Server IP : 66.29.132.122 / Your IP : 18.119.109.232 Web Server : LiteSpeed System : Linux business142.web-hosting.com 4.18.0-553.lve.el8.x86_64 #1 SMP Mon May 27 15:27:34 UTC 2024 x86_64 User : admazpex ( 531) PHP Version : 7.2.34 Disable Function : NONE MySQL : OFF | cURL : ON | WGET : ON | Perl : ON | Python : ON | Sudo : OFF | Pkexec : OFF Directory : /proc/self/root/proc/self/root/proc/thread-self/root/proc/thread-self/root/proc/self/root/proc/thread-self/root/proc/self/root/proc/self/root/opt/cloudlinux/venv/lib64/python3.11/site-packages/pylint/checkers/ |
Upload File : |
# Licensed under the GPL: https://www.gnu.org/licenses/old-licenses/gpl-2.0.html # For details: https://github.com/PyCQA/pylint/blob/main/LICENSE # Copyright (c) https://github.com/PyCQA/pylint/blob/main/CONTRIBUTORS.txt """Checker for string formatting operations.""" from __future__ import annotations import collections import re import sys import tokenize from collections import Counter from collections.abc import Iterable, Sequence from typing import TYPE_CHECKING import astroid from astroid import bases, nodes, util from astroid.typing import SuccessfulInferenceResult from pylint.checkers import BaseChecker, BaseRawFileChecker, BaseTokenChecker, utils from pylint.checkers.utils import only_required_for_messages from pylint.interfaces import HIGH from pylint.typing import MessageDefinitionTuple if TYPE_CHECKING: from pylint.lint import PyLinter if sys.version_info >= (3, 8): from typing import Literal else: from typing_extensions import Literal _AST_NODE_STR_TYPES = ("__builtin__.unicode", "__builtin__.str", "builtins.str") # Prefixes for both strings and bytes literals per # https://docs.python.org/3/reference/lexical_analysis.html#string-and-bytes-literals _PREFIXES = { "r", "u", "R", "U", "f", "F", "fr", "Fr", "fR", "FR", "rf", "rF", "Rf", "RF", "b", "B", "br", "Br", "bR", "BR", "rb", "rB", "Rb", "RB", } SINGLE_QUOTED_REGEX = re.compile(f"({'|'.join(_PREFIXES)})?'''") DOUBLE_QUOTED_REGEX = re.compile(f"({'|'.join(_PREFIXES)})?\"\"\"") QUOTE_DELIMITER_REGEX = re.compile(f"({'|'.join(_PREFIXES)})?(\"|')", re.DOTALL) MSGS: dict[ str, MessageDefinitionTuple ] = { # pylint: disable=consider-using-namedtuple-or-dataclass "E1300": ( "Unsupported format character %r (%#02x) at index %d", "bad-format-character", "Used when an unsupported format character is used in a format string.", ), "E1301": ( "Format string ends in middle of conversion specifier", "truncated-format-string", "Used when a format string terminates before the end of a " "conversion specifier.", ), "E1302": ( "Mixing named and unnamed conversion specifiers in format string", "mixed-format-string", "Used when a format string contains both named (e.g. '%(foo)d') " "and unnamed (e.g. '%d') conversion specifiers. This is also " "used when a named conversion specifier contains * for the " "minimum field width and/or precision.", ), "E1303": ( "Expected mapping for format string, not %s", "format-needs-mapping", "Used when a format string that uses named conversion specifiers " "is used with an argument that is not a mapping.", ), "W1300": ( "Format string dictionary key should be a string, not %s", "bad-format-string-key", "Used when a format string that uses named conversion specifiers " "is used with a dictionary whose keys are not all strings.", ), "W1301": ( "Unused key %r in format string dictionary", "unused-format-string-key", "Used when a format string that uses named conversion specifiers " "is used with a dictionary that contains keys not required by the " "format string.", ), "E1304": ( "Missing key %r in format string dictionary", "missing-format-string-key", "Used when a format string that uses named conversion specifiers " "is used with a dictionary that doesn't contain all the keys " "required by the format string.", ), "E1305": ( "Too many arguments for format string", "too-many-format-args", "Used when a format string that uses unnamed conversion " "specifiers is given too many arguments.", ), "E1306": ( "Not enough arguments for format string", "too-few-format-args", "Used when a format string that uses unnamed conversion " "specifiers is given too few arguments", ), "E1307": ( "Argument %r does not match format type %r", "bad-string-format-type", "Used when a type required by format string " "is not suitable for actual argument type", ), "E1310": ( "Suspicious argument in %s.%s call", "bad-str-strip-call", "The argument to a str.{l,r,}strip call contains a duplicate character,", ), "W1302": ( "Invalid format string", "bad-format-string", "Used when a PEP 3101 format string is invalid.", ), "W1303": ( "Missing keyword argument %r for format string", "missing-format-argument-key", "Used when a PEP 3101 format string that uses named fields " "doesn't receive one or more required keywords.", ), "W1304": ( "Unused format argument %r", "unused-format-string-argument", "Used when a PEP 3101 format string that uses named " "fields is used with an argument that " "is not required by the format string.", ), "W1305": ( "Format string contains both automatic field numbering " "and manual field specification", "format-combined-specification", "Used when a PEP 3101 format string contains both automatic " "field numbering (e.g. '{}') and manual field " "specification (e.g. '{0}').", ), "W1306": ( "Missing format attribute %r in format specifier %r", "missing-format-attribute", "Used when a PEP 3101 format string uses an " "attribute specifier ({0.length}), but the argument " "passed for formatting doesn't have that attribute.", ), "W1307": ( "Using invalid lookup key %r in format specifier %r", "invalid-format-index", "Used when a PEP 3101 format string uses a lookup specifier " "({a[1]}), but the argument passed for formatting " "doesn't contain or doesn't have that key as an attribute.", ), "W1308": ( "Duplicate string formatting argument %r, consider passing as named argument", "duplicate-string-formatting-argument", "Used when we detect that a string formatting is " "repeating an argument instead of using named string arguments", ), "W1309": ( "Using an f-string that does not have any interpolated variables", "f-string-without-interpolation", "Used when we detect an f-string that does not use any interpolation variables, " "in which case it can be either a normal string or a bug in the code.", ), "W1310": ( "Using formatting for a string that does not have any interpolated variables", "format-string-without-interpolation", "Used when we detect a string that does not have any interpolation variables, " "in which case it can be either a normal string without formatting or a bug in the code.", ), } OTHER_NODES = ( nodes.Const, nodes.List, nodes.Lambda, nodes.FunctionDef, nodes.ListComp, nodes.SetComp, nodes.GeneratorExp, ) def get_access_path(key: str | Literal[0], parts: list[tuple[bool, str]]) -> str: """Given a list of format specifiers, returns the final access path (e.g. a.b.c[0][1]). """ path = [] for is_attribute, specifier in parts: if is_attribute: path.append(f".{specifier}") else: path.append(f"[{specifier!r}]") return str(key) + "".join(path) def arg_matches_format_type( arg_type: SuccessfulInferenceResult, format_type: str ) -> bool: if format_type in "sr": # All types can be printed with %s and %r return True if isinstance(arg_type, astroid.Instance): arg_type = arg_type.pytype() if arg_type == "builtins.str": return format_type == "c" if arg_type == "builtins.float": return format_type in "deEfFgGn%" if arg_type == "builtins.int": # Integers allow all types return True return False return True class StringFormatChecker(BaseChecker): """Checks string formatting operations to ensure that the format string is valid and the arguments match the format string. """ name = "string" msgs = MSGS # pylint: disable = too-many-branches, too-many-locals, too-many-statements @only_required_for_messages( "bad-format-character", "truncated-format-string", "mixed-format-string", "bad-format-string-key", "missing-format-string-key", "unused-format-string-key", "bad-string-format-type", "format-needs-mapping", "too-many-format-args", "too-few-format-args", "format-string-without-interpolation", ) def visit_binop(self, node: nodes.BinOp) -> None: if node.op != "%": return left = node.left args = node.right if not (isinstance(left, nodes.Const) and isinstance(left.value, str)): return format_string = left.value try: ( required_keys, required_num_args, required_key_types, required_arg_types, ) = utils.parse_format_string(format_string) except utils.UnsupportedFormatCharacter as exc: formatted = format_string[exc.index] self.add_message( "bad-format-character", node=node, args=(formatted, ord(formatted), exc.index), ) return except utils.IncompleteFormatString: self.add_message("truncated-format-string", node=node) return if not required_keys and not required_num_args: self.add_message("format-string-without-interpolation", node=node) return if required_keys and required_num_args: # The format string uses both named and unnamed format # specifiers. self.add_message("mixed-format-string", node=node) elif required_keys: # The format string uses only named format specifiers. # Check that the RHS of the % operator is a mapping object # that contains precisely the set of keys required by the # format string. if isinstance(args, nodes.Dict): keys = set() unknown_keys = False for k, _ in args.items: if isinstance(k, nodes.Const): key = k.value if isinstance(key, str): keys.add(key) else: self.add_message( "bad-format-string-key", node=node, args=key ) else: # One of the keys was something other than a # constant. Since we can't tell what it is, # suppress checks for missing keys in the # dictionary. unknown_keys = True if not unknown_keys: for key in required_keys: if key not in keys: self.add_message( "missing-format-string-key", node=node, args=key ) for key in keys: if key not in required_keys: self.add_message( "unused-format-string-key", node=node, args=key ) for key, arg in args.items: if not isinstance(key, nodes.Const): continue format_type = required_key_types.get(key.value, None) arg_type = utils.safe_infer(arg) if ( format_type is not None and arg_type and not isinstance(arg_type, util.UninferableBase) and not arg_matches_format_type(arg_type, format_type) ): self.add_message( "bad-string-format-type", node=node, args=(arg_type.pytype(), format_type), ) elif isinstance(args, (OTHER_NODES, nodes.Tuple)): type_name = type(args).__name__ self.add_message("format-needs-mapping", node=node, args=type_name) # else: # The RHS of the format specifier is a name or # expression. It may be a mapping object, so # there's nothing we can check. else: # The format string uses only unnamed format specifiers. # Check that the number of arguments passed to the RHS of # the % operator matches the number required by the format # string. args_elts = [] if isinstance(args, nodes.Tuple): rhs_tuple = utils.safe_infer(args) num_args = None if isinstance(rhs_tuple, nodes.BaseContainer): args_elts = rhs_tuple.elts num_args = len(args_elts) elif isinstance(args, (OTHER_NODES, (nodes.Dict, nodes.DictComp))): args_elts = [args] num_args = 1 elif isinstance(args, nodes.Name): inferred = utils.safe_infer(args) if isinstance(inferred, nodes.Tuple): # The variable is a tuple, so we need to get the elements # from it for further inspection args_elts = inferred.elts num_args = len(args_elts) elif isinstance(inferred, nodes.Const): args_elts = [inferred] num_args = 1 else: num_args = None else: # The RHS of the format specifier is an expression. # It could be a tuple of unknown size, so # there's nothing we can check. num_args = None if num_args is not None: if num_args > required_num_args: self.add_message("too-many-format-args", node=node) elif num_args < required_num_args: self.add_message("too-few-format-args", node=node) for arg, format_type in zip(args_elts, required_arg_types): if not arg: continue arg_type = utils.safe_infer(arg) if ( arg_type and not isinstance(arg_type, util.UninferableBase) and not arg_matches_format_type(arg_type, format_type) ): self.add_message( "bad-string-format-type", node=node, args=(arg_type.pytype(), format_type), ) @only_required_for_messages("f-string-without-interpolation") def visit_joinedstr(self, node: nodes.JoinedStr) -> None: self._check_interpolation(node) def _check_interpolation(self, node: nodes.JoinedStr) -> None: if isinstance(node.parent, nodes.FormattedValue): return for value in node.values: if isinstance(value, nodes.FormattedValue): return self.add_message("f-string-without-interpolation", node=node) def visit_call(self, node: nodes.Call) -> None: func = utils.safe_infer(node.func) if ( isinstance(func, astroid.BoundMethod) and isinstance(func.bound, astroid.Instance) and func.bound.name in {"str", "unicode", "bytes"} ): if func.name in {"strip", "lstrip", "rstrip"} and node.args: arg = utils.safe_infer(node.args[0]) if not isinstance(arg, nodes.Const) or not isinstance(arg.value, str): return if len(arg.value) != len(set(arg.value)): self.add_message( "bad-str-strip-call", node=node, args=(func.bound.name, func.name), ) elif func.name == "format": self._check_new_format(node, func) def _detect_vacuous_formatting( self, node: nodes.Call, positional_arguments: list[SuccessfulInferenceResult] ) -> None: counter = collections.Counter( arg.name for arg in positional_arguments if isinstance(arg, nodes.Name) ) for name, count in counter.items(): if count == 1: continue self.add_message( "duplicate-string-formatting-argument", node=node, args=(name,) ) def _check_new_format(self, node: nodes.Call, func: bases.BoundMethod) -> None: """Check the new string formatting.""" # Skip format nodes which don't have an explicit string on the # left side of the format operation. # We do this because our inference engine can't properly handle # redefinition of the original string. # Note that there may not be any left side at all, if the format method # has been assigned to another variable. See issue 351. For example: # # fmt = 'some string {}'.format # fmt('arg') if isinstance(node.func, nodes.Attribute) and not isinstance( node.func.expr, nodes.Const ): return if node.starargs or node.kwargs: return try: strnode = next(func.bound.infer()) except astroid.InferenceError: return if not (isinstance(strnode, nodes.Const) and isinstance(strnode.value, str)): return try: call_site = astroid.arguments.CallSite.from_call(node) except astroid.InferenceError: return try: fields, num_args, manual_pos = utils.parse_format_method_string( strnode.value ) except utils.IncompleteFormatString: self.add_message("bad-format-string", node=node) return positional_arguments = call_site.positional_arguments named_arguments = call_site.keyword_arguments named_fields = {field[0] for field in fields if isinstance(field[0], str)} if num_args and manual_pos: self.add_message("format-combined-specification", node=node) return check_args = False # Consider "{[0]} {[1]}" as num_args. num_args += sum(1 for field in named_fields if not field) if named_fields: for field in named_fields: if field and field not in named_arguments: self.add_message( "missing-format-argument-key", node=node, args=(field,) ) for field in named_arguments: if field not in named_fields: self.add_message( "unused-format-string-argument", node=node, args=(field,) ) # num_args can be 0 if manual_pos is not. num_args = num_args or manual_pos if positional_arguments or num_args: empty = not all(field for field in named_fields) if named_arguments or empty: # Verify the required number of positional arguments # only if the .format got at least one keyword argument. # This means that the format strings accepts both # positional and named fields and we should warn # when one of them is missing or is extra. check_args = True else: check_args = True if check_args: # num_args can be 0 if manual_pos is not. num_args = num_args or manual_pos if not num_args: self.add_message("format-string-without-interpolation", node=node) return if len(positional_arguments) > num_args: self.add_message("too-many-format-args", node=node) elif len(positional_arguments) < num_args: self.add_message("too-few-format-args", node=node) self._detect_vacuous_formatting(node, positional_arguments) self._check_new_format_specifiers(node, fields, named_arguments) # pylint: disable = too-many-statements def _check_new_format_specifiers( self, node: nodes.Call, fields: list[tuple[str, list[tuple[bool, str]]]], named: dict[str, SuccessfulInferenceResult], ) -> None: """Check attribute and index access in the format string ("{0.a}" and "{0[a]}"). """ key: Literal[0] | str for key, specifiers in fields: # Obtain the argument. If it can't be obtained # or inferred, skip this check. if not key: # {[0]} will have an unnamed argument, defaulting # to 0. It will not be present in `named`, so use the value # 0 for it. key = 0 if isinstance(key, int): try: argname = utils.get_argument_from_call(node, key) except utils.NoSuchArgumentError: continue else: if key not in named: continue argname = named[key] if argname is None or isinstance(argname, util.UninferableBase): continue try: argument = utils.safe_infer(argname) except astroid.InferenceError: continue if not specifiers or not argument: # No need to check this key if it doesn't # use attribute / item access continue if argument.parent and isinstance(argument.parent, nodes.Arguments): # Ignore any object coming from an argument, # because we can't infer its value properly. continue previous = argument parsed: list[tuple[bool, str]] = [] for is_attribute, specifier in specifiers: if isinstance(previous, util.UninferableBase): break parsed.append((is_attribute, specifier)) if is_attribute: try: previous = previous.getattr(specifier)[0] except astroid.NotFoundError: if ( hasattr(previous, "has_dynamic_getattr") and previous.has_dynamic_getattr() ): # Don't warn if the object has a custom __getattr__ break path = get_access_path(key, parsed) self.add_message( "missing-format-attribute", args=(specifier, path), node=node, ) break else: warn_error = False if hasattr(previous, "getitem"): try: previous = previous.getitem(nodes.Const(specifier)) except ( astroid.AstroidIndexError, astroid.AstroidTypeError, astroid.AttributeInferenceError, ): warn_error = True except astroid.InferenceError: break if isinstance(previous, util.UninferableBase): break else: try: # Lookup __getitem__ in the current node, # but skip further checks, because we can't # retrieve the looked object previous.getattr("__getitem__") break except astroid.NotFoundError: warn_error = True if warn_error: path = get_access_path(key, parsed) self.add_message( "invalid-format-index", args=(specifier, path), node=node ) break try: previous = next(previous.infer()) except astroid.InferenceError: # can't check further if we can't infer it break class StringConstantChecker(BaseTokenChecker, BaseRawFileChecker): """Check string literals.""" name = "string" msgs = { "W1401": ( "Anomalous backslash in string: '%s'. " "String constant might be missing an r prefix.", "anomalous-backslash-in-string", "Used when a backslash is in a literal string but not as an escape.", ), "W1402": ( "Anomalous Unicode escape in byte string: '%s'. " "String constant might be missing an r or u prefix.", "anomalous-unicode-escape-in-string", "Used when an escape like \\u is encountered in a byte " "string where it has no effect.", ), "W1404": ( "Implicit string concatenation found in %s", "implicit-str-concat", "String literals are implicitly concatenated in a " "literal iterable definition : " "maybe a comma is missing ?", {"old_names": [("W1403", "implicit-str-concat-in-sequence")]}, ), "W1405": ( "Quote delimiter %s is inconsistent with the rest of the file", "inconsistent-quotes", "Quote delimiters are not used consistently throughout a module " "(with allowances made for avoiding unnecessary escaping).", ), "W1406": ( "The u prefix for strings is no longer necessary in Python >=3.0", "redundant-u-string-prefix", "Used when we detect a string with a u prefix. These prefixes were necessary " "in Python 2 to indicate a string was Unicode, but since Python 3.0 strings " "are Unicode by default.", ), } options = ( ( "check-str-concat-over-line-jumps", { "default": False, "type": "yn", "metavar": "<y or n>", "help": "This flag controls whether the " "implicit-str-concat should generate a warning " "on implicit string concatenation in sequences defined over " "several lines.", }, ), ( "check-quote-consistency", { "default": False, "type": "yn", "metavar": "<y or n>", "help": "This flag controls whether inconsistent-quotes generates a " "warning when the character used as a quote delimiter is used " "inconsistently within a module.", }, ), ) # Characters that have a special meaning after a backslash in either # Unicode or byte strings. ESCAPE_CHARACTERS = "abfnrtvx\n\r\t\\'\"01234567" # Characters that have a special meaning after a backslash but only in # Unicode strings. UNICODE_ESCAPE_CHARACTERS = "uUN" def __init__(self, linter: PyLinter) -> None: super().__init__(linter) self.string_tokens: dict[ tuple[int, int], tuple[str, tokenize.TokenInfo | None] ] = {} """Token position -> (token value, next token).""" def process_module(self, node: nodes.Module) -> None: self._unicode_literals = "unicode_literals" in node.future_imports def process_tokens(self, tokens: list[tokenize.TokenInfo]) -> None: encoding = "ascii" for i, (token_type, token, start, _, line) in enumerate(tokens): if token_type == tokenize.ENCODING: # this is always the first token processed encoding = token elif token_type == tokenize.STRING: # 'token' is the whole un-parsed token; we can look at the start # of it to see whether it's a raw or unicode string etc. self.process_string_token(token, start[0], start[1]) # We figure the next token, ignoring comments & newlines: j = i + 1 while j < len(tokens) and tokens[j].type in ( tokenize.NEWLINE, tokenize.NL, tokenize.COMMENT, ): j += 1 next_token = tokens[j] if j < len(tokens) else None if encoding != "ascii": # We convert `tokenize` character count into a byte count, # to match with astroid `.col_offset` start = (start[0], len(line[: start[1]].encode(encoding))) self.string_tokens[start] = (str_eval(token), next_token) if self.linter.config.check_quote_consistency: self.check_for_consistent_string_delimiters(tokens) @only_required_for_messages("implicit-str-concat") def visit_call(self, node: nodes.Call) -> None: self.check_for_concatenated_strings(node.args, "call") @only_required_for_messages("implicit-str-concat") def visit_list(self, node: nodes.List) -> None: self.check_for_concatenated_strings(node.elts, "list") @only_required_for_messages("implicit-str-concat") def visit_set(self, node: nodes.Set) -> None: self.check_for_concatenated_strings(node.elts, "set") @only_required_for_messages("implicit-str-concat") def visit_tuple(self, node: nodes.Tuple) -> None: self.check_for_concatenated_strings(node.elts, "tuple") def visit_assign(self, node: nodes.Assign) -> None: if isinstance(node.value, nodes.Const) and isinstance(node.value.value, str): self.check_for_concatenated_strings([node.value], "assignment") def check_for_consistent_string_delimiters( self, tokens: Iterable[tokenize.TokenInfo] ) -> None: """Adds a message for each string using inconsistent quote delimiters. Quote delimiters are used inconsistently if " and ' are mixed in a module's shortstrings without having done so to avoid escaping an internal quote character. Args: tokens: The tokens to be checked against for consistent usage. """ string_delimiters: Counter[str] = collections.Counter() # First, figure out which quote character predominates in the module for tok_type, token, _, _, _ in tokens: if tok_type == tokenize.STRING and _is_quote_delimiter_chosen_freely(token): string_delimiters[_get_quote_delimiter(token)] += 1 if len(string_delimiters) > 1: # Ties are broken arbitrarily most_common_delimiter = string_delimiters.most_common(1)[0][0] for tok_type, token, start, _, _ in tokens: if tok_type != tokenize.STRING: continue quote_delimiter = _get_quote_delimiter(token) if ( _is_quote_delimiter_chosen_freely(token) and quote_delimiter != most_common_delimiter ): self.add_message( "inconsistent-quotes", line=start[0], args=(quote_delimiter,) ) def check_for_concatenated_strings( self, elements: Sequence[nodes.NodeNG], iterable_type: str ) -> None: for elt in elements: if not ( isinstance(elt, nodes.Const) and elt.pytype() in _AST_NODE_STR_TYPES ): continue if elt.col_offset < 0: # This can happen in case of escaped newlines continue token_index = (elt.lineno, elt.col_offset) if token_index not in self.string_tokens: # This may happen with Latin1 encoding # cf. https://github.com/PyCQA/pylint/issues/2610 continue matching_token, next_token = self.string_tokens[token_index] # We detect string concatenation: the AST Const is the # combination of 2 string tokens if matching_token != elt.value and next_token is not None: if next_token.type == tokenize.STRING and ( next_token.start[0] == elt.lineno or self.linter.config.check_str_concat_over_line_jumps ): self.add_message( "implicit-str-concat", line=elt.lineno, args=(iterable_type,), confidence=HIGH, ) def process_string_token(self, token: str, start_row: int, start_col: int) -> None: quote_char = None for _index, char in enumerate(token): if char in "'\"": quote_char = char break if quote_char is None: return # pylint: disable=undefined-loop-variable prefix = token[:_index].lower() # markers like u, b, r. after_prefix = token[_index:] # pylint: enable=undefined-loop-variable # Chop off quotes quote_length = ( 3 if after_prefix[:3] == after_prefix[-3:] == 3 * quote_char else 1 ) string_body = after_prefix[quote_length:-quote_length] # No special checks on raw strings at the moment. if "r" not in prefix: self.process_non_raw_string_token( prefix, string_body, start_row, start_col + len(prefix) + quote_length, ) def process_non_raw_string_token( self, prefix: str, string_body: str, start_row: int, string_start_col: int ) -> None: """Check for bad escapes in a non-raw string. prefix: lowercase string of string prefix markers ('ur'). string_body: the un-parsed body of the string, not including the quote marks. start_row: line number in the source. string_start_col: col number of the string start in the source. """ # Walk through the string; if we see a backslash then escape the next # character, and skip over it. If we see a non-escaped character, # alert, and continue. # # Accept a backslash when it escapes a backslash, or a quote, or # end-of-line, or one of the letters that introduce a special escape # sequence <https://docs.python.org/reference/lexical_analysis.html> # index = 0 while True: index = string_body.find("\\", index) if index == -1: break # There must be a next character; having a backslash at the end # of the string would be a SyntaxError. next_char = string_body[index + 1] match = string_body[index : index + 2] # The column offset will vary depending on whether the string token # is broken across lines. Calculate relative to the nearest line # break or relative to the start of the token's line. last_newline = string_body.rfind("\n", 0, index) if last_newline == -1: line = start_row col_offset = index + string_start_col else: line = start_row + string_body.count("\n", 0, index) col_offset = index - last_newline - 1 if next_char in self.UNICODE_ESCAPE_CHARACTERS: if "u" in prefix: pass elif "b" not in prefix: pass # unicode by default else: self.add_message( "anomalous-unicode-escape-in-string", line=line, args=(match,), col_offset=col_offset, ) elif next_char not in self.ESCAPE_CHARACTERS: self.add_message( "anomalous-backslash-in-string", line=line, args=(match,), col_offset=col_offset, ) # Whether it was a valid escape or not, backslash followed by # another character can always be consumed whole: the second # character can never be the start of a new backslash escape. index += 2 @only_required_for_messages("redundant-u-string-prefix") def visit_const(self, node: nodes.Const) -> None: if node.pytype() == "builtins.str" and not isinstance( node.parent, nodes.JoinedStr ): self._detect_u_string_prefix(node) def _detect_u_string_prefix(self, node: nodes.Const) -> None: """Check whether strings include a 'u' prefix like u'String'.""" if node.kind == "u": self.add_message( "redundant-u-string-prefix", line=node.lineno, col_offset=node.col_offset, ) def register(linter: PyLinter) -> None: linter.register_checker(StringFormatChecker(linter)) linter.register_checker(StringConstantChecker(linter)) def str_eval(token: str) -> str: """Mostly replicate `ast.literal_eval(token)` manually to avoid any performance hit. This supports f-strings, contrary to `ast.literal_eval`. We have to support all string literal notations: https://docs.python.org/3/reference/lexical_analysis.html#string-and-bytes-literals """ if token[0:2].lower() in {"fr", "rf"}: token = token[2:] elif token[0].lower() in {"r", "u", "f"}: token = token[1:] if token[0:3] in {'"""', "'''"}: return token[3:-3] return token[1:-1] def _is_long_string(string_token: str) -> bool: """Is this string token a "longstring" (is it triple-quoted)? Long strings are triple-quoted as defined in https://docs.python.org/3/reference/lexical_analysis.html#string-and-bytes-literals This function only checks characters up through the open quotes. Because it's meant to be applied only to tokens that represent string literals, it doesn't bother to check for close-quotes (demonstrating that the literal is a well-formed string). Args: string_token: The string token to be parsed. Returns: A boolean representing whether this token matches a longstring regex. """ return bool( SINGLE_QUOTED_REGEX.match(string_token) or DOUBLE_QUOTED_REGEX.match(string_token) ) def _get_quote_delimiter(string_token: str) -> str: """Returns the quote character used to delimit this token string. This function checks whether the token is a well-formed string. Args: string_token: The token to be parsed. Returns: A string containing solely the first quote delimiter character in the given string. Raises: ValueError: No quote delimiter characters are present. """ match = QUOTE_DELIMITER_REGEX.match(string_token) if not match: raise ValueError(f"string token {string_token} is not a well-formed string") return match.group(2) def _is_quote_delimiter_chosen_freely(string_token: str) -> bool: """Was there a non-awkward option for the quote delimiter? Args: string_token: The quoted string whose delimiters are to be checked. Returns: Whether there was a choice in this token's quote character that would not have involved backslash-escaping an interior quote character. Long strings are excepted from this analysis under the assumption that their quote characters are set by policy. """ quote_delimiter = _get_quote_delimiter(string_token) unchosen_delimiter = '"' if quote_delimiter == "'" else "'" return bool( quote_delimiter and not _is_long_string(string_token) and unchosen_delimiter not in str_eval(string_token) )