Thu, 10 Jan 2019 14:18:48 +0100
Updated copyright for 2019.
# -*- coding: utf-8 -*- # Copyright (c) 2015 - 2019 Detlev Offenbach <detlev@die-offenbachs.de> # """ Module implementing a checker for miscellaneous checks. """ import sys import ast import re import itertools from string import Formatter def composeCallPath(node): """ Generator function to assemble the call path of a given node. @param node node to assemble call path for @type ast.Node @return call path components @rtype str """ if isinstance(node, ast.Attribute): for v in composeCallPath(node.value): yield v yield node.attr elif isinstance(node, ast.Name): yield node.id class MiscellaneousChecker(object): """ Class implementing a checker for miscellaneous checks. """ Codes = [ "M101", "M102", "M111", "M112", "M131", "M132", "M191", "M192", "M193", "M194", "M195", "M196", "M197", "M198", "M201", "M501", "M502", "M503", "M504", "M505", "M506", "M507", "M511", "M512", "M513", "M514", "M601", "M611", "M612", "M613", "M621", "M622", "M623", "M624", "M625", "M631", "M632", "M651", "M652", "M653", "M654", "M655", "M701", "M702", "M711", "M801", "M811", "M821", "M822", "M901", ] Formatter = Formatter() FormatFieldRegex = re.compile(r'^((?:\s|.)*?)(\..*|\[.*\])?$') BuiltinsWhiteList = [ "__name__", "__doc__", "credits", ] def __init__(self, source, filename, select, ignore, expected, repeat, args): """ Constructor @param source source code to be checked @type list of str @param filename name of the source file @type str @param select list of selected codes @type list of str @param ignore list of codes to be ignored @type list of str @param expected list of expected codes @type list of str @param repeat flag indicating to report each occurrence of a code @type bool @param args dictionary of arguments for the miscellaneous checks @type dict """ self.__select = tuple(select) self.__ignore = ('',) if select else tuple(ignore) self.__expected = expected[:] self.__repeat = repeat self.__filename = filename self.__source = source[:] self.__args = args self.__pep3101FormatRegex = re.compile( r'^(?:[^\'"]*[\'"][^\'"]*[\'"])*\s*%|^\s*%') if sys.version_info >= (3, 0): import builtins self.__builtins = [b for b in dir(builtins) if b not in self.BuiltinsWhiteList] else: import __builtin__ self.__builtins = [b for b in dir(__builtin__) if b not in self.BuiltinsWhiteList] # statistics counters self.counters = {} # collection of detected errors self.errors = [] checkersWithCodes = [ (self.__checkCoding, ("M101", "M102")), (self.__checkCopyright, ("M111", "M112")), (self.__checkBuiltins, ("M131", "M132")), (self.__checkComprehensions, ("M191", "M192", "M193", "M194", "M195", "M196", "M197", "M198")), (self.__checkDictWithSortedKeys, ("M201",)), (self.__checkPep3101, ("M601",)), (self.__checkFormatString, ("M611", "M612", "M613", "M621", "M622", "M623", "M624", "M625", "M631", "M632")), (self.__checkBugBear, ("M501", "M502", "M503", "M504", "M505", "M506", "M507", "M511", "M512", "M513", "M514")), (self.__checkLogging, ("M651", "M652", "M653", "M654", "M655")), (self.__checkFuture, ("M701", "M702")), (self.__checkGettext, ("M711",)), (self.__checkPrintStatements, ("M801",)), (self.__checkTuple, ("M811", )), (self.__checkMutableDefault, ("M821", "M822")), ] self.__defaultArgs = { "BuiltinsChecker": { "chr": ["unichr", ], "str": ["unicode", ], }, "CodingChecker": 'latin-1, utf-8', "CopyrightChecker": { "Author": "", "MinFilesize": 0, }, } self.__checkers = [] for checker, codes in checkersWithCodes: if any(not (code and self.__ignoreCode(code)) for code in codes): self.__checkers.append(checker) def __ignoreCode(self, code): """ Private method to check if the message code should be ignored. @param code message code to check for @type str @return flag indicating to ignore the given code @rtype bool """ return (code.startswith(self.__ignore) and not code.startswith(self.__select)) def __error(self, lineNumber, offset, code, *args): """ Private method to record an issue. @param lineNumber line number of the issue @type int @param offset position within line of the issue @type int @param code message code @type str @param args arguments for the message @type list """ if self.__ignoreCode(code): return if code in self.counters: self.counters[code] += 1 else: self.counters[code] = 1 # Don't care about expected codes if code in self.__expected: return if code and (self.counters[code] == 1 or self.__repeat): # record the issue with one based line number self.errors.append( (self.__filename, lineNumber + 1, offset, (code, args))) def __reportInvalidSyntax(self): """ Private method to report a syntax error. """ exc_type, exc = sys.exc_info()[:2] if len(exc.args) > 1: offset = exc.args[1] if len(offset) > 2: offset = offset[1:3] else: offset = (1, 0) self.__error(offset[0] - 1, offset[1] or 0, 'M901', exc_type.__name__, exc.args[0]) def run(self): """ Public method to check the given source against miscellaneous conditions. """ if not self.__filename: # don't do anything, if essential data is missing return if not self.__checkers: # don't do anything, if no codes were selected return source = "".join(self.__source) # Check type for py2: if not str it's unicode if sys.version_info[0] == 2: try: source = source.encode('utf-8') except UnicodeError: pass try: self.__tree = compile(source, self.__filename, 'exec', ast.PyCF_ONLY_AST) except (SyntaxError, TypeError): self.__reportInvalidSyntax() return for check in self.__checkers: check() def __getCoding(self): """ Private method to get the defined coding of the source. @return tuple containing the line number and the coding @rtype tuple of int and str """ for lineno, line in enumerate(self.__source[:5]): matched = re.search(r'coding[:=]\s*([-\w_.]+)', line, re.IGNORECASE) if matched: return lineno, matched.group(1) else: return 0, "" def __checkCoding(self): """ Private method to check the presence of a coding line and valid encodings. """ if len(self.__source) == 0: return encodings = [e.lower().strip() for e in self.__args.get( "CodingChecker", self.__defaultArgs["CodingChecker"]) .split(",")] lineno, coding = self.__getCoding() if coding: if coding.lower() not in encodings: self.__error(lineno, 0, "M102", coding) else: self.__error(0, 0, "M101") def __checkCopyright(self): """ Private method to check the presence of a copyright statement. """ source = "".join(self.__source) copyrightArgs = self.__args.get( "CopyrightChecker", self.__defaultArgs["CopyrightChecker"]) copyrightMinFileSize = copyrightArgs.get( "MinFilesize", self.__defaultArgs["CopyrightChecker"]["MinFilesize"]) copyrightAuthor = copyrightArgs.get( "Author", self.__defaultArgs["CopyrightChecker"]["Author"]) copyrightRegexStr = \ r"Copyright\s+(\(C\)\s+)?(\d{{4}}\s+-\s+)?\d{{4}}\s+{author}" tocheck = max(1024, copyrightMinFileSize) topOfSource = source[:tocheck] if len(topOfSource) < copyrightMinFileSize: return copyrightRe = re.compile(copyrightRegexStr.format(author=r".*"), re.IGNORECASE) if not copyrightRe.search(topOfSource): self.__error(0, 0, "M111") return if copyrightAuthor: copyrightAuthorRe = re.compile( copyrightRegexStr.format(author=copyrightAuthor), re.IGNORECASE) if not copyrightAuthorRe.search(topOfSource): self.__error(0, 0, "M112") def __checkPrintStatements(self): """ Private method to check for print statements. """ for node in ast.walk(self.__tree): if (isinstance(node, ast.Call) and getattr(node.func, 'id', None) == 'print') or \ (hasattr(ast, 'Print') and isinstance(node, ast.Print)): self.__error(node.lineno - 1, node.col_offset, "M801") def __checkTuple(self): """ Private method to check for one element tuples. """ for node in ast.walk(self.__tree): if isinstance(node, ast.Tuple) and \ len(node.elts) == 1: self.__error(node.lineno - 1, node.col_offset, "M811") def __checkFuture(self): """ Private method to check the __future__ imports. """ expectedImports = { i.strip() for i in self.__args.get("FutureChecker", "").split(",") if bool(i.strip())} if len(expectedImports) == 0: # nothing to check for; disabling the check return imports = set() node = None hasCode = False for node in ast.walk(self.__tree): if (isinstance(node, ast.ImportFrom) and node.module == '__future__'): imports |= {name.name for name in node.names} elif isinstance(node, ast.Expr): if not isinstance(node.value, ast.Str): hasCode = True break elif not isinstance(node, (ast.Module, ast.Str)): hasCode = True break if isinstance(node, ast.Module) or not hasCode: return if not (imports >= expectedImports): if imports: self.__error(node.lineno - 1, node.col_offset, "M701", ", ".join(expectedImports), ", ".join(imports)) else: self.__error(node.lineno - 1, node.col_offset, "M702", ", ".join(expectedImports)) def __checkPep3101(self): """ Private method to check for old style string formatting. """ for lineno, line in enumerate(self.__source): match = self.__pep3101FormatRegex.search(line) if match: lineLen = len(line) pos = line.find('%') formatPos = pos formatter = '%' if line[pos + 1] == "(": pos = line.find(")", pos) c = line[pos] while c not in "diouxXeEfFgGcrs": pos += 1 if pos >= lineLen: break c = line[pos] if c in "diouxXeEfFgGcrs": formatter += c self.__error(lineno, formatPos, "M601", formatter) def __checkFormatString(self): """ Private method to check string format strings. """ coding = self.__getCoding()[1] if not coding: # default to utf-8 coding = "utf-8" visitor = TextVisitor() visitor.visit(self.__tree) for node in visitor.nodes: text = node.s if sys.version_info[0] > 2 and isinstance(text, bytes): try: text = text.decode(coding) except UnicodeDecodeError: continue fields, implicit, explicit = self.__getFields(text) if implicit: if node in visitor.calls: self.__error(node.lineno - 1, node.col_offset, "M611") else: if node.is_docstring: self.__error(node.lineno - 1, node.col_offset, "M612") else: self.__error(node.lineno - 1, node.col_offset, "M613") if node in visitor.calls: call, strArgs = visitor.calls[node] numbers = set() names = set() # Determine which fields require a keyword and which an arg for name in fields: fieldMatch = self.FormatFieldRegex.match(name) try: number = int(fieldMatch.group(1)) except ValueError: number = -1 # negative numbers are considered keywords if number >= 0: numbers.add(number) else: names.add(fieldMatch.group(1)) keywords = {keyword.arg for keyword in call.keywords} numArgs = len(call.args) if strArgs: numArgs -= 1 if sys.version_info < (3, 5): hasKwArgs = bool(call.kwargs) hasStarArgs = bool(call.starargs) else: hasKwArgs = any(kw.arg is None for kw in call.keywords) hasStarArgs = sum(1 for arg in call.args if isinstance(arg, ast.Starred)) if hasKwArgs: keywords.discard(None) if hasStarArgs: numArgs -= 1 # if starargs or kwargs is not None, it can't count the # parameters but at least check if the args are used if hasKwArgs: if not names: # No names but kwargs self.__error(call.lineno - 1, call.col_offset, "M623") if hasStarArgs: if not numbers: # No numbers but args self.__error(call.lineno - 1, call.col_offset, "M624") if not hasKwArgs and not hasStarArgs: # can actually verify numbers and names for number in sorted(numbers): if number >= numArgs: self.__error(call.lineno - 1, call.col_offset, "M621", number) for name in sorted(names): if name not in keywords: self.__error(call.lineno - 1, call.col_offset, "M622", name) for arg in range(numArgs): if arg not in numbers: self.__error(call.lineno - 1, call.col_offset, "M631", arg) for keyword in keywords: if keyword not in names: self.__error(call.lineno - 1, call.col_offset, "M632", keyword) if implicit and explicit: self.__error(call.lineno - 1, call.col_offset, "M625") def __getFields(self, string): """ Private method to extract the format field information. @param string format string to be parsed @type str @return format field information as a tuple with fields, implicit field definitions present and explicit field definitions present @rtype tuple of set of str, bool, bool """ fields = set() cnt = itertools.count() implicit = False explicit = False try: for _literal, field, spec, conv in self.Formatter.parse(string): if field is not None and (conv is None or conv in 'rsa'): if not field: field = str(next(cnt)) implicit = True else: explicit = True fields.add(field) fields.update(parsedSpec[1] for parsedSpec in self.Formatter.parse(spec) if parsedSpec[1] is not None) except ValueError: return set(), False, False else: return fields, implicit, explicit def __checkBuiltins(self): """ Private method to check, if built-ins are shadowed. """ functionDefs = [ast.FunctionDef] try: functionDefs.append(ast.AsyncFunctionDef) except AttributeError: pass ignoreBuiltinAssignments = self.__args.get( "BuiltinsChecker", self.__defaultArgs["BuiltinsChecker"]) for node in ast.walk(self.__tree): if isinstance(node, ast.Assign): # assign statement for element in node.targets: if isinstance(element, ast.Name) and \ element.id in self.__builtins: value = node.value if isinstance(value, ast.Name) and \ element.id in ignoreBuiltinAssignments and \ value.id in ignoreBuiltinAssignments[element.id]: # ignore compatibility assignments continue self.__error(element.lineno - 1, element.col_offset, "M131", element.id) elif isinstance(element, (ast.Tuple, ast.List)): for tupleElement in element.elts: if isinstance(tupleElement, ast.Name) and \ tupleElement.id in self.__builtins: self.__error(tupleElement.lineno - 1, tupleElement.col_offset, "M131", tupleElement.id) elif isinstance(node, ast.For): # for loop target = node.target if isinstance(target, ast.Name) and \ target.id in self.__builtins: self.__error(target.lineno - 1, target.col_offset, "M131", target.id) elif isinstance(target, (ast.Tuple, ast.List)): for element in target.elts: if isinstance(element, ast.Name) and \ element.id in self.__builtins: self.__error(element.lineno - 1, element.col_offset, "M131", element.id) elif any(isinstance(node, functionDef) for functionDef in functionDefs): # (asynchronous) function definition if sys.version_info >= (3, 0): for arg in node.args.args: if isinstance(arg, ast.arg) and \ arg.arg in self.__builtins: self.__error(arg.lineno - 1, arg.col_offset, "M132", arg.arg) else: for arg in node.args.args: if isinstance(arg, ast.Name) and \ arg.id in self.__builtins: self.__error(arg.lineno - 1, arg.col_offset, "M132", arg.id) def __checkComprehensions(self): """ Private method to check some comprehension related things. """ for node in ast.walk(self.__tree): if (isinstance(node, ast.Call) and len(node.args) == 1 and isinstance(node.func, ast.Name)): if (isinstance(node.args[0], ast.GeneratorExp) and node.func.id in ('list', 'set', 'dict')): errorCode = { "dict": "M193", "list": "M191", "set": "M192", }[node.func.id] self.__error(node.lineno - 1, node.col_offset, errorCode) elif (isinstance(node.args[0], ast.ListComp) and node.func.id in ('set', 'dict')): errorCode = { 'dict': 'M195', 'set': 'M194', }[node.func.id] self.__error(node.lineno - 1, node.col_offset, errorCode) elif (isinstance(node.args[0], ast.List) and node.func.id in ('set', 'dict')): errorCode = { 'dict': 'M197', 'set': 'M196', }[node.func.id] self.__error(node.lineno - 1, node.col_offset, errorCode) elif (isinstance(node.args[0], ast.ListComp) and node.func.id in ('all', 'any', 'frozenset', 'max', 'min', 'sorted', 'sum', 'tuple',)): self.__error(node.lineno - 1, node.col_offset, "M198", node.func.id) def __checkMutableDefault(self): """ Private method to check for use of mutable types as default arguments. """ mutableTypes = ( ast.Call, ast.Dict, ast.List, ast.Set, ) mutableCalls = ( "Counter", "OrderedDict", "collections.Counter", "collections.OrderedDict", "collections.defaultdict", "collections.deque", "defaultdict", "deque", "dict", "list", "set", ) functionDefs = [ast.FunctionDef] try: functionDefs.append(ast.AsyncFunctionDef) except AttributeError: pass for node in ast.walk(self.__tree): if any(isinstance(node, functionDef) for functionDef in functionDefs): for default in node.args.defaults: if any(isinstance(default, mutableType) for mutableType in mutableTypes): typeName = type(default).__name__ if isinstance(default, ast.Call): callPath = '.'.join(composeCallPath(default.func)) if callPath in mutableCalls: self.__error(default.lineno - 1, default.col_offset, "M823", callPath + "()") else: self.__error(default.lineno - 1, default.col_offset, "M822", typeName) else: self.__error(default.lineno - 1, default.col_offset, "M821", typeName) def __dictShouldBeChecked(self, node): """ Private function to test, if the node should be checked. @param node reference to the AST node @return flag indicating to check the node @rtype bool """ if not all(isinstance(key, ast.Str) for key in node.keys): return False if "__IGNORE_WARNING__" in self.__source[node.lineno - 1] or \ "__IGNORE_WARNING_M201__" in self.__source[node.lineno - 1]: return False lineNumbers = [key.lineno for key in node.keys] return len(lineNumbers) == len(set(lineNumbers)) def __checkDictWithSortedKeys(self): """ Private method to check, if dictionary keys appear in sorted order. """ for node in ast.walk(self.__tree): if isinstance(node, ast.Dict) and self.__dictShouldBeChecked(node): for key1, key2 in zip(node.keys, node.keys[1:]): if key2.s < key1.s: self.__error(key2.lineno - 1, key2.col_offset, "M201", key2.s, key1.s) def __checkLogging(self): """ Private method to check logging statements. """ visitor = LoggingVisitor() visitor.visit(self.__tree) for node, reason in visitor.violations: self.__error(node.lineno - 1, node.col_offset, reason) def __checkGettext(self): """ Private method to check the 'gettext' import statement. """ for node in ast.walk(self.__tree): if isinstance(node, ast.ImportFrom) and \ any(name.asname == '_' for name in node.names): self.__error(node.lineno - 1, node.col_offset, "M711", node.names[0].name) def __checkBugBear(self): """ Private method to bugbear checks. """ visitor = BugBearVisitor() visitor.visit(self.__tree) for violation in visitor.violations: node = violation[0] reason = violation[1] params = violation[2:] self.__error(node.lineno - 1, node.col_offset, reason, *params) class TextVisitor(ast.NodeVisitor): """ Class implementing a node visitor for bytes and str instances. It tries to detect docstrings as string of the first expression of each module, class or function. """ # modelled after the string format flake8 extension def __init__(self): """ Constructor """ super(TextVisitor, self).__init__() self.nodes = [] self.calls = {} def __addNode(self, node): """ Private method to add a node to our list of nodes. @param node reference to the node to add @type ast.AST """ if not hasattr(node, 'is_docstring'): node.is_docstring = False self.nodes.append(node) def __isBaseString(self, node): """ Private method to determine, if a node is a base string node. @param node reference to the node to check @type ast.AST @return flag indicating a base string @rtype bool """ typ = (ast.Str,) if sys.version_info[0] > 2: typ += (ast.Bytes,) return isinstance(node, typ) def visit_Str(self, node): """ Public method to record a string node. @param node reference to the string node @type ast.Str """ self.__addNode(node) def visit_Bytes(self, node): """ Public method to record a bytes node. @param node reference to the bytes node @type ast.Bytes """ self.__addNode(node) def __visitDefinition(self, node): """ Private method handling class and function definitions. @param node reference to the node to handle @type ast.FunctionDef, ast.AsyncFunctionDef or ast.ClassDef """ # Manually traverse class or function definition # * Handle decorators normally # * Use special check for body content # * Don't handle the rest (e.g. bases) for decorator in node.decorator_list: self.visit(decorator) self.__visitBody(node) def __visitBody(self, node): """ Private method to traverse the body of the node manually. If the first node is an expression which contains a string or bytes it marks that as a docstring. @param node reference to the node to traverse @type ast.AST """ if (node.body and isinstance(node.body[0], ast.Expr) and self.__isBaseString(node.body[0].value)): node.body[0].value.is_docstring = True for subnode in node.body: self.visit(subnode) def visit_Module(self, node): """ Public method to handle a module. @param node reference to the node to handle @type ast.Module """ self.__visitBody(node) def visit_ClassDef(self, node): """ Public method to handle a class definition. @param node reference to the node to handle @type ast.ClassDef """ # Skipped nodes: ('name', 'bases', 'keywords', 'starargs', 'kwargs') self.__visitDefinition(node) def visit_FunctionDef(self, node): """ Public method to handle a function definition. @param node reference to the node to handle @type ast.FunctionDef """ # Skipped nodes: ('name', 'args', 'returns') self.__visitDefinition(node) def visit_AsyncFunctionDef(self, node): """ Public method to handle an asynchronous function definition. @param node reference to the node to handle @type ast.AsyncFunctionDef """ # Skipped nodes: ('name', 'args', 'returns') self.__visitDefinition(node) def visit_Call(self, node): """ Public method to handle a function call. @param node reference to the node to handle @type ast.Call """ if (isinstance(node.func, ast.Attribute) and node.func.attr == 'format'): if self.__isBaseString(node.func.value): self.calls[node.func.value] = (node, False) elif (isinstance(node.func.value, ast.Name) and node.func.value.id == 'str' and node.args and self.__isBaseString(node.args[0])): self.calls[node.args[0]] = (node, True) super(TextVisitor, self).generic_visit(node) class LoggingVisitor(ast.NodeVisitor): """ Class implementing a node visitor to check logging statements. """ LoggingLevels = { "debug", "critical", "error", "info", "warn", "warning", } def __init__(self): """ Constructor """ super(LoggingVisitor, self).__init__() self.__currentLoggingCall = None self.__currentLoggingArgument = None self.__currentLoggingLevel = None self.__currentExtraKeyword = None self.violations = [] def __withinLoggingStatement(self): """ Private method to check, if we are inside a logging statement. @return flag indicating we are inside a logging statement @rtype bool """ return self.__currentLoggingCall is not None def __withinLoggingArgument(self): """ Private method to check, if we are inside a logging argument. @return flag indicating we are inside a logging argument @rtype bool """ return self.__currentLoggingArgument is not None def __withinExtraKeyword(self, node): """ Private method to check, if we are inside the extra keyword. @param node reference to the node to be checked @type ast.keyword @return flag indicating we are inside the extra keyword @rtype bool """ return self.__currentExtraKeyword is not None and \ self.__currentExtraKeyword != node def __detectLoggingLevel(self, node): """ Private method to decide whether an AST Call is a logging call. @param node reference to the node to be processed @type ast.Call @return logging level @rtype str or None """ try: if node.func.value.id == "warnings": return None if node.func.attr in LoggingVisitor.LoggingLevels: return node.func.attr except AttributeError: pass return None def __isFormatCall(self, node): """ Private method to check if a function call uses format. @param node reference to the node to be processed @type ast.Call @return flag indicating the function call uses format @rtype bool """ try: return node.func.attr == "format" except AttributeError: return False def visit_Call(self, node): """ Public method to handle a function call. Every logging statement and string format is expected to be a function call. @param node reference to the node to be processed @type ast.Call """ # we are in a logging statement if self.__withinLoggingStatement(): if self.__withinLoggingArgument() and self.__isFormatCall(node): self.violations.append((node, "M651")) super(LoggingVisitor, self).generic_visit(node) return loggingLevel = self.__detectLoggingLevel(node) if loggingLevel and self.__currentLoggingLevel is None: self.__currentLoggingLevel = loggingLevel # we are in some other statement if loggingLevel is None: super(LoggingVisitor, self).generic_visit(node) return # we are entering a new logging statement self.__currentLoggingCall = node if loggingLevel == "warn": self.violations.append((node, "M655")) for index, child in enumerate(ast.iter_child_nodes(node)): if index == 1: self.__currentLoggingArgument = child if index > 1 and isinstance(child, ast.keyword) and \ child.arg == "extra": self.__currentExtraKeyword = child super(LoggingVisitor, self).visit(child) self.__currentLoggingArgument = None self.__currentExtraKeyword = None self.__currentLoggingCall = None self.__currentLoggingLevel = None def visit_BinOp(self, node): """ Public method to handle binary operations while processing the first logging argument. @param node reference to the node to be processed @type ast.BinOp """ if self.__withinLoggingStatement() and self.__withinLoggingArgument(): # handle percent format if isinstance(node.op, ast.Mod): self.violations.append((node, "M652")) # handle string concat if isinstance(node.op, ast.Add): self.violations.append((node, "M653")) super(LoggingVisitor, self).generic_visit(node) def visit_JoinedStr(self, node): """ Public method to handle f-string arguments. @param node reference to the node to be processed @type ast.JoinedStr """ if sys.version_info >= (3, 6): if self.__withinLoggingStatement(): if any(isinstance(i, ast.FormattedValue) for i in node.values): if self.__withinLoggingArgument(): self.violations.append((node, "M654")) super(LoggingVisitor, self).generic_visit(node) class BugBearVisitor(ast.NodeVisitor): """ Class implementing a node visitor to check for various topics. """ # # This class was implemented along the BugBear flake8 extension (v 18.2.0). # Original: Copyright (c) 2016 Ćukasz Langa # NodeWindowSize = 4 def __init__(self): """ Constructor """ super(BugBearVisitor, self).__init__() self.__nodeStack = [] self.__nodeWindow = [] self.violations = [] def visit(self, node): """ Public method to traverse a given AST node. @param node AST node to be traversed @type ast.Node """ self.__nodeStack.append(node) self.__nodeWindow.append(node) self.__nodeWindow = \ self.__nodeWindow[-BugBearVisitor.NodeWindowSize:] super(BugBearVisitor, self).visit(node) self.__nodeStack.pop() def visit_UAdd(self, node): """ Public method to handle unary additions. @param node reference to the node to be processed @type ast.UAdd """ trailingNodes = list(map(type, self.__nodeWindow[-4:])) if trailingNodes == [ast.UnaryOp, ast.UAdd, ast.UnaryOp, ast.UAdd]: originator = self.__nodeWindow[-4] self.violations.append((originator, "M501")) self.generic_visit(node) def visit_Call(self, node): """ Public method to handle a function call. @param node reference to the node to be processed @type ast.Call """ if sys.version_info >= (3, 0): validPaths = ("six", "future.utils", "builtins") methodsDict = { "M511": ("iterkeys", "itervalues", "iteritems", "iterlists"), "M512": ("viewkeys", "viewvalues", "viewitems", "viewlists"), "M513": ("next",), } else: validPaths = () methodsDict = {} if isinstance(node.func, ast.Attribute): for code, methods in methodsDict.items(): if node.func.attr in methods: callPath = ".".join(composeCallPath(node.func.value)) if callPath not in validPaths: self.violations.append((node, code)) break else: self.__checkForM502(node) else: try: if ( node.func.id in ("getattr", "hasattr") and node.args[1].s == "__call__" ): self.violations.append((node, "M503")) except (AttributeError, IndexError): pass self.generic_visit(node) def visit_Attribute(self, node): """ Public method to handle attributes. @param node reference to the node to be processed @type ast.Attribute """ callPath = list(composeCallPath(node)) if '.'.join(callPath) == 'sys.maxint' and sys.version_info >= (3, 0): self.violations.append((node, "M504")) elif len(callPath) == 2 and callPath[1] == 'message' and \ sys.version_info >= (2, 6): name = callPath[0] for elem in reversed(self.__nodeStack[:-1]): if isinstance(elem, ast.ExceptHandler) and elem.name == name: self.violations.append((node, "M505")) break def visit_Assign(self, node): """ Public method to handle assignments. @param node reference to the node to be processed @type ast.Assign """ if isinstance(self.__nodeStack[-2], ast.ClassDef): # By using 'hasattr' below we're ignoring starred arguments, slices # and tuples for simplicity. assignTargets = {t.id for t in node.targets if hasattr(t, 'id')} if '__metaclass__' in assignTargets and sys.version_info >= (3, 0): self.violations.append((node, "M514")) elif len(node.targets) == 1: target = node.targets[0] if isinstance(target, ast.Attribute) and \ isinstance(target.value, ast.Name): if (target.value.id, target.attr) == ('os', 'environ'): self.violations.append((node, "M506")) self.generic_visit(node) def visit_For(self, node): """ Public method to handle 'for' statements. @param node reference to the node to be processed @type ast.For """ self.__checkForM507(node) self.generic_visit(node) def __checkForM502(self, node): """ Private method to check the use of *strip(). @param node reference to the node to be processed @type ast.Call """ if node.func.attr not in ("lstrip", "rstrip", "strip"): return # method name doesn't match if len(node.args) != 1 or not isinstance(node.args[0], ast.Str): return # used arguments don't match the builtin strip s = node.args[0].s if len(s) == 1: return # stripping just one character if len(s) == len(set(s)): return # no characters appear more than once self.violations.append((node, "M502")) def __checkForM507(self, node): """ Private method to check for unused loop variables. @param node reference to the node to be processed @type ast.For """ targets = NameFinder() targets.visit(node.target) ctrlNames = set(filter(lambda s: not s.startswith('_'), targets.getNames())) body = NameFinder() for expr in node.body: body.visit(expr) usedNames = set(body.getNames()) for name in sorted(ctrlNames - usedNames): n = targets.getNames()[name][0] self.violations.append((n, "M507", name)) class NameFinder(ast.NodeVisitor): """ Class to extract a name out of a tree of nodes. """ def __init__(self): """ Constructor """ super(NameFinder, self).__init__() self.__names = {} def visit_Name(self, node): """ Public method to handle 'Name' nodes. @param node reference to the node to be processed @type ast.Name """ self.__names.setdefault(node.id, []).append(node) def visit(self, node): """ Public method to traverse a given AST node. @param node AST node to be traversed @type ast.Node """ if isinstance(node, list): for elem in node: super(NameFinder, self).visit(elem) else: super(NameFinder, self).visit(node) def getNames(self): """ Public method to return the extracted names and Name nodes. @return dictionary containing the names as keys and the list of nodes @rtype dict """ return self.__names # # eflag: noqa = M702