"""isort.py. Exposes a simple library to sort through imports within Python code usage: SortImports(file_name) or: sorted = SortImports(file_contents=file_contents).output Copyright (C) 2013 Timothy Edmund Crosley Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. """ from __future__ import absolute_import, division, print_function, unicode_literals import copy import io import itertools import os import re import sys import sysconfig from collections import OrderedDict, namedtuple from datetime import datetime from difflib import unified_diff from fnmatch import fnmatch from glob import glob from . import settings from .natural import nsorted from .pie_slice import OrderedSet, input, itemsview KNOWN_SECTION_MAPPING = { 'STDLIB': 'STANDARD_LIBRARY', 'FUTURE': 'FUTURE_LIBRARY', 'FIRSTPARTY': 'FIRST_PARTY', 'THIRDPARTY': 'THIRD_PARTY', } class SortImports(object): incorrectly_sorted = False skipped = False def __init__(self, file_path=None, file_contents=None, write_to_stdout=False, check=False, show_diff=False, settings_path=None, ask_to_apply=False, **setting_overrides): if not settings_path and file_path: settings_path = os.path.dirname(os.path.abspath(file_path)) settings_path = settings_path or os.getcwd() self.config = settings.from_path(settings_path).copy() for key, value in itemsview(setting_overrides): access_key = key.replace('not_', '').lower() # The sections config needs to retain order and can't be converted to a set. if access_key != 'sections' and type(self.config.get(access_key)) in (list, tuple): if key.startswith('not_'): self.config[access_key] = list(set(self.config[access_key]).difference(value)) else: self.config[access_key] = list(set(self.config[access_key]).union(value)) else: self.config[key] = value if self.config['force_alphabetical_sort']: self.config.update({'force_alphabetical_sort_within_sections': True, 'no_sections': True, 'lines_between_types': 1, 'from_first': True}) indent = str(self.config['indent']) if indent.isdigit(): indent = " " * int(indent) else: indent = indent.strip("'").strip('"') if indent.lower() == "tab": indent = "\t" self.config['indent'] = indent self.config['comment_prefix'] = self.config['comment_prefix'].strip("'").strip('"') self.place_imports = {} self.import_placements = {} self.remove_imports = [self._format_simplified(removal) for removal in self.config['remove_imports']] self.add_imports = [self._format_natural(addition) for addition in self.config['add_imports']] self._section_comments = ["# " + value for key, value in itemsview(self.config) if key.startswith('import_heading') and value] self.file_encoding = 'utf-8' file_name = file_path self.file_path = file_path or "" if file_path: file_path = os.path.abspath(file_path) if settings.should_skip(file_path, self.config): self.skipped = True if self.config['verbose']: print("WARNING: {0} was skipped as it's listed in 'skip' setting" " or matches a glob in 'skip_glob' setting".format(file_path)) file_contents = None elif not file_contents: self.file_path = file_path self.file_encoding = coding_check(file_path) with io.open(file_path, encoding=self.file_encoding, newline='') as file_to_import_sort: file_contents = file_to_import_sort.read() if file_contents is None or ("isort:" + "skip_file") in file_contents: self.skipped = True self.output = None return if self.config['line_ending']: self.line_separator = self.config['line_ending'] else: if '\r\n' in file_contents: self.line_separator = '\r\n' elif '\r' in file_contents: self.line_separator = '\r' else: self.line_separator = '\n' self.in_lines = file_contents.split(self.line_separator) self.original_length = len(self.in_lines) if (self.original_length > 1 or self.in_lines[:1] not in ([], [""])) or self.config['force_adds']: for add_import in self.add_imports: self.in_lines.append(add_import) self.number_of_lines = len(self.in_lines) self.out_lines = [] self.comments = {'from': {}, 'straight': {}, 'nested': {}, 'above': {'straight': {}, 'from': {}}} self.imports = OrderedDict() self.as_map = {} section_names = self.config['sections'] self.sections = namedtuple('Sections', section_names)(*[name for name in section_names]) for section in itertools.chain(self.sections, self.config['forced_separate']): self.imports[section] = {'straight': OrderedSet(), 'from': OrderedDict()} self.known_patterns = [] for placement in reversed(self.sections): known_placement = KNOWN_SECTION_MAPPING.get(placement, placement) config_key = 'known_{0}'.format(known_placement.lower()) known_patterns = self.config.get(config_key, []) for known_pattern in known_patterns: self.known_patterns.append((re.compile('^' + known_pattern.replace('*', '.*').replace('?', '.?') + '$'), placement)) self.index = 0 self.import_index = -1 self._first_comment_index_start = -1 self._first_comment_index_end = -1 self._parse() if self.import_index != -1: self._add_formatted_imports() self.length_change = len(self.out_lines) - self.original_length while self.out_lines and self.out_lines[-1].strip() == "": self.out_lines.pop(-1) self.out_lines.append("") self.output = self.line_separator.join(self.out_lines) if self.config['atomic']: try: compile(self._strip_top_comments(self.out_lines, self.line_separator), self.file_path, 'exec', 0, 1) except SyntaxError: self.output = file_contents self.incorrectly_sorted = True try: compile(self._strip_top_comments(self.in_lines, self.line_separator), self.file_path, 'exec', 0, 1) print("ERROR: {0} isort would have introduced syntax errors, please report to the project!". format(self.file_path)) except SyntaxError: print("ERROR: {0} File contains syntax errors.".format(self.file_path)) return if check: check_output = self.output check_against = file_contents if self.config['ignore_whitespace']: check_output = check_output.replace(self.line_separator, "").replace(" ", "") check_against = check_against.replace(self.line_separator, "").replace(" ", "") if check_output == check_against: if self.config['verbose']: print("SUCCESS: {0} Everything Looks Good!".format(self.file_path)) return print("ERROR: {0} Imports are incorrectly sorted.".format(self.file_path)) self.incorrectly_sorted = True if show_diff or self.config['show_diff']: self._show_diff(file_contents) elif write_to_stdout: sys.stdout.write(self.output) elif file_name and not check: if self.output == file_contents: return if ask_to_apply: self._show_diff(file_contents) answer = None while answer not in ('yes', 'y', 'no', 'n', 'quit', 'q'): answer = input("Apply suggested changes to '{0}' [y/n/q]? ".format(self.file_path)).lower() if answer in ('no', 'n'): return if answer in ('quit', 'q'): sys.exit(1) with io.open(self.file_path, encoding=self.file_encoding, mode='w', newline='') as output_file: print("Fixing {0}".format(self.file_path)) output_file.write(self.output) def _show_diff(self, file_contents): for line in unified_diff( file_contents.splitlines(1), self.output.splitlines(1), fromfile=self.file_path + ':before', tofile=self.file_path + ':after', fromfiledate=str(datetime.fromtimestamp(os.path.getmtime(self.file_path)) if self.file_path else datetime.now()), tofiledate=str(datetime.now()) ): sys.stdout.write(line) @staticmethod def _strip_top_comments(lines, line_separator): """Strips # comments that exist at the top of the given lines""" lines = copy.copy(lines) while lines and lines[0].startswith("#"): lines = lines[1:] return line_separator.join(lines) def place_module(self, module_name): """Tries to determine if a module is a python std import, third party import, or project code: if it can't determine - it assumes it is project code """ for forced_separate in self.config['forced_separate']: # Ensure all forced_separate patterns will match to end of string path_glob = forced_separate if not forced_separate.endswith('*'): path_glob = '%s*' % forced_separate if fnmatch(module_name, path_glob) or fnmatch(module_name, '.' + path_glob): return forced_separate if module_name.startswith("."): return self.sections.LOCALFOLDER # Try to find most specific placement instruction match (if any) parts = module_name.split('.') module_names_to_check = ['.'.join(parts[:first_k]) for first_k in range(len(parts), 0, -1)] for module_name_to_check in module_names_to_check: for pattern, placement in self.known_patterns: if pattern.match(module_name_to_check): return placement # Use a copy of sys.path to avoid any unintended modifications # to it - e.g. `+=` used below will change paths in place and # if not copied, consequently sys.path, which will grow unbounded # with duplicates on every call to this method. paths = list(sys.path) virtual_env = self.config.get('virtual_env') or os.environ.get('VIRTUAL_ENV') virtual_env_src = False if virtual_env: paths += [path for path in glob('{0}/lib/python*/site-packages'.format(virtual_env)) if path not in paths] paths += [path for path in glob('{0}/src/*'.format(virtual_env)) if os.path.isdir(path)] virtual_env_src = '{0}/src/'.format(virtual_env) # handle case-insensitive paths on windows stdlib_lib_prefix = os.path.normcase(sysconfig.get_paths()['stdlib']) for prefix in paths: package_path = "/".join((prefix, module_name.split(".")[0])) is_module = (exists_case_sensitive(package_path + ".py") or exists_case_sensitive(package_path + ".so")) is_package = exists_case_sensitive(package_path) and os.path.isdir(package_path) if is_module or is_package: if ('site-packages' in prefix or 'dist-packages' in prefix or (virtual_env and virtual_env_src in prefix)): return self.sections.THIRDPARTY elif os.path.normcase(prefix).startswith(stdlib_lib_prefix): return self.sections.STDLIB else: return self.config['default_section'] return self.config['default_section'] def _get_line(self): """Returns the current line from the file while incrementing the index.""" line = self.in_lines[self.index] self.index += 1 return line @staticmethod def _import_type(line): """If the current line is an import line it will return its type (from or straight)""" if "isort:skip" in line: return elif line.startswith('import '): return "straight" elif line.startswith('from '): return "from" def _at_end(self): """returns True if we are at the end of the file.""" return self.index == self.number_of_lines @staticmethod def _module_key(module_name, config, sub_imports=False, ignore_case=False): prefix = "" if ignore_case: module_name = str(module_name).lower() else: module_name = str(module_name) if sub_imports and config['order_by_type']: if module_name.isupper() and len(module_name) > 1: prefix = "A" elif module_name[0:1].isupper(): prefix = "B" else: prefix = "C" module_name = module_name.lower() return "{0}{1}{2}".format(module_name in config['force_to_top'] and "A" or "B", prefix, config['length_sort'] and (str(len(module_name)) + ":" + module_name) or module_name) def _add_comments(self, comments, original_string=""): """ Returns a string with comments added """ return comments and "{0}{1} {2}".format(self._strip_comments(original_string)[0], self.config['comment_prefix'], "; ".join(comments)) or original_string def _wrap(self, line): """ Returns an import wrapped to the specified line-length, if possible. """ wrap_mode = self.config['multi_line_output'] if len(line) > self.config['line_length'] and wrap_mode != settings.WrapModes.NOQA: line_without_comment = line comment = None if '#' in line: line_without_comment, comment = line.split('#', 1) for splitter in ("import ", ".", "as "): exp = r"\b" + re.escape(splitter) + r"\b" if re.search(exp, line_without_comment) and not line_without_comment.strip().startswith(splitter): line_parts = re.split(exp, line_without_comment) if comment: line_parts[-1] = '{0}#{1}'.format(line_parts[-1], comment) next_line = [] while (len(line) + 2) > (self.config['wrap_length'] or self.config['line_length']) and line_parts: next_line.append(line_parts.pop()) line = splitter.join(line_parts) if not line: line = next_line.pop() cont_line = self._wrap(self.config['indent'] + splitter.join(next_line).lstrip()) if self.config['use_parentheses']: output = "{0}{1}({2}{3}{4}{5})".format( line, splitter, self.line_separator, cont_line, "," if self.config['include_trailing_comma'] else "", self.line_separator if wrap_mode in ( settings.WrapModes.VERTICAL_HANGING_INDENT, settings.WrapModes.VERTICAL_GRID_GROUPED, ) else "") lines = output.split(self.line_separator) if self.config['comment_prefix'] in lines[-1] and lines[-1].endswith(')'): line, comment = lines[-1].split(self.config['comment_prefix'], 1) lines[-1] = line + ')' + self.config['comment_prefix'] + comment[:-1] return self.line_separator.join(lines) return "{0}{1}\\{2}{3}".format(line, splitter, self.line_separator, cont_line) elif len(line) > self.config['line_length'] and wrap_mode == settings.WrapModes.NOQA: if "# NOQA" not in line: return "{0}{1} NOQA".format(line, self.config['comment_prefix']) return line def _add_straight_imports(self, straight_modules, section, section_output): for module in straight_modules: if module in self.remove_imports: continue if module in self.as_map: import_definition = '' if self.config['keep_direct_and_as_imports']: import_definition = "import {0}\n".format(module) import_definition += "import {0} as {1}".format(module, self.as_map[module]) else: import_definition = "import {0}".format(module) comments_above = self.comments['above']['straight'].pop(module, None) if comments_above: section_output.extend(comments_above) section_output.append(self._add_comments(self.comments['straight'].get(module), import_definition)) def _add_from_imports(self, from_modules, section, section_output, ignore_case): for module in from_modules: if module in self.remove_imports: continue import_start = "from {0} import ".format(module) from_imports = list(self.imports[section]['from'][module]) if not self.config['no_inline_sort'] or self.config['force_single_line']: from_imports = nsorted(from_imports, key=lambda key: self._module_key(key, self.config, True, ignore_case)) if self.remove_imports: from_imports = [line for line in from_imports if not "{0}.{1}".format(module, line) in self.remove_imports] sub_modules = ['{0}.{1}'.format(module, from_import) for from_import in from_imports] as_imports = dict((from_import, "{0} as {1}".format(from_import, self.as_map[sub_module])) for from_import, sub_module in zip(from_imports, sub_modules) if sub_module in self.as_map) if self.config['combine_as_imports'] and not ("*" in from_imports and self.config['combine_star']): for from_import in copy.copy(from_imports): if from_import in as_imports: from_imports[from_imports.index(from_import)] = as_imports.pop(from_import) while from_imports: comments = self.comments['from'].pop(module, ()) if "*" in from_imports and self.config['combine_star']: import_statement = self._wrap(self._add_comments(comments, "{0}*".format(import_start))) from_imports = None elif self.config['force_single_line']: import_statements = [] while from_imports: from_import = from_imports.pop(0) if from_import in as_imports: from_comments = self.comments['straight'].get('{}.{}'.format(module, from_import)) import_statements.append(self._add_comments(from_comments, self._wrap(import_start + as_imports[from_import]))) continue single_import_line = self._add_comments(comments, import_start + from_import) comment = self.comments['nested'].get(module, {}).pop(from_import, None) if comment: single_import_line += "{0} {1}".format(comments and ";" or self.config['comment_prefix'], comment) import_statements.append(self._wrap(single_import_line)) comments = None import_statement = self.line_separator.join(import_statements) else: while from_imports and from_imports[0] in as_imports: from_import = from_imports.pop(0) from_comments = self.comments['straight'].get('{}.{}'.format(module, from_import)) section_output.append(self._add_comments(from_comments, self._wrap(import_start + as_imports[from_import]))) star_import = False if "*" in from_imports: section_output.append(self._add_comments(comments, "{0}*".format(import_start))) from_imports.remove('*') star_import = True comments = None for from_import in copy.copy(from_imports): if from_import in as_imports: continue comment = self.comments['nested'].get(module, {}).pop(from_import, None) if comment: single_import_line = self._add_comments(comments, import_start + from_import) single_import_line += "{0} {1}".format(comments and ";" or self.config['comment_prefix'], comment) above_comments = self.comments['above']['from'].pop(module, None) if above_comments: section_output.extend(above_comments) section_output.append(self._wrap(single_import_line)) from_imports.remove(from_import) comments = None from_import_section = [] while from_imports and from_imports[0] not in as_imports: from_import_section.append(from_imports.pop(0)) if star_import: import_statement = import_start + (", ").join(from_import_section) else: import_statement = self._add_comments(comments, import_start + (", ").join(from_import_section)) if not from_import_section: import_statement = "" do_multiline_reformat = False force_grid_wrap = self.config['force_grid_wrap'] if force_grid_wrap and len(from_import_section) >= force_grid_wrap: do_multiline_reformat = True if len(import_statement) > self.config['line_length'] and len(from_import_section) > 1: do_multiline_reformat = True # If line too long AND have imports AND we are NOT using GRID or VERTICAL wrap modes if (len(import_statement) > self.config['line_length'] and len(from_import_section) > 0 and self.config['multi_line_output'] not in (1, 0)): do_multiline_reformat = True if do_multiline_reformat: import_statement = self._multi_line_reformat(import_start, from_import_section, comments) if not do_multiline_reformat and len(import_statement) > self.config['line_length']: import_statement = self._wrap(import_statement) if import_statement: above_comments = self.comments['above']['from'].pop(module, None) if above_comments: section_output.extend(above_comments) section_output.append(import_statement) def _multi_line_reformat(self, import_start, from_imports, comments): output_mode = settings.WrapModes._fields[self.config['multi_line_output']].lower() formatter = getattr(self, "_output_" + output_mode, self._output_grid) dynamic_indent = " " * (len(import_start) + 1) indent = self.config['indent'] line_length = self.config['wrap_length'] or self.config['line_length'] import_statement = formatter(import_start, copy.copy(from_imports), dynamic_indent, indent, line_length, comments) if self.config['balanced_wrapping']: lines = import_statement.split(self.line_separator) line_count = len(lines) if len(lines) > 1: minimum_length = min(len(line) for line in lines[:-1]) else: minimum_length = 0 new_import_statement = import_statement while (len(lines[-1]) < minimum_length and len(lines) == line_count and line_length > 10): import_statement = new_import_statement line_length -= 1 new_import_statement = formatter(import_start, copy.copy(from_imports), dynamic_indent, indent, line_length, comments) lines = new_import_statement.split(self.line_separator) if import_statement.count(self.line_separator) == 0: return self._wrap(import_statement) return import_statement def _add_formatted_imports(self): """Adds the imports back to the file. (at the index of the first import) sorted alphabetically and split between groups """ sort_ignore_case = self.config['force_alphabetical_sort_within_sections'] sections = itertools.chain(self.sections, self.config['forced_separate']) if self.config['no_sections']: self.imports['no_sections'] = {'straight': [], 'from': {}} for section in sections: self.imports['no_sections']['straight'].extend(self.imports[section].get('straight', [])) self.imports['no_sections']['from'].update(self.imports[section].get('from', {})) sections = ('no_sections', ) output = [] prev_section_has_imports = False for section in sections: straight_modules = self.imports[section]['straight'] straight_modules = nsorted(straight_modules, key=lambda key: self._module_key(key, self.config)) from_modules = self.imports[section]['from'] from_modules = nsorted(from_modules, key=lambda key: self._module_key(key, self.config)) section_output = [] if self.config['from_first']: self._add_from_imports(from_modules, section, section_output, sort_ignore_case) if self.config['lines_between_types'] and from_modules and straight_modules: section_output.extend([''] * self.config['lines_between_types']) self._add_straight_imports(straight_modules, section, section_output) else: self._add_straight_imports(straight_modules, section, section_output) if self.config['lines_between_types'] and from_modules and straight_modules: section_output.extend([''] * self.config['lines_between_types']) self._add_from_imports(from_modules, section, section_output, sort_ignore_case) if self.config['force_sort_within_sections']: def by_module(line): section = 'B' if line.startswith('#'): return 'AA' line = re.sub('^from ', '', line) line = re.sub('^import ', '', line) if line.split(' ')[0] in self.config['force_to_top']: section = 'A' if not self.config['order_by_type']: line = line.lower() return '{0}{1}'.format(section, line) section_output = nsorted(section_output, key=by_module) if section_output: section_name = section if section_name in self.place_imports: self.place_imports[section_name] = section_output continue section_title = self.config.get('import_heading_' + str(section_name).lower(), '') if section_title: section_comment = "# {0}".format(section_title) if section_comment not in self.out_lines[0:1] and section_comment not in self.in_lines[0:1]: section_output.insert(0, section_comment) if prev_section_has_imports and section_name in self.config['no_lines_before']: while output and output[-1].strip() == '': output.pop() output += section_output + ([''] * self.config['lines_between_sections']) prev_section_has_imports = bool(section_output) while output and output[-1].strip() == '': output.pop() while output and output[0].strip() == '': output.pop(0) output_at = 0 if self.import_index < self.original_length: output_at = self.import_index elif self._first_comment_index_end != -1 and self._first_comment_index_start <= 2: output_at = self._first_comment_index_end self.out_lines[output_at:0] = output imports_tail = output_at + len(output) while [character.strip() for character in self.out_lines[imports_tail: imports_tail + 1]] == [""]: self.out_lines.pop(imports_tail) if len(self.out_lines) > imports_tail: next_construct = "" self._in_quote = False tail = self.out_lines[imports_tail:] for index, line in enumerate(tail): in_quote = self._in_quote if not self._skip_line(line) and line.strip(): if line.strip().startswith("#") and len(tail) > (index + 1) and tail[index + 1].strip(): continue next_construct = line break elif not in_quote: parts = line.split() if len(parts) >= 3 and parts[1] == '=' and "'" not in parts[0] and '"' not in parts[0]: next_construct = line break if self.config['lines_after_imports'] != -1: self.out_lines[imports_tail:0] = ["" for line in range(self.config['lines_after_imports'])] elif next_construct.startswith("def ") or next_construct.startswith("class ") or \ next_construct.startswith("@") or next_construct.startswith("async def"): self.out_lines[imports_tail:0] = ["", ""] else: self.out_lines[imports_tail:0] = [""] if self.place_imports: new_out_lines = [] for index, line in enumerate(self.out_lines): new_out_lines.append(line) if line in self.import_placements: new_out_lines.extend(self.place_imports[self.import_placements[line]]) if len(self.out_lines) <= index or self.out_lines[index + 1].strip() != "": new_out_lines.append("") self.out_lines = new_out_lines def _output_grid(self, statement, imports, white_space, indent, line_length, comments): statement += "(" + imports.pop(0) while imports: next_import = imports.pop(0) next_statement = self._add_comments(comments, statement + ", " + next_import) if len(next_statement.split(self.line_separator)[-1]) + 1 > line_length: lines = ['{0}{1}'.format(white_space, next_import.split(" ")[0])] for part in next_import.split(" ")[1:]: new_line = '{0} {1}'.format(lines[-1], part) if len(new_line) + 1 > line_length: lines.append('{0}{1}'.format(white_space, part)) else: lines[-1] = new_line next_import = self.line_separator.join(lines) statement = (self._add_comments(comments, "{0},".format(statement)) + "{0}{1}".format(self.line_separator, next_import)) comments = None else: statement += ", " + next_import return statement + ("," if self.config['include_trailing_comma'] else "") + ")" def _output_vertical(self, statement, imports, white_space, indent, line_length, comments): first_import = self._add_comments(comments, imports.pop(0) + ",") + self.line_separator + white_space return "{0}({1}{2}{3})".format( statement, first_import, ("," + self.line_separator + white_space).join(imports), "," if self.config['include_trailing_comma'] else "", ) def _output_hanging_indent(self, statement, imports, white_space, indent, line_length, comments): statement += imports.pop(0) while imports: next_import = imports.pop(0) next_statement = self._add_comments(comments, statement + ", " + next_import) if len(next_statement.split(self.line_separator)[-1]) + 3 > line_length: next_statement = (self._add_comments(comments, "{0}, \\".format(statement)) + "{0}{1}{2}".format(self.line_separator, indent, next_import)) comments = None statement = next_statement return statement def _output_vertical_hanging_indent(self, statement, imports, white_space, indent, line_length, comments): return "{0}({1}{2}{3}{4}{5}{2})".format( statement, self._add_comments(comments), self.line_separator, indent, ("," + self.line_separator + indent).join(imports), "," if self.config['include_trailing_comma'] else "", ) def _output_vertical_grid_common(self, statement, imports, white_space, indent, line_length, comments, need_trailing_char): statement += self._add_comments(comments, "(") + self.line_separator + indent + imports.pop(0) while imports: next_import = imports.pop(0) next_statement = "{0}, {1}".format(statement, next_import) current_line_length = len(next_statement.split(self.line_separator)[-1]) if imports or need_trailing_char: # If we have more imports we need to account for a comma after this import # We might also need to account for a closing ) we're going to add. current_line_length += 1 if current_line_length > line_length: next_statement = "{0},{1}{2}{3}".format(statement, self.line_separator, indent, next_import) statement = next_statement if self.config['include_trailing_comma']: statement += ',' return statement def _output_vertical_grid(self, statement, imports, white_space, indent, line_length, comments): return self._output_vertical_grid_common(statement, imports, white_space, indent, line_length, comments, True) + ")" def _output_vertical_grid_grouped(self, statement, imports, white_space, indent, line_length, comments): return self._output_vertical_grid_common(statement, imports, white_space, indent, line_length, comments, True) + self.line_separator + ")" def _output_vertical_grid_grouped_no_comma(self, statement, imports, white_space, indent, line_length, comments): return self._output_vertical_grid_common(statement, imports, white_space, indent, line_length, comments, False) + self.line_separator + ")" def _output_noqa(self, statement, imports, white_space, indent, line_length, comments): retval = '{0}{1}'.format(statement, ', '.join(imports)) comment_str = ' '.join(comments) if comments: if len(retval) + len(self.config['comment_prefix']) + 1 + len(comment_str) <= line_length: return '{0}{1} {2}'.format(retval, self.config['comment_prefix'], comment_str) else: if len(retval) <= line_length: return retval if comments: if "NOQA" in comments: return '{0}{1} {2}'.format(retval, self.config['comment_prefix'], comment_str) else: return '{0}{1} NOQA {2}'.format(retval, self.config['comment_prefix'], comment_str) else: return '{0}{1} NOQA'.format(retval, self.config['comment_prefix']) @staticmethod def _strip_comments(line, comments=None): """Removes comments from import line.""" if comments is None: comments = [] new_comments = False comment_start = line.find("#") if comment_start != -1: comments.append(line[comment_start + 1:].strip()) new_comments = True line = line[:comment_start] return line, comments, new_comments @staticmethod def _format_simplified(import_line): import_line = import_line.strip() if import_line.startswith("from "): import_line = import_line.replace("from ", "") import_line = import_line.replace(" import ", ".") elif import_line.startswith("import "): import_line = import_line.replace("import ", "") return import_line @staticmethod def _format_natural(import_line): import_line = import_line.strip() if not import_line.startswith("from ") and not import_line.startswith("import "): if "." not in import_line: return "import {0}".format(import_line) parts = import_line.split(".") end = parts.pop(-1) return "from {0} import {1}".format(".".join(parts), end) return import_line def _skip_line(self, line): skip_line = self._in_quote if self.index == 1 and line.startswith("#"): self._in_top_comment = True return True elif self._in_top_comment: if not line.startswith("#"): self._in_top_comment = False self._first_comment_index_end = self.index - 1 if '"' in line or "'" in line: index = 0 if self._first_comment_index_start == -1 and (line.startswith('"') or line.startswith("'")): self._first_comment_index_start = self.index while index < len(line): if line[index] == "\\": index += 1 elif self._in_quote: if line[index:index + len(self._in_quote)] == self._in_quote: self._in_quote = False if self._first_comment_index_end < self._first_comment_index_start: self._first_comment_index_end = self.index elif line[index] in ("'", '"'): long_quote = line[index:index + 3] if long_quote in ('"""', "'''"): self._in_quote = long_quote index += 2 else: self._in_quote = line[index] elif line[index] == "#": break index += 1 return skip_line or self._in_quote or self._in_top_comment def _strip_syntax(self, import_string): import_string = import_string.replace("_import", "[[i]]") for remove_syntax in ['\\', '(', ')', ',']: import_string = import_string.replace(remove_syntax, " ") import_list = import_string.split() for key in ('from', 'import'): if key in import_list: import_list.remove(key) import_string = ' '.join(import_list) import_string = import_string.replace("[[i]]", "_import") return import_string.replace("{ ", "{|").replace(" }", "|}") def _parse(self): """Parses a python file taking out and categorizing imports.""" self._in_quote = False self._in_top_comment = False while not self._at_end(): raw_line = line = self._get_line() line = line.replace("from.import ", "from . import ") line = line.replace("\t", " ").replace('import*', 'import *') line = line.replace(" .import ", " . import ") statement_index = self.index skip_line = self._skip_line(line) if line in self._section_comments and not skip_line: if self.import_index == -1: self.import_index = self.index - 1 continue if "isort:imports-" in line and line.startswith("#"): section = line.split("isort:imports-")[-1].split()[0].upper() self.place_imports[section] = [] self.import_placements[line] = section if ";" in line: for part in (part.strip() for part in line.split(";")): if part and not part.startswith("from ") and not part.startswith("import "): skip_line = True import_type = self._import_type(line) if not import_type or skip_line: self.out_lines.append(raw_line) continue for line in (line.strip() for line in line.split(";")): import_type = self._import_type(line) if not import_type: self.out_lines.append(line) continue if self.import_index == -1: self.import_index = self.index - 1 nested_comments = {} import_string, comments, new_comments = self._strip_comments(line) stripped_line = [part for part in self._strip_syntax(import_string).strip().split(" ") if part] if import_type == "from" and len(stripped_line) == 2 and stripped_line[1] != "*" and new_comments: nested_comments[stripped_line[-1]] = comments[0] if "(" in line.split("#")[0] and not self._at_end(): while not line.strip().endswith(")") and not self._at_end(): line, comments, new_comments = self._strip_comments(self._get_line(), comments) stripped_line = self._strip_syntax(line).strip() if import_type == "from" and stripped_line and " " not in stripped_line and new_comments: nested_comments[stripped_line] = comments[-1] import_string += self.line_separator + line else: while line.strip().endswith("\\"): line, comments, new_comments = self._strip_comments(self._get_line(), comments) # Still need to check for parentheses after an escaped line if "(" in line.split("#")[0] and not self._at_end(): stripped_line = self._strip_syntax(line).strip() if import_type == "from" and stripped_line and " " not in stripped_line and new_comments: nested_comments[stripped_line] = comments[-1] import_string += self.line_separator + line while not line.strip().endswith(")") and not self._at_end(): line, comments, new_comments = self._strip_comments(self._get_line(), comments) stripped_line = self._strip_syntax(line).strip() if import_type == "from" and stripped_line and " " not in stripped_line and new_comments: nested_comments[stripped_line] = comments[-1] import_string += self.line_separator + line stripped_line = self._strip_syntax(line).strip() if import_type == "from" and stripped_line and " " not in stripped_line and new_comments: nested_comments[stripped_line] = comments[-1] if import_string.strip().endswith(" import") or line.strip().startswith("import "): import_string += self.line_separator + line else: import_string = import_string.rstrip().rstrip("\\") + " " + line.lstrip() if import_type == "from": import_string = import_string.replace("import(", "import (") parts = import_string.split(" import ") from_import = parts[0].split(" ") import_string = " import ".join([from_import[0] + " " + "".join(from_import[1:])] + parts[1:]) imports = [item.replace("{|", "{ ").replace("|}", " }") for item in self._strip_syntax(import_string).split()] if "as" in imports and (imports.index('as') + 1) < len(imports): while "as" in imports: index = imports.index('as') if import_type == "from": module = imports[0] + "." + imports[index - 1] self.as_map[module] = imports[index + 1] else: module = imports[index - 1] self.as_map[module] = imports[index + 1] if not self.config['combine_as_imports']: self.comments['straight'][module] = comments comments = [] del imports[index:index + 2] if import_type == "from": import_from = imports.pop(0) placed_module = self.place_module(import_from) if self.config['verbose']: print("from-type place_module for %s returned %s" % (import_from, placed_module)) if placed_module == '': print( "WARNING: could not place module {0} of line {1} --" " Do you need to define a default section?".format(import_from, line) ) root = self.imports[placed_module][import_type] for import_name in imports: associated_comment = nested_comments.get(import_name) if associated_comment: self.comments['nested'].setdefault(import_from, {})[import_name] = associated_comment comments.pop(comments.index(associated_comment)) if comments: self.comments['from'].setdefault(import_from, []).extend(comments) if len(self.out_lines) > max(self.import_index, self._first_comment_index_end + 1, 1) - 1: last = self.out_lines and self.out_lines[-1].rstrip() or "" while (last.startswith("#") and not last.endswith('"""') and not last.endswith("'''") and 'isort:imports-' not in last): self.comments['above']['from'].setdefault(import_from, []).insert(0, self.out_lines.pop(-1)) if len(self.out_lines) > max(self.import_index - 1, self._first_comment_index_end + 1, 1) - 1: last = self.out_lines[-1].rstrip() else: last = "" if statement_index - 1 == self.import_index: self.import_index -= len(self.comments['above']['from'].get(import_from, [])) if root.get(import_from, False): root[import_from].update(imports) else: root[import_from] = OrderedSet(imports) else: for module in imports: if comments: self.comments['straight'][module] = comments comments = None if len(self.out_lines) > max(self.import_index, self._first_comment_index_end + 1, 1) - 1: last = self.out_lines and self.out_lines[-1].rstrip() or "" while (last.startswith("#") and not last.endswith('"""') and not last.endswith("'''") and 'isort:imports-' not in last): self.comments['above']['straight'].setdefault(module, []).insert(0, self.out_lines.pop(-1)) if len(self.out_lines) > 0: last = self.out_lines[-1].rstrip() else: last = "" if self.index - 1 == self.import_index: self.import_index -= len(self.comments['above']['straight'].get(module, [])) placed_module = self.place_module(module) if self.config['verbose']: print("else-type place_module for %s returned %s" % (module, placed_module)) if placed_module == '': print( "WARNING: could not place module {0} of line {1} --" " Do you need to define a default section?".format(import_from, line) ) self.imports[placed_module][import_type].add(module) def coding_check(fname, default='utf-8'): # see https://www.python.org/dev/peps/pep-0263/ pattern = re.compile(br'coding[:=]\s*([-\w.]+)') coding = default with io.open(fname, 'rb') as f: for line_number, line in enumerate(f, 1): groups = re.findall(pattern, line) if groups: coding = groups[0].decode('ascii') break if line_number > 2: break return coding def exists_case_sensitive(path): """ Returns if the given path exists and also matches the case on Windows. When finding files that can be imported, it is important for the cases to match because while file os.path.exists("module.py") and os.path.exists("MODULE.py") both return True on Windows, Python can only import using the case of the real file. """ result = os.path.exists(path) if (sys.platform.startswith('win') or sys.platform == 'darwin') and result: directory, basename = os.path.split(path) result = basename in os.listdir(directory) return result