## @file
# This file is used to parse and evaluate expression in directive or PCD value.
#
# Copyright (c) 2011, Intel Corporation. All rights reserved.<BR>
# This program and the accompanying materials
# are licensed and made available under the terms and conditions of the BSD License
# which accompanies this distribution. The full text of the license may be found at
# http://opensource.org/licenses/bsd-license.php
#
# THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS,
# WITHOUT WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED.
## Import Modules
#
from Common.GlobalData import *
from CommonDataClass.Exceptions import BadExpression
from CommonDataClass.Exceptions import WrnExpression
from Misc import GuidStringToGuidStructureString
ERR_STRING_EXPR = 'This operator cannot be used in string expression: [%s].'
ERR_SNYTAX = 'Syntax error, the rest of expression cannot be evaluated: [%s].'
ERR_MATCH = 'No matching right parenthesis.'
ERR_STRING_TOKEN = 'Bad string token: [%s].'
ERR_MACRO_TOKEN = 'Bad macro token: [%s].'
ERR_EMPTY_TOKEN = 'Empty token is not allowed.'
ERR_PCD_RESOLVE = 'PCD token cannot be resolved: [%s].'
ERR_VALID_TOKEN = 'No more valid token found from rest of string: [%s].'
ERR_EXPR_TYPE = 'Different types found in expression.'
ERR_OPERATOR_UNSUPPORT = 'Unsupported operator: [%s]'
ERR_REL_NOT_IN = 'Expect "IN" after "not" operator.'
WRN_BOOL_EXPR = 'Operand of boolean type cannot be used in arithmetic expression.'
WRN_EQCMP_STR_OTHERS = '== Comparison between Operand of string type and Boolean/Number Type always return False.'
WRN_NECMP_STR_OTHERS = '!= Comparison between Operand of string type and Boolean/Number Type always return True.'
ERR_RELCMP_STR_OTHERS = 'Operator taking Operand of string type and Boolean/Number Type is not allowed: [%s].'
ERR_STRING_CMP = 'Unicode string and general string cannot be compared: [%s %s %s]'
ERR_ARRAY_TOKEN = 'Bad C array or C format GUID token: [%s].'
ERR_ARRAY_ELE = 'This must be HEX value for NList or Array: [%s].'
ERR_EMPTY_EXPR = 'Empty expression is not allowed.'
ERR_IN_OPERAND = 'Macro after IN operator can only be: $(FAMILY), $(ARCH), $(TOOL_CHAIN_TAG) and $(TARGET).'
## SplitString
# Split string to list according double quote
# For example: abc"de\"f"ghi"jkl"mn will be: ['abc', '"de\"f"', 'ghi', '"jkl"', 'mn']
#
def SplitString(String):
# There might be escaped quote: "abc\"def\\\"ghi"
Str = String.replace('\\\\', '//').replace('\\\"', '\\\'')
RetList = []
InQuote = False
Item = ''
for i, ch in enumerate(Str):
if ch == '"':
InQuote = not InQuote
if not InQuote:
Item += String[i]
RetList.append(Item)
Item = ''
continue
if Item:
RetList.append(Item)
Item = ''
Item += String[i]
if InQuote:
raise BadExpression(ERR_STRING_TOKEN % Item)
if Item:
RetList.append(Item)
return RetList
## ReplaceExprMacro
#
def ReplaceExprMacro(String, Macros, ExceptionList = None):
StrList = SplitString(String)
for i, String in enumerate(StrList):
InQuote = False
if String.startswith('"'):
InQuote = True
MacroStartPos = String.find('$(')
if MacroStartPos < 0:
continue
RetStr = ''
while MacroStartPos >= 0:
RetStr = String[0:MacroStartPos]
MacroEndPos = String.find(')', MacroStartPos)
if MacroEndPos < 0:
raise BadExpression(ERR_MACRO_TOKEN % String[MacroStartPos:])
Macro = String[MacroStartPos+2:MacroEndPos]
if Macro not in Macros:
# From C reference manual:
# If an undefined macro name appears in the constant-expression of
# !if or !elif, it is replaced by the integer constant 0.
RetStr += '0'
elif not InQuote:
Tklst = RetStr.split()
if Tklst and Tklst[-1] in ['IN', 'in'] and ExceptionList and Macro not in ExceptionList:
raise BadExpression(ERR_IN_OPERAND)
# Make sure the macro in exception list is encapsulated by double quote
# For example: DEFINE ARCH = IA32 X64
# $(ARCH) is replaced with "IA32 X64"
if ExceptionList and Macro in ExceptionList:
RetStr += '"' + Macros[Macro] + '"'
elif Macros[Macro].strip():
RetStr += Macros[Macro]
else:
RetStr += '""'
else:
RetStr += Macros[Macro]
RetStr += String[MacroEndPos+1:]
String = RetStr
MacroStartPos = String.find('$(')
StrList[i] = RetStr
return ''.join(StrList)
SupportedInMacroList = ['TARGET', 'TOOL_CHAIN_TAG', 'ARCH', 'FAMILY']
class ValueExpression(object):
# Logical operator mapping
LogicalOperators = {
'&&' : 'and', '||' : 'or',
'!' : 'not', 'AND': 'and',
'OR' : 'or' , 'NOT': 'not',
'XOR': '^' , 'xor': '^',
'EQ' : '==' , 'NE' : '!=',
'GT' : '>' , 'LT' : '<',
'GE' : '>=' , 'LE' : '<=',
'IN' : 'in'
}
NonLetterOpLst = ['+', '-', '&', '|', '^', '!', '=', '>', '<']
PcdPattern = re.compile(r'[_a-zA-Z][0-9A-Za-z_]*\.[_a-zA-Z][0-9A-Za-z_]*$')
HexPattern = re.compile(r'0[xX][0-9a-fA-F]+$')
RegGuidPattern = re.compile(r'[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}')
SymbolPattern = re.compile("("
"\$\([A-Z][A-Z0-9_]*\)|\$\(\w+\.\w+\)|\w+\.\w+|"
"&&|\|\||!(?!=)|"
"(?<=\W)AND(?=\W)|(?<=\W)OR(?=\W)|(?<=\W)NOT(?=\W)|(?<=\W)XOR(?=\W)|"
"(?<=\W)EQ(?=\W)|(?<=\W)NE(?=\W)|(?<=\W)GT(?=\W)|(?<=\W)LT(?=\W)|(?<=\W)GE(?=\W)|(?<=\W)LE(?=\W)"
")")
@staticmethod
def Eval(Operator, Oprand1, Oprand2 = None):
WrnExp = None
if Operator not in ["==", "!=", ">=", "<=", ">", "<", "in", "not in"] and \
(type(Oprand1) == type('') or type(Oprand2) == type('')):
raise BadExpression(ERR_STRING_EXPR % Operator)
TypeDict = {
type(0) : 0,
type(0L) : 0,
type('') : 1,
type(True) : 2
}
EvalStr = ''
if Operator in ["!", "NOT", "not"]:
if type(Oprand1) == type(''):
raise BadExpression(ERR_STRING_EXPR % Operator)
EvalStr = 'not Oprand1'
else:
if Operator in ["+", "-"] and (type(True) in [type(Oprand1), type(Oprand2)]):
# Boolean in '+'/'-' will be evaluated but raise warning
WrnExp = WrnExpression(WRN_BOOL_EXPR)
elif type('') in [type(Oprand1), type(Oprand2)] and type(Oprand1)!= type(Oprand2):
# == between string and number/boolean will always return False, != return True
if Operator == "==":
WrnExp = WrnExpression(WRN_EQCMP_STR_OTHERS)
WrnExp.result = False
raise WrnExp
elif Operator == "!=":
WrnExp = WrnExpression(WRN_NECMP_STR_OTHERS)
WrnExp.result = True
raise WrnExp
else:
raise BadExpression(ERR_RELCMP_STR_OTHERS % Operator)
elif TypeDict[type(Oprand1)] != TypeDict[type(Oprand2)]:
if Operator in ["==", "!=", ">=", "<=", ">", "<"] and set((TypeDict[type(Oprand1)], TypeDict[type(Oprand2)])) == set((TypeDict[type(True)], TypeDict[type(0)])):
# comparison between number and boolean is allowed
pass
elif Operator in ['&', '|', '^', "and", "or"] and set((TypeDict[type(Oprand1)], TypeDict[type(Oprand2)])) == set((TypeDict[type(True)], TypeDict[type(0)])):
# bitwise and logical operation between number and boolean is allowed
pass
else:
raise BadExpression(ERR_EXPR_TYPE)
if type(Oprand1) == type('') and type(Oprand2) == type(''):
if (Oprand1.startswith('L"') and not Oprand2.startswith('L"')) or \
(not Oprand1.startswith('L"') and Oprand2.startswith('L"')):
raise BadExpression(ERR_STRING_CMP % (Oprand1, Operator, Oprand2))
if 'in' in Operator and type(Oprand2) == type(''):
Oprand2 = Oprand2.split()
EvalStr = 'Oprand1 ' + Operator + ' Oprand2'
# Local symbols used by built in eval function
Dict = {
'Oprand1' : Oprand1,
'Oprand2' : Oprand2
}
try:
Val = eval(EvalStr, {}, Dict)
except Exception, Excpt:
raise BadExpression(str(Excpt))
if Operator in ['and', 'or']:
if Val:
Val = True
else:
Val = False
if WrnExp:
WrnExp.result = Val
raise WrnExp
return Val
def __init__(self, Expression, SymbolTable={}):
self._NoProcess = False
if type(Expression) != type(''):
self._Expr = Expression
self._NoProcess = True
return
self._Expr = ReplaceExprMacro(Expression.strip(),
SymbolTable,
SupportedInMacroList)
if not self._Expr.strip():
raise BadExpression(ERR_EMPTY_EXPR)
#
# The symbol table including PCD and macro mapping
#
self._Symb = SymbolTable
self._Symb.update(self.LogicalOperators)
self._Idx = 0
self._Len = len(self._Expr)
self._Token = ''
self._WarnExcept = None
# Literal token without any conversion
self._LiteralToken = ''
# Public entry for this class
# @param RealValue: False: only evaluate if the expression is true or false, used for conditional expression
# True : return the evaluated str(value), used for PCD value
#
# @return: True or False if RealValue is False
# Evaluated value of string format if RealValue is True
#
def __call__(self, RealValue=False):
if self._NoProcess:
return self._Expr
self._Expr = self._Expr.strip()
if RealValue:
self._Token = self._Expr
if self.__IsNumberToken():
return self._Expr
try:
Token = self._GetToken()
if type(Token) == type('') and Token.startswith('{') and Token.endswith('}') and self._Idx >= self._Len:
return self._Expr
except BadExpression:
pass
self._Idx = 0
self._Token = ''
Val = self._OrExpr()
RealVal = Val
if type(Val) == type(''):
if Val == 'L""':
Val = False
elif not Val:
Val = False
RealVal = '""'
elif not Val.startswith('L"') and not Val.startswith('{'):
Val = True
RealVal = '"' + RealVal + '"'
# The expression has been parsed, but the end of expression is not reached
# It means the rest does not comply EBNF of <Expression>
if self._Idx != self._Len:
raise BadExpression(ERR_SNYTAX % self._Expr[self._Idx:])
if RealValue:
RetVal = str(RealVal)
elif Val:
RetVal = True
else:
RetVal = False
if self._WarnExcept:
self._WarnExcept.result = RetVal
raise self._WarnExcept
else:
return RetVal
# Template function to parse binary operators which have same precedence
# Expr [Operator Expr]*
def _ExprFuncTemplate(self, EvalFunc, OpLst):
Val = EvalFunc()
while self._IsOperator(OpLst):
Op = self._Token
try:
Val = self.Eval(Op, Val, EvalFunc())
except WrnExpression, Warn:
self._WarnExcept = Warn
Val = Warn.result
return Val
# A [|| B]*
def _OrExpr(self):
return self._ExprFuncTemplate(self._AndExpr, ["OR", "or", "||"])
# A [&& B]*
def _AndExpr(self):
return self._ExprFuncTemplate(self._BitOr, ["AND", "and", "&&"])
# A [ | B]*
def _BitOr(self):
return self._ExprFuncTemplate(self._BitXor, ["|"])
# A [ ^ B]*
def _BitXor(self):
return self._ExprFuncTemplate(self._BitAnd, ["XOR", "xor", "^"])
# A [ & B]*
def _BitAnd(self):
return self._ExprFuncTemplate(self._EqExpr, ["&"])
# A [ == B]*
def _EqExpr(self):
Val = self._RelExpr()
while self._IsOperator(["==", "!=", "EQ", "NE", "IN", "in", "!", "NOT", "not"]):
Op = self._Token
if Op in ["!", "NOT", "not"]:
if not self._IsOperator(["IN", "in"]):
raise BadExpression(ERR_REL_NOT_IN)
Op += ' ' + self._Token
try:
Val = self.Eval(Op, Val, self._RelExpr())
except WrnExpression, Warn:
self._WarnExcept = Warn
Val = Warn.result
return Val
# A [ > B]*
def _RelExpr(self):
return self._ExprFuncTemplate(self._AddExpr, ["<=", ">=", "<", ">", "LE", "GE", "LT", "GT"])
# A [ + B]*
def _AddExpr(self):
return self._ExprFuncTemplate(self._UnaryExpr, ["+", "-"])
# [!]*A
def _UnaryExpr(self):
if self._IsOperator(["!", "NOT", "not"]):
Val = self._UnaryExpr()
try:
return self.Eval('not', Val)
except WrnExpression, Warn:
self._WarnExcept = Warn
return Warn.result
return self._IdenExpr()
# Parse identifier or encapsulated expression
def _IdenExpr(self):
Tk = self._GetToken()
if Tk == '(':
Val = self._OrExpr()
try:
# _GetToken may also raise BadExpression
if self._GetToken() != ')':
raise BadExpression(ERR_MATCH)
except BadExpression:
raise BadExpression(ERR_MATCH)
return Val
return Tk
# Skip whitespace or tab
def __SkipWS(self):
for Char in self._Expr[self._Idx:]:
if Char not in ' \t':
break
self._Idx += 1
# Try to convert string to number
def __IsNumberToken(self):
Radix = 10
if self._Token.lower()[0:2] == '0x' and len(self._Token) > 2:
Radix = 16
try:
self._Token = int(self._Token, Radix)
return True
except ValueError:
return False
except TypeError:
return False
# Parse array: {...}
def __GetArray(self):
Token = '{'
self._Idx += 1
self.__GetNList(True)
Token += self._LiteralToken
if self._Idx >= self._Len or self._Expr[self._Idx] != '}':
raise BadExpression(ERR_ARRAY_TOKEN % Token)
Token += '}'
# All whitespace and tabs in array are already stripped.
IsArray = IsGuid = False
if len(Token.split(',')) == 11 and len(Token.split(',{')) == 2 \
and len(Token.split('},')) == 1:
HexLen = [11,6,6,5,4,4,4,4,4,4,6]
HexList= Token.split(',')
if HexList[3].startswith('{') and \
not [Index for Index, Hex in enumerate(HexList) if len(Hex) > HexLen[Index]]:
IsGuid = True
if Token.lstrip('{').rstrip('}').find('{') == -1:
if not [Hex for Hex in Token.lstrip('{').rstrip('}').split(',') if len(Hex) > 4]:
IsArray = True
if not IsArray and not IsGuid:
raise BadExpression(ERR_ARRAY_TOKEN % Token)
self._Idx += 1
self._Token = self._LiteralToken = Token
return self._Token
# Parse string, the format must be: "..."
def __GetString(self):
Idx = self._Idx
# Skip left quote
self._Idx += 1
# Replace escape \\\", \"
Expr = self._Expr[self._Idx:].replace('\\\\', '//').replace('\\\"', '\\\'')
for Ch in Expr:
self._Idx += 1
if Ch == '"':
break
self._Token = self._LiteralToken = self._Expr[Idx:self._Idx]
if not self._Token.endswith('"'):
raise BadExpression(ERR_STRING_TOKEN % self._Token)
self._Token = self._Token[1:-1]
return self._Token
# Get token that is comprised by alphanumeric, underscore or dot(used by PCD)
# @param IsAlphaOp: Indicate if parsing general token or script operator(EQ, NE...)
def __GetIdToken(self, IsAlphaOp = False):
IdToken = ''
for Ch in self._Expr[self._Idx:]:
if not self.__IsIdChar(Ch):
break
self._Idx += 1
IdToken += Ch
self._Token = self._LiteralToken = IdToken
if not IsAlphaOp:
self.__ResolveToken()
return self._Token
# Try to resolve token
def __ResolveToken(self):
if not self._Token:
raise BadExpression(ERR_EMPTY_TOKEN)
# PCD token
if self.PcdPattern.match(self._Token):
if self._Token not in self._Symb:
Ex = BadExpression(ERR_PCD_RESOLVE % self._Token)
Ex.Pcd = self._Token
raise Ex
self._Token = ValueExpression(self._Symb[self._Token], self._Symb)(True)
if type(self._Token) != type(''):
self._LiteralToken = hex(self._Token)
return
if self._Token.startswith('"'):
self._Token = self._Token[1:-1]
elif self._Token in ["FALSE", "false", "False"]:
self._Token = False
elif self._Token in ["TRUE", "true", "True"]:
self._Token = True
else:
self.__IsNumberToken()
def __GetNList(self, InArray=False):
self._GetSingleToken()
if not self.__IsHexLiteral():
if InArray:
raise BadExpression(ERR_ARRAY_ELE % self._Token)
return self._Token
self.__SkipWS()
Expr = self._Expr[self._Idx:]
if not Expr.startswith(','):
return self._Token
NList = self._LiteralToken
while Expr.startswith(','):
NList += ','
self._Idx += 1
self.__SkipWS()
self._GetSingleToken()
if not self.__IsHexLiteral():
raise BadExpression(ERR_ARRAY_ELE % self._Token)
NList += self._LiteralToken
self.__SkipWS()
Expr = self._Expr[self._Idx:]
self._Token = self._LiteralToken = NList
return self._Token
def __IsHexLiteral(self):
if self._LiteralToken.startswith('{') and \
self._LiteralToken.endswith('}'):
return True
if self.HexPattern.match(self._LiteralToken):
Token = self._LiteralToken[2:]
Token = Token.lstrip('0')
if not Token:
self._LiteralToken = '0x0'
else:
self._LiteralToken = '0x' + Token.lower()
return True
return False
def _GetToken(self):
return self.__GetNList()
@staticmethod
def __IsIdChar(Ch):
return Ch in '._/:' or Ch.isalnum()
# Parse operand
def _GetSingleToken(self):
self.__SkipWS()
Expr = self._Expr[self._Idx:]
if Expr.startswith('L"'):
# Skip L
self._Idx += 1
UStr = self.__GetString()
self._Token = 'L"' + UStr + '"'
return self._Token
self._Token = ''
if Expr:
Ch = Expr[0]
Match = self.RegGuidPattern.match(Expr)
if Match and not Expr[Match.end():Match.end()+1].isalnum() \
and Expr[Match.end():Match.end()+1] != '_':
self._Idx += Match.end()
self._Token = ValueExpression(GuidStringToGuidStructureString(Expr[0:Match.end()]))(True)
return self._Token
elif self.__IsIdChar(Ch):
return self.__GetIdToken()
elif Ch == '"':
return self.__GetString()
elif Ch == '{':
return self.__GetArray()
elif Ch == '(' or Ch == ')':
self._Idx += 1
self._Token = Ch
return self._Token
raise BadExpression(ERR_VALID_TOKEN % Expr)
# Parse operator
def _GetOperator(self):
self.__SkipWS()
LegalOpLst = ['&&', '||', '!=', '==', '>=', '<='] + self.NonLetterOpLst
self._Token = ''
Expr = self._Expr[self._Idx:]
# Reach end of expression
if not Expr:
return ''
# Script operator: LT, GT, LE, GE, EQ, NE, and, or, xor, not
if Expr[0].isalpha():
return self.__GetIdToken(True)
# Start to get regular operator: +, -, <, > ...
if Expr[0] not in self.NonLetterOpLst:
return ''
OpToken = ''
for Ch in Expr:
if Ch in self.NonLetterOpLst:
if '!' == Ch and OpToken:
break
self._Idx += 1
OpToken += Ch
else:
break
if OpToken not in LegalOpLst:
raise BadExpression(ERR_OPERATOR_UNSUPPORT % OpToken)
self._Token = OpToken
return OpToken
# Check if current token matches the operators given from OpList
def _IsOperator(self, OpList):
Idx = self._Idx
self._GetOperator()
if self._Token in OpList:
if self._Token in self.LogicalOperators:
self._Token = self.LogicalOperators[self._Token]
return True
self._Idx = Idx
return False
if __name__ == '__main__':
pass
while True:
input = raw_input('Input expr: ')
if input in 'qQ':
break
try:
print ValueExpression(input)(True)
print ValueExpression(input)(False)
except WrnExpression, Ex:
print Ex.result
print str(Ex)
except Exception, Ex:
print str(Ex)