cppcheck/addons/misra.py

1225 lines
35 KiB
Python

#!/usr/bin/env python
#
# MISRA C 2012 checkers
#
# Example usage of this addon (scan a sourcefile main.cpp)
# cppcheck --dump main.cpp
# python misra.py --rule-texts=<path-to-rule-texts> main.cpp.dump
# python misra.py --misra-pdf=<path-to-misra.pdf> main.cpp.dump
#
# Limitations: This addon is released as open source. Rule texts can't be freely
# distributed. https://www.misra.org.uk/forum/viewtopic.php?f=56&t=1189
#
# Total number of rules: 153
import cppcheckdata
import sys
import re
import os
import tempfile
import subprocess
ruleTexts = {}
VERIFY = False
VERIFY_EXPECTED = []
VERIFY_ACTUAL = []
def reportError(location, num1, num2):
if VERIFY:
VERIFY_ACTUAL.append(str(location.linenr) + ':' + str(num1) + '.' + str(num2))
else:
num = num1 * 100 + num2
id = 'misra-c2012-' + str(num1) + '.' + str(num2)
if num in ruleTexts:
errmsg = ruleTexts[num] + ' [' + id + ']'
else:
errmsg = 'misra violation (use --misra-pdf=<file> or --rule-texts=<file> to get proper output) [' + id + ']'
sys.stderr.write('[' + location.file + ':' + str(location.linenr) + '] (style): ' + errmsg + '\n')
def simpleMatch(token, pattern):
for p in pattern.split(' '):
if not token or token.str != p:
return False
token = token.next
return True
KEYWORDS = {
'auto',
'break',
'case',
'char',
'const',
'continue',
'default',
'do',
'double',
'else',
'enum',
'extern',
'float',
'for',
'goto',
'if',
'int',
'long',
'register',
'return',
'short',
'signed',
'sizeof',
'static',
'struct',
'switch',
'typedef',
'union',
'unsigned',
'void',
'volatile',
'while'
}
def getEssentialType(expr):
if not expr:
return None
if expr.variable:
typeToken = expr.variable.typeStartToken
while typeToken and typeToken.isName:
if typeToken.str in {'char', 'short', 'int', 'long', 'float', 'double'}:
return typeToken.str
typeToken = typeToken.next
elif expr.astOperand1 and expr.astOperand2 and expr.str in {'+', '-', '*', '/', '%', '&', '|', '^'}:
e1 = getEssentialType(expr.astOperand1)
e2 = getEssentialType(expr.astOperand2)
if not e1 or not e2:
return None
types = ['bool', 'char', 'short', 'int', 'long', 'long long']
try:
i1 = types.index(e1)
i2 = types.index(e2)
if i2 >= i1:
return types[i2]
return types[i1]
except ValueError:
return None
return None
def bitsOfEssentialType(expr):
type = getEssentialType(expr)
if type is None:
return 0
if type == 'char':
return CHAR_BIT
if type == 'short':
return SHORT_BIT
if type == 'int':
return INT_BIT
if type == 'long':
return LONG_BIT
if type == 'long long':
return LONG_LONG_BIT
return 0
def isCast(expr):
if not expr or expr.str != '(' or not expr.astOperand1 or expr.astOperand2:
return False
if simpleMatch(expr, '( )'):
return False
return True
def isFunctionCall(expr):
if not expr:
return False
if expr.str != '(' or not expr.astOperand1:
return False
if expr.astOperand1 != expr.previous:
return False
if expr.astOperand1.str in KEYWORDS:
return False
return True
def countSideEffects(expr):
if not expr or expr.str in {',', ';'}:
return 0
ret = 0
if expr.str in {'++', '--', '='}:
ret = 1
return ret + countSideEffects(expr.astOperand1) + countSideEffects(expr.astOperand2)
def getForLoopExpressions(forToken):
if not forToken or forToken.str != 'for':
return None
lpar = forToken.next
if not lpar or lpar.str != '(':
return None
if not lpar.astOperand2 or lpar.astOperand2.str != ';':
return None
if not lpar.astOperand2.astOperand2 or lpar.astOperand2.astOperand2.str != ';':
return None
return [lpar.astOperand2.astOperand1,
lpar.astOperand2.astOperand2.astOperand1,
lpar.astOperand2.astOperand2.astOperand2]
def hasFloatComparison(expr):
if not expr:
return False
if expr.isLogicalOp:
return hasFloatComparison(expr.astOperand1) or hasFloatComparison(expr.astOperand2)
if expr.isComparisonOp:
# TODO: Use ValueType
return cppcheckdata.astIsFloat(expr.astOperand1) or cppcheckdata.astIsFloat(expr.astOperand2)
return False
def hasSideEffectsRecursive(expr):
if not expr:
return False
if expr.str == '=' and expr.astOperand1 and expr.astOperand1.str == '[':
prev = expr.astOperand1.previous
if prev and (prev.str == '{' or prev.str == '{'):
return hasSideEffectsRecursive(expr.astOperand2)
if expr.str in {'++', '--', '='}:
return True
# Todo: Check function calls
return hasSideEffectsRecursive(expr.astOperand1) or hasSideEffectsRecursive(expr.astOperand2)
def isBoolExpression(expr):
return expr and expr.str in {'!', '==', '!=', '<', '<=', '>', '>=', '&&', '||'}
def isConstantExpression(expr):
if expr.isNumber:
return True
if expr.isName:
return False
if simpleMatch(expr.previous, 'sizeof ('):
return True
if expr.astOperand1 and not isConstantExpression(expr.astOperand1):
return False
if expr.astOperand2 and not isConstantExpression(expr.astOperand2):
return False
return True
def isUnsignedInt(expr):
# TODO this function is very incomplete. use ValueType?
if not expr:
return False
if expr.isNumber:
return 'u' in expr.str or 'U' in expr.str
if expr.str in {'+', '-', '*', '/', '%'}:
return isUnsignedInt(expr.astOperand1) or isUnsignedInt(expr.astOperand2)
return False
def getPrecedence(expr):
if not expr:
return 16
if not expr.astOperand1 or not expr.astOperand2:
return 16
if expr.str in {'*', '/', '%'}:
return 12
if expr.str in {'+', '-'}:
return 11
if expr.str in {'<<', '>>'}:
return 10
if expr.str in {'<', '>', '<=', '>='}:
return 9
if expr.str in {'==', '!='}:
return 8
if expr.str == '&':
return 7
if expr.str == '^':
return 6
if expr.str == '|':
return 5
if expr.str == '&&':
return 4
if expr.str == '||':
return 3
if expr.str in {'?', ':'}:
return 2
if expr.isAssignmentOp:
return 1
if expr.str == ',':
return 0
return -1
def noParentheses(tok1, tok2):
while tok1 and tok1 != tok2:
if tok1.str == '(' or tok1.str == ')':
return False
tok1 = tok1.next
return tok1 == tok2
def findGotoLabel(gotoToken):
label = gotoToken.next.str
tok = gotoToken.next.next
while tok:
if tok.str == '}' and tok.scope.type == 'Function':
break
if tok.str == label and tok.next.str == ':':
return tok
tok = tok.next
return None
def findInclude(directives, header):
for directive in directives:
if directive.str == '#include ' + header:
return directive
return None
def misra_3_1(rawTokens):
for token in rawTokens:
if token.str.startswith('/*') or token.str.startswith('//'):
if '//' in token.str[2:] or '/*' in token.str[2:]:
reportError(token, 3, 1)
def misra_5_1(data):
for token in data.tokenlist:
if token.isName and len(token.str) > 31:
reportError(token, 5, 1)
def misra_5_3(data):
scopeVars = {}
for var in data.variables:
if var.isArgument:
# TODO
continue
if var.nameToken.scope not in scopeVars:
scopeVars[var.nameToken.scope] = []
scopeVars[var.nameToken.scope].append(var)
for innerScope in data.scopes:
if innerScope.type == 'Global':
continue
if innerScope not in scopeVars:
continue
for innerVar in scopeVars[innerScope]:
outerScope = innerScope.nestedIn
while outerScope:
if outerScope not in scopeVars:
outerScope = outerScope.nestedIn
continue
found = False
for outerVar in scopeVars[outerScope]:
if innerVar.nameToken.str == outerVar.nameToken.str:
found = True
break
if found:
reportError(innerVar.nameToken, 5, 3)
break
outerScope = outerScope.nestedIn
def misra_5_4(data):
for dir in data.directives:
if re.match(r'#define [a-zA-Z0-9_]{64,}', dir.str):
reportError(dir, 5, 4)
def misra_5_5(data):
macroNames = []
for dir in data.directives:
res = re.match(r'#define ([A-Za-z0-9_]+)', dir.str)
if res:
macroNames.append(res.group(1))
for var in data.variables:
if var.nameToken.str in macroNames:
reportError(var.nameToken, 5, 5)
def misra_7_1(rawTokens):
for tok in rawTokens:
if re.match(r'^0[0-7]+$', tok.str):
reportError(tok, 7, 1)
def misra_7_3(rawTokens):
for tok in rawTokens:
if re.match(r'^[0-9]+l', tok.str):
reportError(tok, 7, 3)
def misra_8_11(data):
for var in data.variables:
if var.isExtern and simpleMatch(var.nameToken.next, '[ ]') and var.nameToken.scope.type == 'Global':
reportError(var.nameToken, 8, 11)
def misra_8_12(data):
for token in data.tokenlist:
if token.str != '{':
continue
if not token.scope or token.scope.type != 'Enum':
continue
etok = token
values = []
while etok:
if etok.str == '}':
break
if etok.str == '=':
rhsValues = etok.astOperand2.values
if rhsValues and len(rhsValues) == 1:
if rhsValues[0].intvalue in values:
reportError(etok, 8, 12)
break
values.append(rhsValues[0].intvalue)
etok = etok.next
def misra_8_14(rawTokens):
for token in rawTokens:
if token.str == 'restrict':
reportError(token, 8, 14)
def misra_9_5(rawTokens):
for token in rawTokens:
if simpleMatch(token, '[ ] = { ['):
reportError(token, 9, 5)
def misra_10_4(data):
for token in data.tokenlist:
if token.str not in {'+', '-', '*', '/', '%', '&', '|', '^'} and not token.isComparisonOp:
continue
if not token.astOperand1 or not token.astOperand2:
continue
if not token.astOperand1.valueType or not token.astOperand2.valueType:
continue
if not token.astOperand1.valueType.isIntegral() or not token.astOperand2.valueType.isIntegral():
continue
e1 = getEssentialType(token.astOperand1)
e2 = getEssentialType(token.astOperand2)
if e1 and e2 and e1 != e2:
reportError(token, 10, 4)
def misra_10_6(data):
for token in data.tokenlist:
if token.str != '=' or not token.astOperand1 or not token.astOperand2:
continue
vt1 = token.astOperand1.valueType
vt2 = token.astOperand2.valueType
if not vt1 or vt1.pointer > 0:
continue
if not vt2 or vt2.pointer > 0:
continue
try:
intTypes = ['char', 'short', 'int', 'long', 'long long']
index1 = intTypes.index(vt1.type)
e = getEssentialType(token.astOperand2)
if not e:
continue
index2 = intTypes.index(e)
if index1 > index2:
reportError(token, 10, 6)
except ValueError:
pass
def misra_10_8(data):
for token in data.tokenlist:
if not isCast(token):
continue
if not token.valueType or token.valueType.pointer > 0:
continue
if not token.astOperand1.valueType or token.astOperand1.valueType.pointer > 0:
continue
if not token.astOperand1.astOperand1:
continue
try:
intTypes = ['char', 'short', 'int', 'long', 'long long']
index1 = intTypes.index(token.valueType.type)
e = getEssentialType(token.astOperand1)
if not e:
continue
index2 = intTypes.index(e)
if index1 > index2:
reportError(token, 10, 8)
except ValueError:
pass
def misra_11_3(data):
for token in data.tokenlist:
if not isCast(token):
continue
vt1 = token.valueType
vt2 = token.astOperand1.valueType
if not vt1 or not vt2:
continue
if vt1.pointer == vt2.pointer and vt1.pointer > 0 and vt1.type != vt2.type and\
vt1.isIntegral() and vt2.isIntegral() and vt1.type != 'char':
reportError(token, 11, 3)
def misra_11_4(data):
for token in data.tokenlist:
if not isCast(token):
continue
vt1 = token.valueType
vt2 = token.astOperand1.valueType
if not vt1 or not vt2:
continue
if vt1.pointer == 0 and vt2.pointer > 0 and vt2.type != 'void':
reportError(token, 11, 4)
def misra_11_5(data):
for token in data.tokenlist:
if not isCast(token):
continue
vt1 = token.valueType
vt2 = token.astOperand1.valueType
if not vt1 or not vt2:
continue
if vt1.pointer > 0 and vt1.type != 'void' and vt2.pointer == vt1.pointer and vt2.type == 'void':
reportError(token, 11, 5)
def misra_11_6(data):
for token in data.tokenlist:
if not isCast(token):
continue
vt1 = token.valueType
vt2 = token.astOperand1.valueType
if not vt1 or not vt2:
continue
if vt1.pointer == 1 and vt1.type == 'void' and vt2.pointer == 0:
reportError(token, 11, 6)
elif vt1.pointer == 0 and vt2.pointer == 1 and vt2.type == 'void':
reportError(token, 11, 6)
def misra_11_7(data):
for token in data.tokenlist:
if not isCast(token):
continue
vt1 = token.valueType
vt2 = token.astOperand1.valueType
if not vt1 or not vt2:
continue
if vt1.pointer > 0 and vt1.type == 'record' and\
vt2.pointer > 0 and vt2.type == 'record' and vt1.typeScopeId != vt2.typeScopeId:
reportError(token, 11, 7)
def misra_11_8(data):
for token in data.tokenlist:
if not isCast(token):
continue
if not token.valueType or not token.astOperand1.valueType:
continue
if token.valueType.pointer == 0 or token.valueType.pointer == 0:
continue
if token.valueType.constness == 0 and token.astOperand1.valueType.constness > 0:
reportError(token, 11, 8)
def misra_11_9(data):
for directive in data.directives:
res1 = re.match(r'#define ([A-Za-z_][A-Za-z_0-9]*) (.*)', directive.str)
if not res1:
continue
name = res1.group(1)
if name == 'NULL':
continue
value = res1.group(2).replace(' ', '')
if value == '((void*)0)':
reportError(directive, 11, 9)
def misra_12_1_sizeof(rawTokens):
state = 0
for tok in rawTokens:
if tok.str.startswith('//') or tok.str.startswith('/*'):
continue
if tok.str == 'sizeof':
state = 1
elif state == 1:
if re.match(r'^[a-zA-Z_]', tok.str):
state = 2
else:
state = 0
elif state == 2:
if tok.str in {'+', '-', '*', '/', '%'}:
reportError(tok, 12, 1)
else:
state = 0
def misra_12_1(data):
for token in data.tokenlist:
p = getPrecedence(token)
if p < 2 or p > 12:
continue
p1 = getPrecedence(token.astOperand1)
if p < p1 <= 12 and noParentheses(token.astOperand1, token):
reportError(token, 12, 1)
continue
p2 = getPrecedence(token.astOperand2)
if p < p2 <= 12 and noParentheses(token, token.astOperand2):
reportError(token, 12, 1)
continue
def misra_12_2(data):
for token in data.tokenlist:
if not (token.str in {'<<', '>>'}):
continue
if (not token.astOperand2) or (not token.astOperand2.values):
continue
maxval = 0
for val in token.astOperand2.values:
if val.intvalue and val.intvalue > maxval:
maxval = val.intvalue
if maxval == 0:
continue
sz = bitsOfEssentialType(token.astOperand1)
if sz <= 0:
continue
if maxval >= sz:
reportError(token, 12, 2)
def misra_12_3(data):
for token in data.tokenlist:
if token.str != ',' or token.scope.type == 'Enum':
continue
if token.astParent and token.astParent.str in {'(', ',', '{'}:
continue
reportError(token, 12, 3)
def misra_12_4(data):
if INT_BIT == 16:
max_uint = 0xffff
elif INT_BIT == 32:
max_uint = 0xffffffff
else:
return
for token in data.tokenlist:
if (not isConstantExpression(token)) or (not isUnsignedInt(token)):
continue
if not token.values:
continue
for value in token.values:
if value.intvalue < 0 or value.intvalue > max_uint:
reportError(token, 12, 4)
break
def misra_13_1(data):
for token in data.tokenlist:
if token.str != '=':
continue
init = token.next
if init and init.str == '{' and hasSideEffectsRecursive(init):
reportError(init, 13, 1)
def misra_13_3(data):
for token in data.tokenlist:
if token.str not in {'++', '--'}:
continue
astTop = token
while astTop.astParent and astTop.astParent.str not in {',', ';'}:
astTop = astTop.astParent
if countSideEffects(astTop) >= 2:
reportError(astTop, 13, 3)
def misra_13_4(data):
for token in data.tokenlist:
if token.str != '=':
continue
if not token.astParent:
continue
if token.astOperand1.str == '[' and token.astOperand1.previous.str in {'{', ','}:
continue
if not (token.astParent.str in [',', ';']):
reportError(token, 13, 4)
def misra_13_5(data):
for token in data.tokenlist:
if token.isLogicalOp and hasSideEffectsRecursive(token.astOperand2):
reportError(token, 13, 5)
def misra_13_6(data):
for token in data.tokenlist:
if token.str == 'sizeof' and hasSideEffectsRecursive(token.next):
reportError(token, 13, 6)
def misra_14_1(data):
for token in data.tokenlist:
if token.str != 'for':
continue
exprs = getForLoopExpressions(token)
if exprs and hasFloatComparison(exprs[1]):
reportError(token, 14, 1)
def misra_14_2(data):
for token in data.tokenlist:
expressions = getForLoopExpressions(token)
if not expressions:
continue
if expressions[0] and not expressions[0].isAssignmentOp:
reportError(token, 14, 2)
elif hasSideEffectsRecursive(expressions[1]):
reportError(token, 14, 2)
def misra_14_4(data):
for token in data.tokenlist:
if token.str != '(':
continue
if not token.astOperand1 or not (token.astOperand1.str in {'if', 'while'}):
continue
if not isBoolExpression(token.astOperand2):
reportError(token, 14, 4)
def misra_15_1(data):
for token in data.tokenlist:
if token.str == "goto":
reportError(token, 15, 1)
def misra_15_2(data):
for token in data.tokenlist:
if token.str != 'goto':
continue
if (not token.next) or (not token.next.isName):
continue
if not findGotoLabel(token):
reportError(token, 15, 2)
def misra_15_3(data):
for token in data.tokenlist:
if token.str != 'goto':
continue
if (not token.next) or (not token.next.isName):
continue
tok = findGotoLabel(token)
if not tok:
continue
scope = token.scope
while scope and scope != tok.scope:
scope = scope.nestedIn
if not scope:
reportError(token, 15, 3)
def misra_15_5(data):
for token in data.tokenlist:
if token.str == 'return' and token.scope.type != 'Function':
reportError(token, 15, 5)
def misra_15_6(rawTokens):
state = 0
indent = 0
tok1 = None
for token in rawTokens:
if token.str in {'if', 'for', 'while'}:
if simpleMatch(token.previous, '# if'):
continue
if simpleMatch(token.previous, "} while"):
continue
state = 1
indent = 0
tok1 = token
elif state == 1:
if indent == 0 and token.str != '(':
state = 0
continue
if token.str == '(':
indent = indent + 1
elif token.str == ')':
if indent == 0:
state = 0
elif indent == 1:
state = 2
indent = indent - 1
elif state == 2:
if token.str.startswith('//') or token.str.startswith('/*'):
continue
state = 0
if token.str != '{':
reportError(tok1, 15, 6)
def misra_15_7(data):
for token in data.tokenlist:
if not simpleMatch(token, '}'):
continue
if not token.scope.type == 'If':
continue
if not token.scope.nestedIn.type == 'Else':
continue
if not token.next.str == 'else':
reportError(token, 15, 7)
# TODO add 16.1 rule
def misra_16_2(data):
for token in data.tokenlist:
if token.str == 'case' and token.scope.type != 'Switch':
reportError(token, 16, 2)
def misra_16_3(rawTokens):
# state: 0=no, 1=break is seen but not its ';', 2=after 'break;', 'comment', '{'
state = 0
for token in rawTokens:
if token.str == 'break':
state = 1
elif token.str == ';':
if state == 1:
state = 2
else:
state = 0
elif token.str.startswith('/*') or token.str.startswith('//'):
if 'fallthrough' in token.str.lower():
state = 2
elif token.str == '{':
state = 2
elif token.str == 'case' and state != 2:
reportError(token, 16, 3)
def misra_16_4(data):
for token in data.tokenlist:
if token.str != 'switch':
continue
if not simpleMatch(token, 'switch ('):
continue
if not simpleMatch(token.next.link, ') {'):
continue
startTok = token.next.link.next
tok = startTok.next
while tok and tok.str != '}':
if tok.str == '{':
tok = tok.link
elif tok.str == 'default':
break
tok = tok.next
if tok and tok.str != 'default':
reportError(token, 16, 4)
def misra_16_5(data):
for token in data.tokenlist:
if token.str != 'default':
continue
if token.previous and token.previous.str == '{':
continue
tok2 = token
while tok2:
if tok2.str in {'}', 'case'}:
break
if tok2.str == '{':
tok2 = tok2.link
tok2 = tok2.next
if tok2 and tok2.str == 'case':
reportError(token, 16, 5)
def misra_16_6(data):
for token in data.tokenlist:
if not (simpleMatch(token, 'switch (') and simpleMatch(token.next.link, ') {')):
continue
tok = token.next.link.next.next
count = 0
while tok:
if tok.str == 'break':
count = count + 1
elif tok.str == '{':
tok = tok.link
if simpleMatch(tok.previous.previous, 'break ;'):
count = count + 1
elif tok.str == '}':
break
tok = tok.next
if count < 2:
reportError(token, 16, 6)
def misra_16_7(data):
for token in data.tokenlist:
if simpleMatch(token, 'switch (') and isBoolExpression(token.next.astOperand2):
reportError(token, 16, 7)
def misra_17_1(data):
for token in data.tokenlist:
if isFunctionCall(token) and token.astOperand1.str in {'va_list', 'va_arg', 'va_start', 'va_end', 'va_copy'}:
reportError(token, 17, 1)
def misra_17_6(rawTokens):
for token in rawTokens:
if simpleMatch(token, '[ static'):
reportError(token, 17, 6)
def misra_17_8(data):
for token in data.tokenlist:
if not (token.isAssignmentOp or (token.str in {'++', '--'})):
continue
if not token.astOperand1:
continue
var = token.astOperand1.variable
if var and var.isArgument:
reportError(token, 17, 8)
def misra_18_5(data):
for var in data.variables:
if not var.isPointer:
continue
typetok = var.nameToken
count = 0
while typetok:
if typetok.str == '*':
count = count + 1
elif not typetok.isName:
break
typetok = typetok.previous
if count > 2:
reportError(var.nameToken, 18, 5)
def misra_18_8(data):
for var in data.variables:
if not var.isArray or not var.isLocal:
continue
# TODO Array dimensions are not available in dump, must look in tokens
typetok = var.nameToken.next
if not typetok or typetok.str != '[':
continue
if not isConstantExpression(typetok.astOperand2):
reportError(var.nameToken, 18, 8)
def misra_19_2(data):
for token in data.tokenlist:
if token.str == 'union':
reportError(token, 19, 2)
def misra_20_1(data):
for directive in data.directives:
if not directive.str.startswith('#include'):
continue
for token in data.tokenlist:
if token.file != directive.file:
continue
if int(token.linenr) < int(directive.linenr):
reportError(directive, 20, 1)
break
def misra_20_2(data):
for directive in data.directives:
if not directive.str.startswith('#include '):
continue
for pattern in {'\\', '//', '/*', "'"}:
if pattern in directive.str:
reportError(directive, 20, 2)
break
def misra_20_3(rawTokens):
linenr = -1
for token in rawTokens:
if token.str.startswith('/') or token.linenr == linenr:
continue
linenr = token.linenr
if not simpleMatch(token, '# include'):
continue
headerToken = token.next.next
if not headerToken or not (headerToken.str.startswith('<') or headerToken.str.startswith('"')):
reportError(token, 20, 3)
def misra_20_4(data):
for directive in data.directives:
res = re.search(r'#define ([a-z][a-z0-9_]+)', directive.str)
if res and (res.group(1) in KEYWORDS):
reportError(directive, 20, 4)
def misra_20_5(data):
for directive in data.directives:
if directive.str.startswith('#undef '):
reportError(directive, 20, 5)
def misra_21_3(data):
for token in data.tokenlist:
if isFunctionCall(token) and (token.astOperand1.str in {'malloc', 'calloc', 'realloc', 'free'}):
reportError(token, 21, 3)
def misra_21_4(data):
directive = findInclude(data.directives, '<setjmp.h>')
if directive:
reportError(directive, 21, 4)
def misra_21_5(data):
directive = findInclude(data.directives, '<signal.h>')
if directive:
reportError(directive, 21, 5)
def misra_21_7(data):
for token in data.tokenlist:
if isFunctionCall(token) and (token.astOperand1.str in {'atof', 'atoi', 'atol', 'atoll'}):
reportError(token, 21, 7)
def misra_21_8(data):
for token in data.tokenlist:
if isFunctionCall(token) and (token.astOperand1.str in {'abort', 'getenv', 'system'}):
reportError(token, 21, 8)
def misra_21_9(data):
for token in data.tokenlist:
if (token.str in {'bsearch', 'qsort'}) and token.next and token.next.str == '(':
reportError(token, 21, 9)
def misra_21_11(data):
directive = findInclude(data.directives, '<tgmath.h>')
if directive:
reportError(directive, 21, 11)
def loadRuleTexts(filename):
num1 = 0
num2 = 0
appendixA = False
for line in open(filename, 'rt'):
line = line.replace('\r', '').replace('\n', '')
if not appendixA:
if line.find('Appendix A') >= 0 and line.find('Summary of guidelines') >= 10:
appendixA = True
continue
if line.find('Appendix B') >= 0:
break
res = re.match(r'^Rule ([0-9]+).([0-9]+)', line)
if res:
num1 = int(res.group(1))
num2 = int(res.group(2))
continue
res = re.match(r'^[ ]*(Advisory|Required|Mandatory)$', line)
if res:
continue
res = re.match(r'^[ ]*([#A-Z].*)', line)
if res:
global ruleTexts
ruleTexts[num1*100+num2] = res.group(1)
num2 = num2 + 1
continue
def loadRuleTextsFromPdf(filename):
f = tempfile.NamedTemporaryFile(delete=False)
f.close()
#print('tempfile:' + f.name)
subprocess.call(['pdftotext', filename, f.name])
loadRuleTexts(f.name)
try:
os.remove(f.name)
except OSError as err:
print('Failed to remove temporary file ' + f.name)
if len(sys.argv) == 1:
print("""
Syntax: misra.py [OPTIONS] <dumpfiles>
OPTIONS:
--rule-texts=<file> Load rule texts from plain text file.
If you have the tool 'pdftotext' you can generate
this textfile with such command:
$ pdftotext MISRA_C_2012.pdf MISRA_C_2012.txt
Otherwise you can more or less copy/paste the chapter
Appendix A Summary of guidelines
from the MISRA pdf.
Format:
<..arbitrary text..>
Appendix A Summary of guidelines
Dir 1.1
Rule text for 1.1
Dir 1.2
Rule text for 1.2
<...>
--misra-pdf=<file> Misra PDF file that rule texts will be extracted from.
The tool 'pdftotext' from xpdf is used and must be installed.
Debian: sudo apt-get install xpdf
Windows: http://gnuwin32.sourceforge.net/packages/xpdf.htm
If you don't have 'pdftotext' and don't want to install it then
you can use --rule-texts=<file>.
""")
sys.exit(1)
for arg in sys.argv[1:]:
if arg == '-verify':
VERIFY = True
elif arg.startswith('--rule-texts='):
loadRuleTexts(arg[13:])
elif arg.startswith('--misra-pdf='):
loadRuleTextsFromPdf(arg[12:])
for arg in sys.argv[1:]:
if not arg.endswith('.dump'):
continue
data = cppcheckdata.parsedump(arg)
CHAR_BIT = data.platform.char_bit
SHORT_BIT = data.platform.short_bit
INT_BIT = data.platform.int_bit
LONG_BIT = data.platform.long_bit
LONG_LONG_BIT = data.platform.long_long_bit
POINTER_BIT = data.platform.pointer_bit
if VERIFY:
VERIFY_ACTUAL = []
VERIFY_EXPECTED = []
for tok in data.rawTokens:
if tok.str.startswith('//') and 'TODO' not in tok.str:
for word in tok.str[2:].split(' '):
if re.match(r'[0-9]+\.[0-9]+', word):
VERIFY_EXPECTED.append(str(tok.linenr) + ':' + word)
else:
print('Checking ' + arg + '...')
cfgNumber = 0
for cfg in data.configurations:
cfgNumber = cfgNumber + 1
if len(data.configurations) > 1:
print('Checking ' + arg + ', config "' + cfg.name + '"...')
if cfgNumber == 1:
misra_3_1(data.rawTokens)
misra_5_1(cfg)
misra_5_3(cfg)
misra_5_4(cfg)
misra_5_5(cfg)
if cfgNumber == 1:
misra_7_1(data.rawTokens)
misra_7_3(data.rawTokens)
misra_8_11(cfg)
misra_8_12(cfg)
if cfgNumber == 1:
misra_8_14(data.rawTokens)
misra_9_5(data.rawTokens)
misra_10_4(cfg)
misra_10_6(cfg)
misra_10_8(cfg)
misra_11_3(cfg)
misra_11_4(cfg)
misra_11_5(cfg)
misra_11_6(cfg)
misra_11_7(cfg)
misra_11_8(cfg)
misra_11_9(cfg)
if cfgNumber == 1:
misra_12_1_sizeof(data.rawTokens)
misra_12_1(cfg)
misra_12_2(cfg)
misra_12_3(cfg)
misra_12_4(cfg)
misra_13_1(cfg)
misra_13_3(cfg)
misra_13_4(cfg)
misra_13_5(cfg)
misra_13_6(cfg)
misra_14_1(cfg)
misra_14_2(cfg)
misra_14_4(cfg)
misra_15_1(cfg)
misra_15_2(cfg)
misra_15_3(cfg)
misra_15_5(cfg)
if cfgNumber == 1:
misra_15_6(data.rawTokens)
misra_15_7(cfg)
misra_16_2(cfg)
if cfgNumber == 1:
misra_16_3(data.rawTokens)
misra_16_4(cfg)
misra_16_5(cfg)
misra_16_6(cfg)
misra_16_7(cfg)
misra_17_1(cfg)
if cfgNumber == 1:
misra_17_6(data.rawTokens)
misra_17_8(cfg)
misra_18_5(cfg)
misra_18_8(cfg)
misra_19_2(cfg)
misra_20_1(cfg)
misra_20_2(cfg)
if cfgNumber == 1:
misra_20_3(data.rawTokens)
misra_20_4(cfg)
misra_20_5(cfg)
misra_21_3(cfg)
misra_21_4(cfg)
misra_21_5(cfg)
misra_21_7(cfg)
misra_21_8(cfg)
misra_21_9(cfg)
misra_21_11(cfg)
if VERIFY:
for expected in VERIFY_EXPECTED:
if expected not in VERIFY_ACTUAL:
print('Expected but not seen: ' + expected)
sys.exit(1)
for actual in VERIFY_ACTUAL:
if actual not in VERIFY_EXPECTED:
print('Not expected: ' + actual)
sys.exit(1)