.M,  # Multi-line search
            's': re.S,  # Allows . to match newline characters
            'u': re.U,  # Treat a pattern as a sequence of unicode code points
            'y': 4096,  # Perform a "sticky" search that matches starting at the current position in the target string
        }

        def __init__(self, pattern_txt, flags=0):
            if isinstance(flags, compat_str):
                flags, _ = self.regex_flags(flags)
            # First, avoid https://github.com/python/cpython/issues/74534
            self.__self = None
            self.__pattern_txt = pattern_txt.replace('[[', r'[\[')
            self.__flags = flags

        def __instantiate(self):
            if self.__self:
                return
            self.__self = re.compile(self.__pattern_txt, self.__flags)
            # Thx: https://stackoverflow.com/questions/44773522/setattr-on-python2-sre-sre-pattern
            for name in dir(self.__self):
                # Only these? Obviously __class__, __init__.
                # PyPy creates a __weakref__ attribute with value None
                # that can't be setattr'd but also can't need to be copied.
                if name in ('__class__', '__init__', '__weakref__'):
                    continue
                setattr(self, name, getattr(self.__self, name))

        def __getattr__(self, name):
            self.__instantiate()
            if hasattr(self, name):
                return getattr(self, name)
            return super(JSInterpreter.JS_RegExp, self).__getattr__(name)

        @classmethod
        def regex_flags(cls, expr):
            flags = 0
            if not expr:
                return flags, expr
            for idx, ch in enumerate(expr):
                if ch not in cls.RE_FLAGS:
                    break
                flags |= cls.RE_FLAGS[ch]
            return flags, expr[idx + 1:]

    @classmethod
    def __op_chars(cls):
        op_chars = set(';,[')
        for op in cls._all_operators():
            for c in op[0]:
                op_chars.add(c)
        return op_chars

    def _named_object(self, namespace, obj):
        self.__named_object_counter += 1
        name = '%s%d' % (self._OBJ_NAME, self.__named_object_counter)
        namespace[name] = obj
        return name

    @classmethod
    def _separate(cls, expr, delim=',', max_split=None, skip_delims=None):
        if not expr:
            return
        # collections.Counter() is ~10% slower in both 2.7 and 3.9
        counters = dict((k, 0) for k in _MATCHING_PARENS.values())
        start, splits, pos, delim_len = 0, 0, 0, len(delim) - 1
        in_quote, escaping, skipping = None, False, 0
        after_op, in_regex_char_group = True, False

        for idx, char in enumerate(expr):
            paren_delta = 0
            if not in_quote:
                if char in _MATCHING_PARENS:
                    counters[_MATCHING_PARENS[char]] += 1
                    paren_delta = 1
                elif char in counters:
                    counters[char] -= 1
                    paren_delta = -1
            if not escaping:
                if char in _QUOTES and in_quote in (char, None):
                    if in_quote or after_op or char != '/':
                        in_quote = None if in_quote and not in_regex_char_group else char
                elif in_quote == '/' and char in '[]':
                    in_regex_char_group = char == '['
            escaping = not escaping and in_quote and char == '\\'
            after_op = not in_quote and (char in cls.OP_CHARS or paren_delta > 0 or (after_op and char.isspace()))

            if char != delim[pos] or any(counters.values()) or in_quote:
                pos = skipping = 0
                continue
            elif skipping > 0:
                skipping -= 1
                continue
            elif pos == 0 and skip_delims:
                here = expr[idx:]
                for s in variadic(skip_delims):
                    if here.startswith(s) and s:
                        skipping = len(s) - 1
                        break
                if skipping > 0:
                    continue
            if pos < delim_len:
                pos += 1
                continue
            yield expr[start: idx - delim_len]
            start, pos = idx + 1, 0
            splits += 1
            if max_split and splits >= max_split:
                break
        yield expr[start:]

    @classmethod
    def _separate_at_paren(cls, expr, delim=None):
        if delim is None:
            delim = expr and _MATCHING_PARENS[expr[0]]
        separated = list(cls._separate(expr, delim, 1))

        if len(separated) < 2:
            raise cls.Exception('No terminating paren {delim} in {expr!r:.5500}'.format(**locals()))
        return separated[0][1:].strip(), separated[1].strip()

    @staticmethod
    def _all_operators():
        return itertools.chain(
            # Ref: https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Operator_Precedence
            _SC_OPERATORS, _LOG_OPERATORS, _COMP_OPERATORS, _OPERATORS)

    def _operator(self, op, left_val, right_expr, expr, local_vars, allow_recursion):
        if op in ('||', '&&'):
            if (op == '&&') ^ _js_ternary(left_val):
                return left_val  # short circuiting
        elif op == '??':
            if left_val not in (None, JS_Undefined):
                return left_val
        elif op == '?':
            right_expr = _js_ternary(left_val, *self._separate(right_expr, ':', 1))

        right_val = self.interpret_expression(right_expr, local_vars, allow_recursion)
        opfunc = op and next((v for k, v in self._all_operators() if k == op), None)
        if not opfunc:
            return right_val

        try:
            # print('Eval:', opfunc.__name__, left_val, right_val)
            return opfunc(left_val, right_val)
        except Exception as e:
            raise self.Exception('Failed to evaluate {left_val!r:.50} {op} {right_val!r:.50}'.format(**locals()), expr, cause=e)

    def _index(self, obj, idx, allow_undefined=False):
        if idx == 'length':
            return len(obj)
        try:
            return obj[int(idx)] if isinstance(obj, list) else obj[idx]
        except Exception as e:
            if allow_undefined:
                return JS_Undefined
            raise self.Exception('Cannot get index {idx:.100}'.format(**locals()), expr=repr(obj), cause=e)

    def _dump(self, obj, namespace):
        try:
            return json.dumps(obj)
        except TypeError:
            return self._named_object(namespace, obj)

    # used below
    _VAR_RET_THROW_RE = re.compile(r'''(?x)
        (?P<var>(?:var|const|let)\s)|return(?:\s+|(?=["'])|$)|(?P<throw>throw\s+)
        ''')
    _COMPOUND_RE = re.compile(r'''(?x)
        (?P<try>try)\s*\{|
        (?P<if>if)\s*\(|
        (?P<switch>switch)\s*\(|
        (?P<for>for)\s*\(|
        (?P<while>while)\s*\(
        ''')
    _FINALLY_RE = re.compile(r'finally\s*\{')
    _SWITCH_RE = re.compile(r'switch\s*\(')

    def interpret_statement(self, stmt, local_vars, allow_recursion=100):
        if allow_recursion < 0:
            raise self.Exception('Recursion limit reached')
        allow_recursion -= 1

        # print('At: ' + stmt[:60])
        should_return = False
        # fails on (eg) if (...) stmt1; else stmt2;
        sub_statements = list(self._separate(stmt, ';')) or ['']
        expr = stmt = sub_statements.pop().strip()
        for sub_stmt in sub_statements:
            ret, should_return = self.interpret_statement(sub_stmt, local_vars, allow_recursion)
            if should_return:
                return ret, should_return

        m = self._VAR_RET_THROW_RE.match(stmt)
        if m:
            expr = stmt[len(m.group(0)):].strip()
            if m.group('throw'):
                raise JS_Throw(self.interpret_expression(expr, local_vars, allow_recursion))
            should_return = not m.group('var')
        if not expr:
            return None, should_return

        if expr[0] in _QUOTES:
            inner, outer = self._separate(expr, expr[0], 1)
            if expr[0] == '/':
                flags, outer = self.JS_RegExp.regex_flags(outer)
                inner = self.JS_RegExp(inner[1:], flags=flags)
            else:
                inner = json.loads(js_to_json(inner + expr[0]))  # , strict=True))
            if not outer:
                return inner, should_return
            expr = self._named_object(local_vars, inner) + outer

        new_kw, _, obj = expr.partition('new ')
        if not new_kw:
            for klass, konstr in (('Date', lambda x: int(unified_timestamp(x, False) * 1000)),
                                  ('RegExp', self.JS_RegExp),
                                  ('Error', self.Exception)):
                if not obj.startswith(klass + '('):
                    continue
                left, right = self._separate_at_paren(obj[len(klass):])
                argvals = self.interpret_iter(left, local_vars, allow_recursion)
                expr = konstr(*argvals)
                if expr is None:
                    raise self.Exception('Failed to parse {klass} {left!r:.100}'.format(**locals()), expr=expr)
                expr = self._dump(expr, local_vars) + right
                break
            else:
                raise self.Exception('Unsupported object {obj:.100}'.format(**locals()), expr=expr)

        if expr.startswith('void '):
            left = self.interpret_expression(expr[5:], local_vars, allow_recursion)
            return None, should_return

        if expr.startswith('{'):
            inner, outer = self._separate_at_paren(expr)
            # try for object expression (Map)
            sub_expressions = [list(self._separate(sub_expr.strip(), ':', 1)) for sub_expr in self._separate(inner)]
            if all(len(sub_expr) == 2 for sub_expr in sub_expressions):
                return dict(
                    (key_expr if re.match(_NAME_RE, key_expr) else key_expr,
                     self.interpret_expression(val_expr, local_vars, allow_recursion))
                    for key_expr, val_expr in sub_expressions), should_return
            # or statement list
            inner, should_abort = self.interpret_statement(inner, local_vars, allow_recursion)
            if not outer or should_abort:
                return inner, should_abort or should_return
            else:
                expr = self._dump(inner, local_vars) + outer

        if expr.startswith('('):
            inner, outer = self._separate_at_paren(expr)
            inner, should_abort = self.interpret_statement(inner, local_vars, allow_recursion)
            if not outer or should_abort:
                return inner, should_abort or should_return
            else:
                expr = self._dump(inner, local_vars) + outer

        if expr.startswith('['):
            inner, outer = self._separate_at_paren(expr)
            name = self._named_object(local_vars, [
                self.interpret_expression(item, local_vars, allow_recursion)
                for item in self._separate(inner)])
            expr = name + outer

        m = self._COMPOUND_RE.match(expr)
        md = m.groupdict() if m else {}
        if md.get('if'):
            cndn, expr = self._separate_at_paren(expr[m.end() - 1:])
            if expr.startswith('{'):
                if_expr, expr = self._separate_at_paren(expr)
            else:
                # may lose ... else ... because of ll.368-374
                if_expr, expr = self._separate_at_paren(expr, delim=';')
            else_expr = None
            m = re.match(r'else\s*(?P<block>\{)?', expr)
            if m:
                if m.group('block'):
                    else_expr, expr = self._separate_at_paren(expr[m.end() - 1:])
                else:
                    # handle subset ... else if (...) {...} else ...
                    # TODO: make interpret_statement do this properly, if possible
                    exprs = list(self._separate(expr[m.end():], delim='}', max_split=2))
                    if len(exprs) > 1:
                        if re.match(r'\s*if\s*\(', exprs[0]) and re.match(r'\s*else\b', exprs[1]):
                            else_expr = exprs[0] + '}' + exprs[1]
                            expr = (exprs[2] + '}') if len(exprs) == 3 else None
                        else:
                            else_expr = exprs[0]
                            exprs.append('')
                            expr = '}'.join(exprs[1:])
                    else:
                        else_expr = exprs[0]
                        expr = None
                    else_expr = else_expr.lstrip() + '}'
            cndn = _js_ternary(self.interpret_expression(cndn, local_vars, allow_recursion))
            ret, should_abort = self.interpret_statement(
                if_expr if cndn else else_expr, local_vars, allow_recursion)
            if should_abort:
                return ret, True

        elif md.get('try'):
            try_expr, expr = self._separate_at_paren(expr[m.end() - 1:])
            err = None
            try:
                ret, should_abort = self.interpret_statement(try_expr, local_vars, allow_recursion)
                if should_abort:
                    return ret, True
            except Exception as e:
                # XXX: This works for now, but makes debugging future issues very hard
                err = e

            pending = (None, False)
            m = re.match(r'catch\s*(?P<err>\(\s*{_NAME_RE}\s*\))?\{{'.format(**globals()), expr)
            if m:
                sub_expr, expr = self._separate_at_paren(expr[m.end() - 1:])
                if err:
                    catch_vars = {}
                    if m.group('err'):
                        catch_vars[m.group('err')] = err.error if isinstance(err, JS_Throw) else err
                    catch_vars = local_vars.new_child(m=catch_vars)
                    err = None
                    pending = self.interpret_statement(sub_expr, catch_vars, allow_recursion)

            m = self._FINALLY_RE.match(expr)
            if m:
                sub_expr, expr = self._separate_at_paren(expr[m.end() - 1:])
                ret, should_abort = self.interpret_statement(sub_expr, local_vars, allow_recursion)
                if should_abort:
                    return ret, True

            ret, should_abort = pending
            if should_abort:
                return ret, True

            if err:
                raise err

        elif md.get('for') or md.get('while'):
            init_or_cond, remaining = self._separate_at_paren(expr[m.end() - 1:])
            if remaining.startswith('{'):
                body, expr = self._separate_at_paren(remaining)
            else:
                switch_m = self._SWITCH_RE.match(remaining)  # FIXME
                if switch_m:
                    switch_val, remaining = self._separate_at_paren(remaining[switch_m.end() - 1:])
                    body, expr = self._separate_at_paren(remaining, '}')
                    body = 'switch(%s){%s}' % (switch_val, body)
                else:
                    body, expr = remaining, ''
            if md.get('for'):
                start, cndn, increment = self._separate(init_or_cond, ';')
                self.interpret_expression(start, local_vars, allow_recursion)
            else:
                cndn, increment = init_or_cond, None
            while _js_ternary(self.interpret_expression(cndn, local_vars, allow_recursion)):
                try:
                    ret, should_abort = self.interpret_statement(body, local_vars, allow_recursion)
                    if should_abort:
                        return ret, True
                except JS_Break:
                    break
                except JS_Continue:
                    pass
                if increment:
                    self.interpret_expression(increment, local_vars, allow_recursion)

        elif md.get('switch'):
            switch_val, remaining = self._separate_at_paren(expr[m.end() - 1:])
            switch_val = self.interpret_expression(switch_val, local_vars, allow_recursion)
            body, expr = self._separate_at_paren(remaining, '}')
            items = body.replace('default:', 'case default:').split('case ')[1:]
            for default in (False, True):
                matched = False
                for item in items:
                    case, stmt = (i.strip() for i in self._separate(item, ':', 1))
                    if default:
                        matched = matched or case == 'default'
                    elif not matched:
                        matched = (case != 'default'
                                   and switch_val == self.interpret_expression(case, local_vars, allow_recursion))
                    if not matched:
                        continue
                    try:
                        ret, should_abort = self.interpret_statement(stmt, local_vars, allow_recursion)
                        if should_abort:
                            return ret
                    except JS_Break:
                        break
                if matched:
                    break

        if md:
            ret, should_abort = self.interpret_statement(expr, local_vars, allow_recursion)
            return ret, should_abort or should_return

        # Comma separated statements
        sub_expressions = list(self._separate(expr))
        if len(sub_expressions) > 1:
            for sub_expr in sub_expressions:
                ret, should_abort = self.interpret_statement(sub_expr, local_vars, allow_recursion)
                if should_abort:
                    return ret, True
            return ret, False

        for m in re.finditer(r'''(?x)
                (?P<pre_sign>\+\+|--)(?P<var1>{_NAME_RE})|
                (?P<var2>{_NAME_RE})(?P<post_sign>\+\+|--)'''.format(**globals()), expr):
            var = m.group('var1') or m.group('var2')
            start, end = m.span()
            sign = m.group('pre_sign') or m.group('post_sign')
            ret = local_vars[var]
            local_vars[var] += 1 if sign[0] == '+' else -1
            if m.group('pre_sign'):
                ret = local_vars[var]
            expr = expr[:start] + self._dump(ret, local_vars) + expr[end:]

        if not expr:
            return None, should_return

        m = re.match(r'''(?x)
            (?P<assign>
                (?P<out>{_NAME_RE})(?:\[(?P<index>[^\]]+?)\])?\s*
                (?P<op>{_OPERATOR_RE})?
                =(?!=)(?P<expr>.*)$
            )|(?P<return>
                (?!if|return|true|false|null|undefined)(?P<name>{_NAME_RE})$
            )|(?P<indexing>
                (?P<in>{_NAME_RE})\[(?P<idx>.+)\]$
            )|(?P<attribute>
                (?P<var>{_NAME_RE})(?:(?P<nullish>\?)?\.(?P<member>[^(]+)|\[(?P<member2>[^\]]+)\])\s*
            )|(?P<function>
                (?P<fname>{_NAME_RE})\((?P<args>.*)\)$
            )'''.format(**globals()), expr)
        md = m.groupdict() if m else {}
        if md.get('assign'):
            left_val = local_vars.get(m.group('out'))

            if not m.group('index'):
                local_vars[m.group('out')] = self._operator(
                    m.group('op'), left_val, m.group('expr'), expr, local_vars, allow_recursion)
                return local_vars[m.group('out')], should_return
            elif left_val in (None, JS_Undefined):
                raise self.Exception('Cannot index undefined variable ' + m.group('out'), expr=expr)

            idx = self.interpret_expression(m.group('index'), local_vars, allow_recursion)
            if not isinstance(idx, (int, float)):
                raise self.Exception('List index %s must be integer' % (idx, ), expr=expr)
            idx = int(idx)
            left_val[idx] = self._operator(
                m.group('op'), self._index(left_val, idx), m.group('expr'), expr, local_vars, allow_recursion)
            return left_val[idx], should_return

        elif expr.isdigit():
            return int(expr), should_return

        elif expr == 'break':
            raise JS_Break()
        elif expr == 'continue':
            raise JS_Continue()

        elif expr == 'undefined':
            return JS_Undefined, should_return
        elif expr == 'NaN':
            return float('NaN'), should_return

        elif md.get('return'):
            return local_vars[m.group('name')], should_return

        try:
            ret = json.loads(js_to_json(expr))  # strict=True)
            if not md.get('attribute'):
                return ret, should_return
        except ValueError:
            pass

        if md.get('indexing'):
            val = local_vars[m.group('in')]
            idx = self.interpret_expression(m.group('idx'), local_vars, allow_recursion)
            return self._index(val, idx), should_return

        for op, _ in self._all_operators():
            # hackety: </> have higher priority than <</>>, but don't confuse them
            skip_delim = (op + op) if op in '<>*?' else None
            if op == '?':
                skip_delim = (skip_delim, '?.')
            separated = list(self._separate(expr, op, skip_delims=skip_delim))
            if len(separated) < 2:
                continue

            right_expr = separated.pop()
            # handle operators that are both unary and binary, minimal BODMAS
            if op in ('+', '-'):
                undone = 0
                while len(separated) > 1 and not separated[-1].strip():
                    undone += 1
                    separated.pop()
                if op == '-' and undone % 2 != 0:
                    right_expr = op + right_expr
                left_val = separated[-1]
                for dm_op in ('*', '%', '/', '**'):
                    bodmas = tuple(self._separate(left_val, dm_op, skip_delims=skip_delim))
                    if len(bodmas) > 1 and not bodmas[-1].strip():
                        expr = op.join(separated) + op + right_expr
                        right_expr = None
                        break
                if right_expr is None:
                    continue

            left_val = self.interpret_expression(op.join(separated), local_vars, allow_recursion)
            return self._operator(op, left_val, right_expr, expr, local_vars, allow_recursion), should_return

        if md.get('attribute'):
            variable, member, nullish = m.group('var', 'member', 'nullish')
            if not member:
                member = self.interpret_expression(m.group('member2'), local_vars, allow_recursion)
            arg_str = expr[m.end():]
            if arg_str.startswith('('):
                arg_str, remaining = self._separate_at_paren(arg_str)
            else:
                arg_str, remaining = None, arg_str

            def assertion(cndn, msg):
                """ assert, but without risk of getting optimized out """
                if not cndn:
                    memb = member
                    raise self.Exception('{memb} {msg}'.format(**locals()), expr=expr)

            def eval_method():
                if (variable, member) == ('console', 'debug'):
                    return
                types = {
                    'String': compat_str,
                    'Math': float,
                }
                obj = local_vars.get(variable)
                if obj in (JS_Undefined, None):
                    obj = types.get(variable, JS_Undefined)
                if obj is JS_Undefined:
                    try:
                        if variable not in self._objects:
                            self._objects[variable] = self.extract_object(variable)
                        obj = self._objects[variable]
                    except self.Exception:
                        if not nullish:
                            raise

                if nullish and obj is JS_Undefined:
                    return JS_Undefined

                # Member access
                if arg_str is None:
                    return self._index(obj, member, nullish)

                # Function call
                argvals = [
                    self.interpret_expression(v, local_vars, allow_recursion)
                    for v in self._separate(arg_str)]

                if obj == compat_str:
                    if member == 'fromCharCode':
                        assertion(argvals, 'takes one or more arguments')
                        return ''.join(map(compat_chr, argvals))
                    raise self.Exception('Unsupported string method ' + member, expr=expr)
                elif obj == float:
                    if member == 'pow':
                        assertion(len(argvals) == 2, 'takes two arguments')
                        return argvals[0] ** argvals[1]
                    raise self.Exception('Unsupported Math method ' + member, expr=expr)

                if member == 'split':
                    assertion(argvals, 'takes one or more arguments')
                    assertion(len(argvals) == 1, 'with limit argument is not implemented')
                    return obj.split(argvals[0]) if argvals[0] else list(obj)
                elif member == 'join':
                    assertion(isinstance(obj, list), 'must be applied on a list')
                    assertion(len(argvals) == 1, 'takes exactly one argument')
                    return argvals[0].join(obj)
                elif member == 'reverse':
                    assertion(not argvals, 'does not take any arguments')
                    obj.reverse()
                    return obj
                elif member == 'slice':
                    assertion(isinstance(obj, list), 'must be applied on a list')
                    assertion(len(argvals) == 1, 'takes exactly one argument')
                    return obj[argvals[0]:]
                elif member == 'splice':
                    assertion(isinstance(obj, list), 'must be applied on a list')
                    assertion(argvals, 'takes one or more arguments')
                    index, howMany = map(int, (argvals + [len(obj)])[:2])
                    if index < 0:
                        index += len(obj)
                    add_items = argvals[2:]
                    res = []
                    for i in range(index, min(index + howMany, len(obj))):
                        res.append(obj.pop(index))
                    for i, item in enumerate(add_items):
                        obj.insert(index + i, item)
                    return res
                elif member == 'unshift':
                    assertion(isinstance(obj, list), 'must be applied on a list')
                    assertion(argvals, 'takes one or more arguments')
                    for item in reversed(argvals):
                        obj.insert(0, item)
                    return obj
                elif member == 'pop':
                    assertion(isinstance(obj, list), 'must be applied on a list')
                    assertion(not argvals, 'does not take any arguments')
                    if not obj:
                        return
                    return obj.pop()
                elif member == 'push':
                    assertion(argvals, 'takes one or more arguments')
                    obj.extend(argvals)
                    return obj
                elif member == 'forEach':
                    assertion(argvals, 'takes one or more arguments')
                    assertion(len(argvals) <= 2, 'takes at-most 2 arguments')
                    f, this = (argvals + [''])[:2]
                    return [f((item, idx, obj), {'this': this}, allow_recursion) for idx, item in enumerate(obj)]
                elif member == 'indexOf':
                    assertion(argvals, 'takes one or more arguments')
                    assertion(len(argvals) <= 2, 'takes at-most 2 arguments')
                    idx, start = (argvals + [0])[:2]
                    try:
                        return obj.index(idx, start)
                    except ValueError:
                        return -1
                elif member == 'charCodeAt':
                    assertion(isinstance(obj, compat_str), 'must be applied on a string')
                    # assertion(len(argvals) == 1, 'takes exactly one argument') # but not enforced
                    idx = argvals[0] if isinstance(argvals[0], int) else 0
                    if idx >= len(obj):
                        return None
                    return ord(obj[idx])
                elif member in ('replace', 'replaceAll'):
                    assertion(isinstance(obj, compat_str), 'must be applied on a string')
                    assertion(len(argvals) == 2, 'takes exactly two arguments')
                    # TODO: argvals[1] callable, other Py vs JS edge cases
                    if isinstance(argvals[0], self.JS_RegExp):
                        count = 0 if argvals[0].flags & self.JS_RegExp.RE_FLAGS['g'] else 1
                        assertion(member != 'replaceAll' or count == 0,
                                  'replaceAll must be called with a global RegExp')
                        return argvals[0].sub(argvals[1], obj, count=count)
                    count = ('replaceAll', 'replace').index(member)
                    return re.sub(re.escape(argvals[0]), argvals[1], obj, count=count)

                idx = int(member) if isinstance(obj, list) else member
                return obj[idx](argvals, allow_recursion=allow_recursion)

            if remaining:
                ret, should_abort = self.interpret_statement(
                    self._named_object(local_vars, eval_method()) + remaining,
                    local_vars, allow_recursion)
                return ret, should_return or should_abort
            else:
                return eval_method(), should_return

        elif md.get('function'):
            fname = m.group('fname')
            argvals = [self.interpret_expression(v, local_vars, allow_recursion)
                       for v in self._separate(m.group('args'))]
            if fname in local_vars:
                return local_vars[fname](argvals, allow_recursion=allow_recursion), should_return
            elif fname not in self._functions:
                self._functions[fname] = self.extract_function(fname)
            return self._functions[fname](argvals, allow_recursion=allow_recursion), should_return

        raise self.Exception(
            'Unsupported JS expression ' + (expr[:40] if expr != stmt else ''), expr=stmt)

    def interpret_expression(self, expr, local_vars, allow_recursion):
        ret, should_return = self.interpret_statement(expr, local_vars, allow_recursion)
        if should_return:
            raise self.Exception('Cannot return from an expression', expr)
        return ret

    def interpret_iter(self, list_txt, local_vars, allow_recursion):
        for v in self._separate(list_txt):
            yield self.interpret_expression(v, local_vars, allow_recursion)

    def extract_object(self, objname):
        _FUNC_NAME_RE = r'''(?:[a-zA-Z$0-9]+|"[a-zA-Z$0-9]+"|'[a-zA-Z$0-9]+')'''
        obj = {}
        obj_m = re.search(
            r'''(?x)
                (?<!this\.)%s\s*=\s*{\s*
                    (?P<fields>(%s\s*:\s*function\s*\(.*?\)\s*{.*?}(?:,\s*)?)*)
                }\s*;
            ''' % (re.escape(objname), _FUNC_NAME_RE),
            self.code)
        if not obj_m:
            raise self.Exception('Could not find object ' + objname)
        fields = obj_m.group('fields')
        # Currently, it only supports function definitions
        fields_m = re.finditer(
            r'''(?x)
                (?P<key>%s)\s*:\s*function\s*\((?P<args>(?:%s|,)*)\){(?P<code>[^}]+)}
            ''' % (_FUNC_NAME_RE, _NAME_RE),
            fields)
        for f in fields_m:
            argnames = self.build_arglist(f.group('args'))
            obj[remove_quotes(f.group('key'))] = self.build_function(argnames, f.group('code'))

        return obj

    def extract_function_code(self, funcname):
        """ @returns argnames, code """
        func_m = re.search(
            r'''(?xs)
                (?:
                    function\s+%(name)s|
                    [{;,]\s*%(name)s\s*=\s*function|
                    (?:var|const|let)\s+%(name)s\s*=\s*function
                )\s*
                \((?P<args>[^)]*)\)\s*
                (?P<code>{.+})''' % {'name': re.escape(funcname)},
            self.code)
        code, _ = self._separate_at_paren(func_m.group('code'))  # refine the match
        if func_m is None:
            raise self.Exception('Could not find JS function "{funcname}"'.format(**locals()))
        return self.build_arglist(func_m.group('args')), code

    def extract_function(self, funcname):
        return self.extract_function_from_code(*self.extract_function_code(funcname))

    def extract_function_from_code(self, argnames, code, *global_stack):
        local_vars = {}
        while True:
            mobj = re.search(r'function\((?P<args>[^)]*)\)\s*{', code)
            if mobj is None:
                break
            start, body_start = mobj.span()
            body, remaining = self._separate_at_paren(code[body_start - 1:], '}')
            name = self._named_object(local_vars, self.extract_function_from_code(
                [x.strip() for x in mobj.group('args').split(',')],
                body, local_vars, *global_stack))
            code = code[:start] + name + remaining
        return self.build_function(argnames, code, local_vars, *global_stack)

    def call_function(self, funcname, *args):
        return self.extract_function(funcname)(args)

    @classmethod
    def build_arglist(cls, arg_text):
        if not arg_text:
            return []

        def valid_arg(y):
            y = y.strip()
            if not y:
                raise cls.Exception('Missing arg in "%s"' % (arg_text, ))
            return y

        return [valid_arg(x) for x in cls._separate(arg_text)]

    def build_function(self, argnames, code, *global_stack):
        global_stack = list(global_stack) or [{}]
        argnames = tuple(argnames)
        # import pdb; pdb.set_trace()

        def resf(args, kwargs={}, allow_recursion=100):
            global_stack[0].update(
                zip_longest(argnames, args, fillvalue=None))
            global_stack[0].update(kwargs)
            var_stack = LocalNameSpace(*global_stack)
            ret, should_abort = self.interpret_statement(code.replace('\n', ' '), var_stack, allow_recursion - 1)
            if should_abort:
                return ret
        return resf