889 lines
28 KiB
Python
889 lines
28 KiB
Python
# pyshlex.py - PLY compatible lexer for pysh.
|
|
#
|
|
# Copyright 2007 Patrick Mezard
|
|
#
|
|
# This software may be used and distributed according to the terms
|
|
# of the GNU General Public License, incorporated herein by reference.
|
|
|
|
# TODO:
|
|
# - review all "char in 'abc'" snippets: the empty string can be matched
|
|
# - test line continuations within quoted/expansion strings
|
|
# - eof is buggy wrt sublexers
|
|
# - the lexer cannot really work in pull mode as it would be required to run
|
|
# PLY in pull mode. It was designed to work incrementally and it would not be
|
|
# that hard to enable pull mode.
|
|
import re
|
|
try:
|
|
s = set()
|
|
del s
|
|
except NameError:
|
|
from Set import Set as set
|
|
|
|
from ply import lex
|
|
from bb.pysh.sherrors import *
|
|
|
|
class NeedMore(Exception):
|
|
pass
|
|
|
|
def is_blank(c):
|
|
return c in (' ', '\t')
|
|
|
|
_RE_DIGITS = re.compile(r'^\d+$')
|
|
|
|
def are_digits(s):
|
|
return _RE_DIGITS.search(s) is not None
|
|
|
|
_OPERATORS = dict([
|
|
('&&', 'AND_IF'),
|
|
('||', 'OR_IF'),
|
|
(';;', 'DSEMI'),
|
|
('<<', 'DLESS'),
|
|
('>>', 'DGREAT'),
|
|
('<&', 'LESSAND'),
|
|
('>&', 'GREATAND'),
|
|
('<>', 'LESSGREAT'),
|
|
('<<-', 'DLESSDASH'),
|
|
('>|', 'CLOBBER'),
|
|
('&', 'AMP'),
|
|
(';', 'COMMA'),
|
|
('<', 'LESS'),
|
|
('>', 'GREATER'),
|
|
('(', 'LPARENS'),
|
|
(')', 'RPARENS'),
|
|
])
|
|
|
|
#Make a function to silence pychecker "Local variable shadows global"
|
|
def make_partial_ops():
|
|
partials = {}
|
|
for k in _OPERATORS:
|
|
for i in range(1, len(k)+1):
|
|
partials[k[:i]] = None
|
|
return partials
|
|
|
|
_PARTIAL_OPERATORS = make_partial_ops()
|
|
|
|
def is_partial_op(s):
|
|
"""Return True if s matches a non-empty subpart of an operator starting
|
|
at its first character.
|
|
"""
|
|
return s in _PARTIAL_OPERATORS
|
|
|
|
def is_op(s):
|
|
"""If s matches an operator, returns the operator identifier. Return None
|
|
otherwise.
|
|
"""
|
|
return _OPERATORS.get(s)
|
|
|
|
_RESERVEDS = dict([
|
|
('if', 'If'),
|
|
('then', 'Then'),
|
|
('else', 'Else'),
|
|
('elif', 'Elif'),
|
|
('fi', 'Fi'),
|
|
('do', 'Do'),
|
|
('done', 'Done'),
|
|
('case', 'Case'),
|
|
('esac', 'Esac'),
|
|
('while', 'While'),
|
|
('until', 'Until'),
|
|
('for', 'For'),
|
|
('{', 'Lbrace'),
|
|
('}', 'Rbrace'),
|
|
('!', 'Bang'),
|
|
('in', 'In'),
|
|
('|', 'PIPE'),
|
|
])
|
|
|
|
def get_reserved(s):
|
|
return _RESERVEDS.get(s)
|
|
|
|
_RE_NAME = re.compile(r'^[0-9a-zA-Z_]+$')
|
|
|
|
def is_name(s):
|
|
return _RE_NAME.search(s) is not None
|
|
|
|
def find_chars(seq, chars):
|
|
for i,v in enumerate(seq):
|
|
if v in chars:
|
|
return i,v
|
|
return -1, None
|
|
|
|
class WordLexer:
|
|
"""WordLexer parse quoted or expansion expressions and return an expression
|
|
tree. The input string can be any well formed sequence beginning with quoting
|
|
or expansion character. Embedded expressions are handled recursively. The
|
|
resulting tree is made of lists and strings. Lists represent quoted or
|
|
expansion expressions. Each list first element is the opening separator,
|
|
the last one the closing separator. In-between can be any number of strings
|
|
or lists for sub-expressions. Non quoted/expansion expression can written as
|
|
strings or as lists with empty strings as starting and ending delimiters.
|
|
"""
|
|
|
|
NAME_CHARSET = 'abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_'
|
|
NAME_CHARSET = dict(zip(NAME_CHARSET, NAME_CHARSET))
|
|
|
|
SPECIAL_CHARSET = '@*#?-$!0'
|
|
|
|
#Characters which can be escaped depends on the current delimiters
|
|
ESCAPABLE = {
|
|
'`': set(['$', '\\', '`']),
|
|
'"': set(['$', '\\', '`', '"']),
|
|
"'": set(),
|
|
}
|
|
|
|
def __init__(self, heredoc = False):
|
|
# _buffer is the unprocessed input characters buffer
|
|
self._buffer = []
|
|
# _stack is empty or contains a quoted list being processed
|
|
# (this is the DFS path to the quoted expression being evaluated).
|
|
self._stack = []
|
|
self._escapable = None
|
|
# True when parsing unquoted here documents
|
|
self._heredoc = heredoc
|
|
|
|
def add(self, data, eof=False):
|
|
"""Feed the lexer with more data. If the quoted expression can be
|
|
delimited, return a tuple (expr, remaining) containing the expression
|
|
tree and the unconsumed data.
|
|
Otherwise, raise NeedMore.
|
|
"""
|
|
self._buffer += list(data)
|
|
self._parse(eof)
|
|
|
|
result = self._stack[0]
|
|
remaining = ''.join(self._buffer)
|
|
self._stack = []
|
|
self._buffer = []
|
|
return result, remaining
|
|
|
|
def _is_escapable(self, c, delim=None):
|
|
if delim is None:
|
|
if self._heredoc:
|
|
# Backslashes works as if they were double quoted in unquoted
|
|
# here-documents
|
|
delim = '"'
|
|
else:
|
|
if len(self._stack)<=1:
|
|
return True
|
|
delim = self._stack[-2][0]
|
|
|
|
escapables = self.ESCAPABLE.get(delim, None)
|
|
return escapables is None or c in escapables
|
|
|
|
def _parse_squote(self, buf, result, eof):
|
|
if not buf:
|
|
raise NeedMore()
|
|
try:
|
|
pos = buf.index("'")
|
|
except ValueError:
|
|
raise NeedMore()
|
|
result[-1] += ''.join(buf[:pos])
|
|
result += ["'"]
|
|
return pos+1, True
|
|
|
|
def _parse_bquote(self, buf, result, eof):
|
|
if not buf:
|
|
raise NeedMore()
|
|
|
|
if buf[0]=='\n':
|
|
#Remove line continuations
|
|
result[:] = ['', '', '']
|
|
elif self._is_escapable(buf[0]):
|
|
result[-1] += buf[0]
|
|
result += ['']
|
|
else:
|
|
#Keep as such
|
|
result[:] = ['', '\\'+buf[0], '']
|
|
|
|
return 1, True
|
|
|
|
def _parse_dquote(self, buf, result, eof):
|
|
if not buf:
|
|
raise NeedMore()
|
|
pos, sep = find_chars(buf, '$\\`"')
|
|
if pos==-1:
|
|
raise NeedMore()
|
|
|
|
result[-1] += ''.join(buf[:pos])
|
|
if sep=='"':
|
|
result += ['"']
|
|
return pos+1, True
|
|
else:
|
|
#Keep everything until the separator and defer processing
|
|
return pos, False
|
|
|
|
def _parse_command(self, buf, result, eof):
|
|
if not buf:
|
|
raise NeedMore()
|
|
|
|
chars = '$\\`"\''
|
|
if result[0] == '$(':
|
|
chars += ')'
|
|
pos, sep = find_chars(buf, chars)
|
|
if pos == -1:
|
|
raise NeedMore()
|
|
|
|
result[-1] += ''.join(buf[:pos])
|
|
if (result[0]=='$(' and sep==')') or (result[0]=='`' and sep=='`'):
|
|
result += [sep]
|
|
return pos+1, True
|
|
else:
|
|
return pos, False
|
|
|
|
def _parse_parameter(self, buf, result, eof):
|
|
if not buf:
|
|
raise NeedMore()
|
|
|
|
pos, sep = find_chars(buf, '$\\`"\'}')
|
|
if pos==-1:
|
|
raise NeedMore()
|
|
|
|
result[-1] += ''.join(buf[:pos])
|
|
if sep=='}':
|
|
result += [sep]
|
|
return pos+1, True
|
|
else:
|
|
return pos, False
|
|
|
|
def _parse_dollar(self, buf, result, eof):
|
|
sep = result[0]
|
|
if sep=='$':
|
|
if not buf:
|
|
#TODO: handle empty $
|
|
raise NeedMore()
|
|
if buf[0]=='(':
|
|
if len(buf)==1:
|
|
raise NeedMore()
|
|
|
|
if buf[1]=='(':
|
|
result[0] = '$(('
|
|
buf[:2] = []
|
|
else:
|
|
result[0] = '$('
|
|
buf[:1] = []
|
|
|
|
elif buf[0]=='{':
|
|
result[0] = '${'
|
|
buf[:1] = []
|
|
else:
|
|
if buf[0] in self.SPECIAL_CHARSET:
|
|
result[-1] = buf[0]
|
|
read = 1
|
|
else:
|
|
for read,c in enumerate(buf):
|
|
if c not in self.NAME_CHARSET:
|
|
break
|
|
else:
|
|
if not eof:
|
|
raise NeedMore()
|
|
read += 1
|
|
|
|
result[-1] += ''.join(buf[0:read])
|
|
|
|
if not result[-1]:
|
|
result[:] = ['', result[0], '']
|
|
else:
|
|
result += ['']
|
|
return read,True
|
|
|
|
sep = result[0]
|
|
if sep=='$(':
|
|
parsefunc = self._parse_command
|
|
elif sep=='${':
|
|
parsefunc = self._parse_parameter
|
|
else:
|
|
raise NotImplementedError(sep)
|
|
|
|
pos, closed = parsefunc(buf, result, eof)
|
|
return pos, closed
|
|
|
|
def _parse(self, eof):
|
|
buf = self._buffer
|
|
stack = self._stack
|
|
recurse = False
|
|
|
|
while 1:
|
|
if not stack or recurse:
|
|
if not buf:
|
|
raise NeedMore()
|
|
if buf[0] not in ('"\\`$\''):
|
|
raise ShellSyntaxError('Invalid quoted string sequence')
|
|
stack.append([buf[0], ''])
|
|
buf[:1] = []
|
|
recurse = False
|
|
|
|
result = stack[-1]
|
|
if result[0]=="'":
|
|
parsefunc = self._parse_squote
|
|
elif result[0]=='\\':
|
|
parsefunc = self._parse_bquote
|
|
elif result[0]=='"':
|
|
parsefunc = self._parse_dquote
|
|
elif result[0]=='`':
|
|
parsefunc = self._parse_command
|
|
elif result[0][0]=='$':
|
|
parsefunc = self._parse_dollar
|
|
else:
|
|
raise NotImplementedError()
|
|
|
|
read, closed = parsefunc(buf, result, eof)
|
|
|
|
buf[:read] = []
|
|
if closed:
|
|
if len(stack)>1:
|
|
#Merge in parent expression
|
|
parsed = stack.pop()
|
|
stack[-1] += [parsed]
|
|
stack[-1] += ['']
|
|
else:
|
|
break
|
|
else:
|
|
recurse = True
|
|
|
|
def normalize_wordtree(wtree):
|
|
"""Fold back every literal sequence (delimited with empty strings) into
|
|
parent sequence.
|
|
"""
|
|
def normalize(wtree):
|
|
result = []
|
|
for part in wtree[1:-1]:
|
|
if isinstance(part, list):
|
|
part = normalize(part)
|
|
if part[0]=='':
|
|
#Move the part content back at current level
|
|
result += part[1:-1]
|
|
continue
|
|
elif not part:
|
|
#Remove empty strings
|
|
continue
|
|
result.append(part)
|
|
if not result:
|
|
result = ['']
|
|
return [wtree[0]] + result + [wtree[-1]]
|
|
|
|
return normalize(wtree)
|
|
|
|
|
|
def make_wordtree(token, here_document=False):
|
|
"""Parse a delimited token and return a tree similar to the ones returned by
|
|
WordLexer. token may contain any combinations of expansion/quoted fields and
|
|
non-ones.
|
|
"""
|
|
tree = ['']
|
|
remaining = token
|
|
delimiters = '\\$`'
|
|
if not here_document:
|
|
delimiters += '\'"'
|
|
|
|
while 1:
|
|
pos, sep = find_chars(remaining, delimiters)
|
|
if pos==-1:
|
|
tree += [remaining, '']
|
|
return normalize_wordtree(tree)
|
|
tree.append(remaining[:pos])
|
|
remaining = remaining[pos:]
|
|
|
|
try:
|
|
result, remaining = WordLexer(heredoc = here_document).add(remaining, True)
|
|
except NeedMore:
|
|
raise ShellSyntaxError('Invalid token "%s"')
|
|
tree.append(result)
|
|
|
|
|
|
def wordtree_as_string(wtree):
|
|
"""Rewrite an expression tree generated by make_wordtree as string."""
|
|
def visit(node, output):
|
|
for child in node:
|
|
if isinstance(child, list):
|
|
visit(child, output)
|
|
else:
|
|
output.append(child)
|
|
|
|
output = []
|
|
visit(wtree, output)
|
|
return ''.join(output)
|
|
|
|
|
|
def unquote_wordtree(wtree):
|
|
"""Fold the word tree while removing quotes everywhere. Other expansion
|
|
sequences are joined as such.
|
|
"""
|
|
def unquote(wtree):
|
|
unquoted = []
|
|
if wtree[0] in ('', "'", '"', '\\'):
|
|
wtree = wtree[1:-1]
|
|
|
|
for part in wtree:
|
|
if isinstance(part, list):
|
|
part = unquote(part)
|
|
unquoted.append(part)
|
|
return ''.join(unquoted)
|
|
|
|
return unquote(wtree)
|
|
|
|
|
|
class HereDocLexer:
|
|
"""HereDocLexer delimits whatever comes from the here-document starting newline
|
|
not included to the closing delimiter line included.
|
|
"""
|
|
def __init__(self, op, delim):
|
|
assert op in ('<<', '<<-')
|
|
if not delim:
|
|
raise ShellSyntaxError('invalid here document delimiter %s' % str(delim))
|
|
|
|
self._op = op
|
|
self._delim = delim
|
|
self._buffer = []
|
|
self._token = []
|
|
|
|
def add(self, data, eof):
|
|
"""If the here-document was delimited, return a tuple (content, remaining).
|
|
Raise NeedMore() otherwise.
|
|
"""
|
|
self._buffer += list(data)
|
|
self._parse(eof)
|
|
token = ''.join(self._token)
|
|
remaining = ''.join(self._buffer)
|
|
self._token, self._remaining = [], []
|
|
return token, remaining
|
|
|
|
def _parse(self, eof):
|
|
while 1:
|
|
#Look for first unescaped newline. Quotes may be ignored
|
|
escaped = False
|
|
for i,c in enumerate(self._buffer):
|
|
if escaped:
|
|
escaped = False
|
|
elif c=='\\':
|
|
escaped = True
|
|
elif c=='\n':
|
|
break
|
|
else:
|
|
i = -1
|
|
|
|
if i==-1 or self._buffer[i]!='\n':
|
|
if not eof:
|
|
raise NeedMore()
|
|
#No more data, maybe the last line is closing delimiter
|
|
line = ''.join(self._buffer)
|
|
eol = ''
|
|
self._buffer[:] = []
|
|
else:
|
|
line = ''.join(self._buffer[:i])
|
|
eol = self._buffer[i]
|
|
self._buffer[:i+1] = []
|
|
|
|
if self._op=='<<-':
|
|
line = line.lstrip('\t')
|
|
|
|
if line==self._delim:
|
|
break
|
|
|
|
self._token += [line, eol]
|
|
if i==-1:
|
|
break
|
|
|
|
class Token:
|
|
#TODO: check this is still in use
|
|
OPERATOR = 'OPERATOR'
|
|
WORD = 'WORD'
|
|
|
|
def __init__(self):
|
|
self.value = ''
|
|
self.type = None
|
|
|
|
def __getitem__(self, key):
|
|
#Behave like a two elements tuple
|
|
if key==0:
|
|
return self.type
|
|
if key==1:
|
|
return self.value
|
|
raise IndexError(key)
|
|
|
|
|
|
class HereDoc:
|
|
def __init__(self, op, name=None):
|
|
self.op = op
|
|
self.name = name
|
|
self.pendings = []
|
|
|
|
TK_COMMA = 'COMMA'
|
|
TK_AMPERSAND = 'AMP'
|
|
TK_OP = 'OP'
|
|
TK_TOKEN = 'TOKEN'
|
|
TK_COMMENT = 'COMMENT'
|
|
TK_NEWLINE = 'NEWLINE'
|
|
TK_IONUMBER = 'IO_NUMBER'
|
|
TK_ASSIGNMENT = 'ASSIGNMENT_WORD'
|
|
TK_HERENAME = 'HERENAME'
|
|
|
|
class Lexer:
|
|
"""Main lexer.
|
|
|
|
Call add() until the script AST is returned.
|
|
"""
|
|
# Here-document handling makes the whole thing more complex because they basically
|
|
# force tokens to be reordered: here-content must come right after the operator
|
|
# and the here-document name, while some other tokens might be following the
|
|
# here-document expression on the same line.
|
|
#
|
|
# So, here-doc states are basically:
|
|
# *self._state==ST_NORMAL
|
|
# - self._heredoc.op is None: no here-document
|
|
# - self._heredoc.op is not None but name is: here-document operator matched,
|
|
# waiting for the document name/delimiter
|
|
# - self._heredoc.op and name are not None: here-document is ready, following
|
|
# tokens are being stored and will be pushed again when the document is
|
|
# completely parsed.
|
|
# *self._state==ST_HEREDOC
|
|
# - The here-document is being delimited by self._herelexer. Once it is done
|
|
# the content is pushed in front of the pending token list then all these
|
|
# tokens are pushed once again.
|
|
ST_NORMAL = 'ST_NORMAL'
|
|
ST_OP = 'ST_OP'
|
|
ST_BACKSLASH = 'ST_BACKSLASH'
|
|
ST_QUOTED = 'ST_QUOTED'
|
|
ST_COMMENT = 'ST_COMMENT'
|
|
ST_HEREDOC = 'ST_HEREDOC'
|
|
|
|
#Match end of backquote strings
|
|
RE_BACKQUOTE_END = re.compile(r'(?<!\\)(`)')
|
|
|
|
def __init__(self, parent_state = None):
|
|
self._input = []
|
|
self._pos = 0
|
|
|
|
self._token = ''
|
|
self._type = TK_TOKEN
|
|
|
|
self._state = self.ST_NORMAL
|
|
self._parent_state = parent_state
|
|
self._wordlexer = None
|
|
|
|
self._heredoc = HereDoc(None)
|
|
self._herelexer = None
|
|
|
|
### Following attributes are not used for delimiting token and can safely
|
|
### be changed after here-document detection (see _push_toke)
|
|
|
|
# Count the number of tokens following a 'For' reserved word. Needed to
|
|
# return an 'In' reserved word if it comes in third place.
|
|
self._for_count = None
|
|
|
|
def add(self, data, eof=False):
|
|
"""Feed the lexer with data.
|
|
|
|
When eof is set to True, returns unconsumed data or raise if the lexer
|
|
is in the middle of a delimiting operation.
|
|
Raise NeedMore otherwise.
|
|
"""
|
|
self._input += list(data)
|
|
self._parse(eof)
|
|
self._input[:self._pos] = []
|
|
return ''.join(self._input)
|
|
|
|
def _parse(self, eof):
|
|
while self._state:
|
|
if self._pos>=len(self._input):
|
|
if not eof:
|
|
raise NeedMore()
|
|
elif self._state not in (self.ST_OP, self.ST_QUOTED, self.ST_HEREDOC):
|
|
#Delimit the current token and leave cleanly
|
|
self._push_token('')
|
|
break
|
|
else:
|
|
#Let the sublexer handle the eof themselves
|
|
pass
|
|
|
|
if self._state==self.ST_NORMAL:
|
|
self._parse_normal()
|
|
elif self._state==self.ST_COMMENT:
|
|
self._parse_comment()
|
|
elif self._state==self.ST_OP:
|
|
self._parse_op(eof)
|
|
elif self._state==self.ST_QUOTED:
|
|
self._parse_quoted(eof)
|
|
elif self._state==self.ST_HEREDOC:
|
|
self._parse_heredoc(eof)
|
|
else:
|
|
assert False, "Unknown state " + str(self._state)
|
|
|
|
if self._heredoc.op is not None:
|
|
raise ShellSyntaxError('missing here-document delimiter')
|
|
|
|
def _parse_normal(self):
|
|
c = self._input[self._pos]
|
|
if c=='\n':
|
|
self._push_token(c)
|
|
self._token = c
|
|
self._type = TK_NEWLINE
|
|
self._push_token('')
|
|
self._pos += 1
|
|
elif c in ('\\', '\'', '"', '`', '$'):
|
|
self._state = self.ST_QUOTED
|
|
elif is_partial_op(c):
|
|
self._push_token(c)
|
|
|
|
self._type = TK_OP
|
|
self._token += c
|
|
self._pos += 1
|
|
self._state = self.ST_OP
|
|
elif is_blank(c):
|
|
self._push_token(c)
|
|
|
|
#Discard blanks
|
|
self._pos += 1
|
|
elif self._token:
|
|
self._token += c
|
|
self._pos += 1
|
|
elif c=='#':
|
|
self._state = self.ST_COMMENT
|
|
self._type = TK_COMMENT
|
|
self._pos += 1
|
|
else:
|
|
self._pos += 1
|
|
self._token += c
|
|
|
|
def _parse_op(self, eof):
|
|
assert self._token
|
|
|
|
while 1:
|
|
if self._pos>=len(self._input):
|
|
if not eof:
|
|
raise NeedMore()
|
|
c = ''
|
|
else:
|
|
c = self._input[self._pos]
|
|
|
|
op = self._token + c
|
|
if c and is_partial_op(op):
|
|
#Still parsing an operator
|
|
self._token = op
|
|
self._pos += 1
|
|
else:
|
|
#End of operator
|
|
self._push_token(c)
|
|
self._state = self.ST_NORMAL
|
|
break
|
|
|
|
def _parse_comment(self):
|
|
while 1:
|
|
if self._pos>=len(self._input):
|
|
raise NeedMore()
|
|
|
|
c = self._input[self._pos]
|
|
if c=='\n':
|
|
#End of comment, do not consume the end of line
|
|
self._state = self.ST_NORMAL
|
|
break
|
|
else:
|
|
self._token += c
|
|
self._pos += 1
|
|
|
|
def _parse_quoted(self, eof):
|
|
"""Precondition: the starting backquote/dollar is still in the input queue."""
|
|
if not self._wordlexer:
|
|
self._wordlexer = WordLexer()
|
|
|
|
if self._pos<len(self._input):
|
|
#Transfer input queue character into the subparser
|
|
input = self._input[self._pos:]
|
|
self._pos += len(input)
|
|
|
|
wtree, remaining = self._wordlexer.add(input, eof)
|
|
self._wordlexer = None
|
|
self._token += wordtree_as_string(wtree)
|
|
|
|
#Put unparsed character back in the input queue
|
|
if remaining:
|
|
self._input[self._pos:self._pos] = list(remaining)
|
|
self._state = self.ST_NORMAL
|
|
|
|
def _parse_heredoc(self, eof):
|
|
assert not self._token
|
|
|
|
if self._herelexer is None:
|
|
self._herelexer = HereDocLexer(self._heredoc.op, self._heredoc.name)
|
|
|
|
if self._pos<len(self._input):
|
|
#Transfer input queue character into the subparser
|
|
input = self._input[self._pos:]
|
|
self._pos += len(input)
|
|
|
|
self._token, remaining = self._herelexer.add(input, eof)
|
|
|
|
#Reset here-document state
|
|
self._herelexer = None
|
|
heredoc, self._heredoc = self._heredoc, HereDoc(None)
|
|
if remaining:
|
|
self._input[self._pos:self._pos] = list(remaining)
|
|
self._state = self.ST_NORMAL
|
|
|
|
#Push pending tokens
|
|
heredoc.pendings[:0] = [(self._token, self._type, heredoc.name)]
|
|
for token, type, delim in heredoc.pendings:
|
|
self._token = token
|
|
self._type = type
|
|
self._push_token(delim)
|
|
|
|
def _push_token(self, delim):
|
|
if not self._token:
|
|
return 0
|
|
|
|
if self._heredoc.op is not None:
|
|
if self._heredoc.name is None:
|
|
#Here-document name
|
|
if self._type!=TK_TOKEN:
|
|
raise ShellSyntaxError("expecting here-document name, got '%s'" % self._token)
|
|
self._heredoc.name = unquote_wordtree(make_wordtree(self._token))
|
|
self._type = TK_HERENAME
|
|
else:
|
|
#Capture all tokens until the newline starting the here-document
|
|
if self._type==TK_NEWLINE:
|
|
assert self._state==self.ST_NORMAL
|
|
self._state = self.ST_HEREDOC
|
|
|
|
self._heredoc.pendings.append((self._token, self._type, delim))
|
|
self._token = ''
|
|
self._type = TK_TOKEN
|
|
return 1
|
|
|
|
# BEWARE: do not change parser state from here to the end of the function:
|
|
# when parsing between an here-document operator to the end of the line
|
|
# tokens are stored in self._heredoc.pendings. Therefore, they will not
|
|
# reach the section below.
|
|
|
|
#Check operators
|
|
if self._type==TK_OP:
|
|
#False positive because of partial op matching
|
|
op = is_op(self._token)
|
|
if not op:
|
|
self._type = TK_TOKEN
|
|
else:
|
|
#Map to the specific operator
|
|
self._type = op
|
|
if self._token in ('<<', '<<-'):
|
|
#Done here rather than in _parse_op because there is no need
|
|
#to change the parser state since we are still waiting for
|
|
#the here-document name
|
|
if self._heredoc.op is not None:
|
|
raise ShellSyntaxError("syntax error near token '%s'" % self._token)
|
|
assert self._heredoc.op is None
|
|
self._heredoc.op = self._token
|
|
|
|
if self._type==TK_TOKEN:
|
|
if '=' in self._token and not delim:
|
|
if self._token.startswith('='):
|
|
#Token is a WORD... a TOKEN that is.
|
|
pass
|
|
else:
|
|
prev = self._token[:self._token.find('=')]
|
|
if is_name(prev):
|
|
self._type = TK_ASSIGNMENT
|
|
else:
|
|
#Just a token (unspecified)
|
|
pass
|
|
else:
|
|
reserved = get_reserved(self._token)
|
|
if reserved is not None:
|
|
if reserved=='In' and self._for_count!=2:
|
|
#Sorry, not a reserved word after all
|
|
pass
|
|
else:
|
|
self._type = reserved
|
|
if reserved in ('For', 'Case'):
|
|
self._for_count = 0
|
|
elif are_digits(self._token) and delim in ('<', '>'):
|
|
#Detect IO_NUMBER
|
|
self._type = TK_IONUMBER
|
|
elif self._token==';':
|
|
self._type = TK_COMMA
|
|
elif self._token=='&':
|
|
self._type = TK_AMPERSAND
|
|
elif self._type==TK_COMMENT:
|
|
#Comments are not part of sh grammar, ignore them
|
|
self._token = ''
|
|
self._type = TK_TOKEN
|
|
return 0
|
|
|
|
if self._for_count is not None:
|
|
#Track token count in 'For' expression to detect 'In' reserved words.
|
|
#Can only be in third position, no need to go beyond
|
|
self._for_count += 1
|
|
if self._for_count==3:
|
|
self._for_count = None
|
|
|
|
self.on_token((self._token, self._type))
|
|
self._token = ''
|
|
self._type = TK_TOKEN
|
|
return 1
|
|
|
|
def on_token(self, token):
|
|
raise NotImplementedError
|
|
|
|
|
|
tokens = [
|
|
TK_TOKEN,
|
|
# To silence yacc unused token warnings
|
|
# TK_COMMENT,
|
|
TK_NEWLINE,
|
|
TK_IONUMBER,
|
|
TK_ASSIGNMENT,
|
|
TK_HERENAME,
|
|
]
|
|
|
|
#Add specific operators
|
|
tokens += _OPERATORS.values()
|
|
#Add reserved words
|
|
tokens += _RESERVEDS.values()
|
|
|
|
class PLYLexer(Lexer):
|
|
"""Bridge Lexer and PLY lexer interface."""
|
|
def __init__(self):
|
|
Lexer.__init__(self)
|
|
self._tokens = []
|
|
self._current = 0
|
|
self.lineno = 0
|
|
|
|
def on_token(self, token):
|
|
value, type = token
|
|
|
|
self.lineno = 0
|
|
t = lex.LexToken()
|
|
t.value = value
|
|
t.type = type
|
|
t.lexer = self
|
|
t.lexpos = 0
|
|
t.lineno = 0
|
|
|
|
self._tokens.append(t)
|
|
|
|
def is_empty(self):
|
|
return not bool(self._tokens)
|
|
|
|
#PLY compliant interface
|
|
def token(self):
|
|
if self._current>=len(self._tokens):
|
|
return None
|
|
t = self._tokens[self._current]
|
|
self._current += 1
|
|
return t
|
|
|
|
|
|
def get_tokens(s):
|
|
"""Parse the input string and return a tuple (tokens, unprocessed) where
|
|
tokens is a list of parsed tokens and unprocessed is the part of the input
|
|
string left untouched by the lexer.
|
|
"""
|
|
lexer = PLYLexer()
|
|
untouched = lexer.add(s, True)
|
|
tokens = []
|
|
while 1:
|
|
token = lexer.token()
|
|
if token is None:
|
|
break
|
|
tokens.append(token)
|
|
|
|
tokens = [(t.value, t.type) for t in tokens]
|
|
return tokens, untouched
|