Poodletooth-iLand/panda/python/Lib/site-packages/ply/cpp.py
Master Jumblespeed d882959bfa switch to remote
2015-05-18 22:11:33 -04:00

898 lines
32 KiB
Python
Executable file

# -----------------------------------------------------------------------------
# cpp.py
#
# Author: David Beazley (http://www.dabeaz.com)
# Copyright (C) 2007
# All rights reserved
#
# This module implements an ANSI-C style lexical preprocessor for PLY.
# -----------------------------------------------------------------------------
from __future__ import generators
# -----------------------------------------------------------------------------
# Default preprocessor lexer definitions. These tokens are enough to get
# a basic preprocessor working. Other modules may import these if they want
# -----------------------------------------------------------------------------
tokens = (
'CPP_ID','CPP_INTEGER', 'CPP_FLOAT', 'CPP_STRING', 'CPP_CHAR', 'CPP_WS', 'CPP_COMMENT', 'CPP_POUND','CPP_DPOUND'
)
literals = "+-*/%|&~^<>=!?()[]{}.,;:\\\'\""
# Whitespace
def t_CPP_WS(t):
r'\s+'
t.lexer.lineno += t.value.count("\n")
return t
t_CPP_POUND = r'\#'
t_CPP_DPOUND = r'\#\#'
# Identifier
t_CPP_ID = r'[A-Za-z_][\w_]*'
# Integer literal
def CPP_INTEGER(t):
r'(((((0x)|(0X))[0-9a-fA-F]+)|(\d+))([uU]|[lL]|[uU][lL]|[lL][uU])?)'
return t
t_CPP_INTEGER = CPP_INTEGER
# Floating literal
t_CPP_FLOAT = r'((\d+)(\.\d+)(e(\+|-)?(\d+))? | (\d+)e(\+|-)?(\d+))([lL]|[fF])?'
# String literal
def t_CPP_STRING(t):
r'\"([^\\\n]|(\\(.|\n)))*?\"'
t.lexer.lineno += t.value.count("\n")
return t
# Character constant 'c' or L'c'
def t_CPP_CHAR(t):
r'(L)?\'([^\\\n]|(\\(.|\n)))*?\''
t.lexer.lineno += t.value.count("\n")
return t
# Comment
def t_CPP_COMMENT(t):
r'(/\*(.|\n)*?\*/)|(//.*?\n)'
t.lexer.lineno += t.value.count("\n")
return t
def t_error(t):
t.type = t.value[0]
t.value = t.value[0]
t.lexer.skip(1)
return t
import re
import copy
import time
import os.path
# -----------------------------------------------------------------------------
# trigraph()
#
# Given an input string, this function replaces all trigraph sequences.
# The following mapping is used:
#
# ??= #
# ??/ \
# ??' ^
# ??( [
# ??) ]
# ??! |
# ??< {
# ??> }
# ??- ~
# -----------------------------------------------------------------------------
_trigraph_pat = re.compile(r'''\?\?[=/\'\(\)\!<>\-]''')
_trigraph_rep = {
'=':'#',
'/':'\\',
"'":'^',
'(':'[',
')':']',
'!':'|',
'<':'{',
'>':'}',
'-':'~'
}
def trigraph(input):
return _trigraph_pat.sub(lambda g: _trigraph_rep[g.group()[-1]],input)
# ------------------------------------------------------------------
# Macro object
#
# This object holds information about preprocessor macros
#
# .name - Macro name (string)
# .value - Macro value (a list of tokens)
# .arglist - List of argument names
# .variadic - Boolean indicating whether or not variadic macro
# .vararg - Name of the variadic parameter
#
# When a macro is created, the macro replacement token sequence is
# pre-scanned and used to create patch lists that are later used
# during macro expansion
# ------------------------------------------------------------------
class Macro(object):
def __init__(self,name,value,arglist=None,variadic=False):
self.name = name
self.value = value
self.arglist = arglist
self.variadic = variadic
if variadic:
self.vararg = arglist[-1]
self.source = None
# ------------------------------------------------------------------
# Preprocessor object
#
# Object representing a preprocessor. Contains macro definitions,
# include directories, and other information
# ------------------------------------------------------------------
class Preprocessor(object):
def __init__(self,lexer=None):
if lexer is None:
lexer = lex.lexer
self.lexer = lexer
self.macros = { }
self.path = []
self.temp_path = []
# Probe the lexer for selected tokens
self.lexprobe()
tm = time.localtime()
self.define("__DATE__ \"%s\"" % time.strftime("%b %d %Y",tm))
self.define("__TIME__ \"%s\"" % time.strftime("%H:%M:%S",tm))
self.parser = None
# -----------------------------------------------------------------------------
# tokenize()
#
# Utility function. Given a string of text, tokenize into a list of tokens
# -----------------------------------------------------------------------------
def tokenize(self,text):
tokens = []
self.lexer.input(text)
while True:
tok = self.lexer.token()
if not tok: break
tokens.append(tok)
return tokens
# ---------------------------------------------------------------------
# error()
#
# Report a preprocessor error/warning of some kind
# ----------------------------------------------------------------------
def error(self,file,line,msg):
print("%s:%d %s" % (file,line,msg))
# ----------------------------------------------------------------------
# lexprobe()
#
# This method probes the preprocessor lexer object to discover
# the token types of symbols that are important to the preprocessor.
# If this works right, the preprocessor will simply "work"
# with any suitable lexer regardless of how tokens have been named.
# ----------------------------------------------------------------------
def lexprobe(self):
# Determine the token type for identifiers
self.lexer.input("identifier")
tok = self.lexer.token()
if not tok or tok.value != "identifier":
print("Couldn't determine identifier type")
else:
self.t_ID = tok.type
# Determine the token type for integers
self.lexer.input("12345")
tok = self.lexer.token()
if not tok or int(tok.value) != 12345:
print("Couldn't determine integer type")
else:
self.t_INTEGER = tok.type
self.t_INTEGER_TYPE = type(tok.value)
# Determine the token type for strings enclosed in double quotes
self.lexer.input("\"filename\"")
tok = self.lexer.token()
if not tok or tok.value != "\"filename\"":
print("Couldn't determine string type")
else:
self.t_STRING = tok.type
# Determine the token type for whitespace--if any
self.lexer.input(" ")
tok = self.lexer.token()
if not tok or tok.value != " ":
self.t_SPACE = None
else:
self.t_SPACE = tok.type
# Determine the token type for newlines
self.lexer.input("\n")
tok = self.lexer.token()
if not tok or tok.value != "\n":
self.t_NEWLINE = None
print("Couldn't determine token for newlines")
else:
self.t_NEWLINE = tok.type
self.t_WS = (self.t_SPACE, self.t_NEWLINE)
# Check for other characters used by the preprocessor
chars = [ '<','>','#','##','\\','(',')',',','.']
for c in chars:
self.lexer.input(c)
tok = self.lexer.token()
if not tok or tok.value != c:
print("Unable to lex '%s' required for preprocessor" % c)
# ----------------------------------------------------------------------
# add_path()
#
# Adds a search path to the preprocessor.
# ----------------------------------------------------------------------
def add_path(self,path):
self.path.append(path)
# ----------------------------------------------------------------------
# group_lines()
#
# Given an input string, this function splits it into lines. Trailing whitespace
# is removed. Any line ending with \ is grouped with the next line. This
# function forms the lowest level of the preprocessor---grouping into text into
# a line-by-line format.
# ----------------------------------------------------------------------
def group_lines(self,input):
lex = self.lexer.clone()
lines = [x.rstrip() for x in input.splitlines()]
for i in xrange(len(lines)):
j = i+1
while lines[i].endswith('\\') and (j < len(lines)):
lines[i] = lines[i][:-1]+lines[j]
lines[j] = ""
j += 1
input = "\n".join(lines)
lex.input(input)
lex.lineno = 1
current_line = []
while True:
tok = lex.token()
if not tok:
break
current_line.append(tok)
if tok.type in self.t_WS and '\n' in tok.value:
yield current_line
current_line = []
if current_line:
yield current_line
# ----------------------------------------------------------------------
# tokenstrip()
#
# Remove leading/trailing whitespace tokens from a token list
# ----------------------------------------------------------------------
def tokenstrip(self,tokens):
i = 0
while i < len(tokens) and tokens[i].type in self.t_WS:
i += 1
del tokens[:i]
i = len(tokens)-1
while i >= 0 and tokens[i].type in self.t_WS:
i -= 1
del tokens[i+1:]
return tokens
# ----------------------------------------------------------------------
# collect_args()
#
# Collects comma separated arguments from a list of tokens. The arguments
# must be enclosed in parenthesis. Returns a tuple (tokencount,args,positions)
# where tokencount is the number of tokens consumed, args is a list of arguments,
# and positions is a list of integers containing the starting index of each
# argument. Each argument is represented by a list of tokens.
#
# When collecting arguments, leading and trailing whitespace is removed
# from each argument.
#
# This function properly handles nested parenthesis and commas---these do not
# define new arguments.
# ----------------------------------------------------------------------
def collect_args(self,tokenlist):
args = []
positions = []
current_arg = []
nesting = 1
tokenlen = len(tokenlist)
# Search for the opening '('.
i = 0
while (i < tokenlen) and (tokenlist[i].type in self.t_WS):
i += 1
if (i < tokenlen) and (tokenlist[i].value == '('):
positions.append(i+1)
else:
self.error(self.source,tokenlist[0].lineno,"Missing '(' in macro arguments")
return 0, [], []
i += 1
while i < tokenlen:
t = tokenlist[i]
if t.value == '(':
current_arg.append(t)
nesting += 1
elif t.value == ')':
nesting -= 1
if nesting == 0:
if current_arg:
args.append(self.tokenstrip(current_arg))
positions.append(i)
return i+1,args,positions
current_arg.append(t)
elif t.value == ',' and nesting == 1:
args.append(self.tokenstrip(current_arg))
positions.append(i+1)
current_arg = []
else:
current_arg.append(t)
i += 1
# Missing end argument
self.error(self.source,tokenlist[-1].lineno,"Missing ')' in macro arguments")
return 0, [],[]
# ----------------------------------------------------------------------
# macro_prescan()
#
# Examine the macro value (token sequence) and identify patch points
# This is used to speed up macro expansion later on---we'll know
# right away where to apply patches to the value to form the expansion
# ----------------------------------------------------------------------
def macro_prescan(self,macro):
macro.patch = [] # Standard macro arguments
macro.str_patch = [] # String conversion expansion
macro.var_comma_patch = [] # Variadic macro comma patch
i = 0
while i < len(macro.value):
if macro.value[i].type == self.t_ID and macro.value[i].value in macro.arglist:
argnum = macro.arglist.index(macro.value[i].value)
# Conversion of argument to a string
if i > 0 and macro.value[i-1].value == '#':
macro.value[i] = copy.copy(macro.value[i])
macro.value[i].type = self.t_STRING
del macro.value[i-1]
macro.str_patch.append((argnum,i-1))
continue
# Concatenation
elif (i > 0 and macro.value[i-1].value == '##'):
macro.patch.append(('c',argnum,i-1))
del macro.value[i-1]
continue
elif ((i+1) < len(macro.value) and macro.value[i+1].value == '##'):
macro.patch.append(('c',argnum,i))
i += 1
continue
# Standard expansion
else:
macro.patch.append(('e',argnum,i))
elif macro.value[i].value == '##':
if macro.variadic and (i > 0) and (macro.value[i-1].value == ',') and \
((i+1) < len(macro.value)) and (macro.value[i+1].type == self.t_ID) and \
(macro.value[i+1].value == macro.vararg):
macro.var_comma_patch.append(i-1)
i += 1
macro.patch.sort(key=lambda x: x[2],reverse=True)
# ----------------------------------------------------------------------
# macro_expand_args()
#
# Given a Macro and list of arguments (each a token list), this method
# returns an expanded version of a macro. The return value is a token sequence
# representing the replacement macro tokens
# ----------------------------------------------------------------------
def macro_expand_args(self,macro,args):
# Make a copy of the macro token sequence
rep = [copy.copy(_x) for _x in macro.value]
# Make string expansion patches. These do not alter the length of the replacement sequence
str_expansion = {}
for argnum, i in macro.str_patch:
if argnum not in str_expansion:
str_expansion[argnum] = ('"%s"' % "".join([x.value for x in args[argnum]])).replace("\\","\\\\")
rep[i] = copy.copy(rep[i])
rep[i].value = str_expansion[argnum]
# Make the variadic macro comma patch. If the variadic macro argument is empty, we get rid
comma_patch = False
if macro.variadic and not args[-1]:
for i in macro.var_comma_patch:
rep[i] = None
comma_patch = True
# Make all other patches. The order of these matters. It is assumed that the patch list
# has been sorted in reverse order of patch location since replacements will cause the
# size of the replacement sequence to expand from the patch point.
expanded = { }
for ptype, argnum, i in macro.patch:
# Concatenation. Argument is left unexpanded
if ptype == 'c':
rep[i:i+1] = args[argnum]
# Normal expansion. Argument is macro expanded first
elif ptype == 'e':
if argnum not in expanded:
expanded[argnum] = self.expand_macros(args[argnum])
rep[i:i+1] = expanded[argnum]
# Get rid of removed comma if necessary
if comma_patch:
rep = [_i for _i in rep if _i]
return rep
# ----------------------------------------------------------------------
# expand_macros()
#
# Given a list of tokens, this function performs macro expansion.
# The expanded argument is a dictionary that contains macros already
# expanded. This is used to prevent infinite recursion.
# ----------------------------------------------------------------------
def expand_macros(self,tokens,expanded=None):
if expanded is None:
expanded = {}
i = 0
while i < len(tokens):
t = tokens[i]
if t.type == self.t_ID:
if t.value in self.macros and t.value not in expanded:
# Yes, we found a macro match
expanded[t.value] = True
m = self.macros[t.value]
if not m.arglist:
# A simple macro
ex = self.expand_macros([copy.copy(_x) for _x in m.value],expanded)
for e in ex:
e.lineno = t.lineno
tokens[i:i+1] = ex
i += len(ex)
else:
# A macro with arguments
j = i + 1
while j < len(tokens) and tokens[j].type in self.t_WS:
j += 1
if tokens[j].value == '(':
tokcount,args,positions = self.collect_args(tokens[j:])
if not m.variadic and len(args) != len(m.arglist):
self.error(self.source,t.lineno,"Macro %s requires %d arguments" % (t.value,len(m.arglist)))
i = j + tokcount
elif m.variadic and len(args) < len(m.arglist)-1:
if len(m.arglist) > 2:
self.error(self.source,t.lineno,"Macro %s must have at least %d arguments" % (t.value, len(m.arglist)-1))
else:
self.error(self.source,t.lineno,"Macro %s must have at least %d argument" % (t.value, len(m.arglist)-1))
i = j + tokcount
else:
if m.variadic:
if len(args) == len(m.arglist)-1:
args.append([])
else:
args[len(m.arglist)-1] = tokens[j+positions[len(m.arglist)-1]:j+tokcount-1]
del args[len(m.arglist):]
# Get macro replacement text
rep = self.macro_expand_args(m,args)
rep = self.expand_macros(rep,expanded)
for r in rep:
r.lineno = t.lineno
tokens[i:j+tokcount] = rep
i += len(rep)
del expanded[t.value]
continue
elif t.value == '__LINE__':
t.type = self.t_INTEGER
t.value = self.t_INTEGER_TYPE(t.lineno)
i += 1
return tokens
# ----------------------------------------------------------------------
# evalexpr()
#
# Evaluate an expression token sequence for the purposes of evaluating
# integral expressions.
# ----------------------------------------------------------------------
def evalexpr(self,tokens):
# tokens = tokenize(line)
# Search for defined macros
i = 0
while i < len(tokens):
if tokens[i].type == self.t_ID and tokens[i].value == 'defined':
j = i + 1
needparen = False
result = "0L"
while j < len(tokens):
if tokens[j].type in self.t_WS:
j += 1
continue
elif tokens[j].type == self.t_ID:
if tokens[j].value in self.macros:
result = "1L"
else:
result = "0L"
if not needparen: break
elif tokens[j].value == '(':
needparen = True
elif tokens[j].value == ')':
break
else:
self.error(self.source,tokens[i].lineno,"Malformed defined()")
j += 1
tokens[i].type = self.t_INTEGER
tokens[i].value = self.t_INTEGER_TYPE(result)
del tokens[i+1:j+1]
i += 1
tokens = self.expand_macros(tokens)
for i,t in enumerate(tokens):
if t.type == self.t_ID:
tokens[i] = copy.copy(t)
tokens[i].type = self.t_INTEGER
tokens[i].value = self.t_INTEGER_TYPE("0L")
elif t.type == self.t_INTEGER:
tokens[i] = copy.copy(t)
# Strip off any trailing suffixes
tokens[i].value = str(tokens[i].value)
while tokens[i].value[-1] not in "0123456789abcdefABCDEF":
tokens[i].value = tokens[i].value[:-1]
expr = "".join([str(x.value) for x in tokens])
expr = expr.replace("&&"," and ")
expr = expr.replace("||"," or ")
expr = expr.replace("!"," not ")
try:
result = eval(expr)
except StandardError:
self.error(self.source,tokens[0].lineno,"Couldn't evaluate expression")
result = 0
return result
# ----------------------------------------------------------------------
# parsegen()
#
# Parse an input string/
# ----------------------------------------------------------------------
def parsegen(self,input,source=None):
# Replace trigraph sequences
t = trigraph(input)
lines = self.group_lines(t)
if not source:
source = ""
self.define("__FILE__ \"%s\"" % source)
self.source = source
chunk = []
enable = True
iftrigger = False
ifstack = []
for x in lines:
for i,tok in enumerate(x):
if tok.type not in self.t_WS: break
if tok.value == '#':
# Preprocessor directive
for tok in x:
if tok in self.t_WS and '\n' in tok.value:
chunk.append(tok)
dirtokens = self.tokenstrip(x[i+1:])
if dirtokens:
name = dirtokens[0].value
args = self.tokenstrip(dirtokens[1:])
else:
name = ""
args = []
if name == 'define':
if enable:
for tok in self.expand_macros(chunk):
yield tok
chunk = []
self.define(args)
elif name == 'include':
if enable:
for tok in self.expand_macros(chunk):
yield tok
chunk = []
oldfile = self.macros['__FILE__']
for tok in self.include(args):
yield tok
self.macros['__FILE__'] = oldfile
self.source = source
elif name == 'undef':
if enable:
for tok in self.expand_macros(chunk):
yield tok
chunk = []
self.undef(args)
elif name == 'ifdef':
ifstack.append((enable,iftrigger))
if enable:
if not args[0].value in self.macros:
enable = False
iftrigger = False
else:
iftrigger = True
elif name == 'ifndef':
ifstack.append((enable,iftrigger))
if enable:
if args[0].value in self.macros:
enable = False
iftrigger = False
else:
iftrigger = True
elif name == 'if':
ifstack.append((enable,iftrigger))
if enable:
result = self.evalexpr(args)
if not result:
enable = False
iftrigger = False
else:
iftrigger = True
elif name == 'elif':
if ifstack:
if ifstack[-1][0]: # We only pay attention if outer "if" allows this
if enable: # If already true, we flip enable False
enable = False
elif not iftrigger: # If False, but not triggered yet, we'll check expression
result = self.evalexpr(args)
if result:
enable = True
iftrigger = True
else:
self.error(self.source,dirtokens[0].lineno,"Misplaced #elif")
elif name == 'else':
if ifstack:
if ifstack[-1][0]:
if enable:
enable = False
elif not iftrigger:
enable = True
iftrigger = True
else:
self.error(self.source,dirtokens[0].lineno,"Misplaced #else")
elif name == 'endif':
if ifstack:
enable,iftrigger = ifstack.pop()
else:
self.error(self.source,dirtokens[0].lineno,"Misplaced #endif")
else:
# Unknown preprocessor directive
pass
else:
# Normal text
if enable:
chunk.extend(x)
for tok in self.expand_macros(chunk):
yield tok
chunk = []
# ----------------------------------------------------------------------
# include()
#
# Implementation of file-inclusion
# ----------------------------------------------------------------------
def include(self,tokens):
# Try to extract the filename and then process an include file
if not tokens:
return
if tokens:
if tokens[0].value != '<' and tokens[0].type != self.t_STRING:
tokens = self.expand_macros(tokens)
if tokens[0].value == '<':
# Include <...>
i = 1
while i < len(tokens):
if tokens[i].value == '>':
break
i += 1
else:
print("Malformed #include <...>")
return
filename = "".join([x.value for x in tokens[1:i]])
path = self.path + [""] + self.temp_path
elif tokens[0].type == self.t_STRING:
filename = tokens[0].value[1:-1]
path = self.temp_path + [""] + self.path
else:
print("Malformed #include statement")
return
for p in path:
iname = os.path.join(p,filename)
try:
data = open(iname,"r").read()
dname = os.path.dirname(iname)
if dname:
self.temp_path.insert(0,dname)
for tok in self.parsegen(data,filename):
yield tok
if dname:
del self.temp_path[0]
break
except IOError:
pass
else:
print("Couldn't find '%s'" % filename)
# ----------------------------------------------------------------------
# define()
#
# Define a new macro
# ----------------------------------------------------------------------
def define(self,tokens):
if isinstance(tokens,(str,unicode)):
tokens = self.tokenize(tokens)
linetok = tokens
try:
name = linetok[0]
if len(linetok) > 1:
mtype = linetok[1]
else:
mtype = None
if not mtype:
m = Macro(name.value,[])
self.macros[name.value] = m
elif mtype.type in self.t_WS:
# A normal macro
m = Macro(name.value,self.tokenstrip(linetok[2:]))
self.macros[name.value] = m
elif mtype.value == '(':
# A macro with arguments
tokcount, args, positions = self.collect_args(linetok[1:])
variadic = False
for a in args:
if variadic:
print("No more arguments may follow a variadic argument")
break
astr = "".join([str(_i.value) for _i in a])
if astr == "...":
variadic = True
a[0].type = self.t_ID
a[0].value = '__VA_ARGS__'
variadic = True
del a[1:]
continue
elif astr[-3:] == "..." and a[0].type == self.t_ID:
variadic = True
del a[1:]
# If, for some reason, "." is part of the identifier, strip off the name for the purposes
# of macro expansion
if a[0].value[-3:] == '...':
a[0].value = a[0].value[:-3]
continue
if len(a) > 1 or a[0].type != self.t_ID:
print("Invalid macro argument")
break
else:
mvalue = self.tokenstrip(linetok[1+tokcount:])
i = 0
while i < len(mvalue):
if i+1 < len(mvalue):
if mvalue[i].type in self.t_WS and mvalue[i+1].value == '##':
del mvalue[i]
continue
elif mvalue[i].value == '##' and mvalue[i+1].type in self.t_WS:
del mvalue[i+1]
i += 1
m = Macro(name.value,mvalue,[x[0].value for x in args],variadic)
self.macro_prescan(m)
self.macros[name.value] = m
else:
print("Bad macro definition")
except LookupError:
print("Bad macro definition")
# ----------------------------------------------------------------------
# undef()
#
# Undefine a macro
# ----------------------------------------------------------------------
def undef(self,tokens):
id = tokens[0].value
try:
del self.macros[id]
except LookupError:
pass
# ----------------------------------------------------------------------
# parse()
#
# Parse input text.
# ----------------------------------------------------------------------
def parse(self,input,source=None,ignore={}):
self.ignore = ignore
self.parser = self.parsegen(input,source)
# ----------------------------------------------------------------------
# token()
#
# Method to return individual tokens
# ----------------------------------------------------------------------
def token(self):
try:
while True:
tok = next(self.parser)
if tok.type not in self.ignore: return tok
except StopIteration:
self.parser = None
return None
if __name__ == '__main__':
import ply.lex as lex
lexer = lex.lex()
# Run a preprocessor
import sys
f = open(sys.argv[1])
input = f.read()
p = Preprocessor(lexer)
p.parse(input,sys.argv[1])
while True:
tok = p.token()
if not tok: break
print(p.source, tok)