%PDF- %PDF-
Direktori : /proc/thread-self/root/opt/alt/python37/lib/python3.7/site-packages/libfuturize/ |
Current File : //proc/thread-self/root/opt/alt/python37/lib/python3.7/site-packages/libfuturize/main.py |
""" futurize: automatic conversion to clean 2/3 code using ``python-future`` ====================================================================== Like Armin Ronacher's modernize.py, ``futurize`` attempts to produce clean standard Python 3 code that runs on both Py2 and Py3. One pass -------- Use it like this on Python 2 code: $ futurize --verbose mypython2script.py This will attempt to port the code to standard Py3 code that also provides Py2 compatibility with the help of the right imports from ``future``. To write changes to the files, use the -w flag. Two stages ---------- The ``futurize`` script can also be called in two separate stages. First: $ futurize --stage1 mypython2script.py This produces more modern Python 2 code that is not yet compatible with Python 3. The tests should still run and the diff should be uncontroversial to apply to most Python projects that are willing to drop support for Python 2.5 and lower. After this, the recommended approach is to explicitly mark all strings that must be byte-strings with a b'' prefix and all text (unicode) strings with a u'' prefix, and then invoke the second stage of Python 2 to 2/3 conversion with:: $ futurize --stage2 mypython2script.py Stage 2 adds a dependency on ``future``. It converts most remaining Python 2-specific code to Python 3 code and adds appropriate imports from ``future`` to restore Py2 support. The command above leaves all unadorned string literals as native strings (byte-strings on Py2, unicode strings on Py3). If instead you would like all unadorned string literals to be promoted to unicode, you can also pass this flag: $ futurize --stage2 --unicode-literals mypython2script.py This adds the declaration ``from __future__ import unicode_literals`` to the top of each file, which implicitly declares all unadorned string literals to be unicode strings (``unicode`` on Py2). All imports ----------- The --all-imports option forces adding all ``__future__`` imports, ``builtins`` imports, and standard library aliases, even if they don't seem necessary for the current state of each module. (This can simplify testing, and can reduce the need to think about Py2 compatibility when editing the code further.) """ from __future__ import (absolute_import, print_function, unicode_literals) import future.utils from future import __version__ import sys import logging import optparse import os from lib2to3.main import warn, StdoutRefactoringTool from lib2to3 import refactor from libfuturize.fixes import (lib2to3_fix_names_stage1, lib2to3_fix_names_stage2, libfuturize_fix_names_stage1, libfuturize_fix_names_stage2) fixer_pkg = 'libfuturize.fixes' def main(args=None): """Main program. Args: fixer_pkg: the name of a package where the fixers are located. args: optional; a list of command line arguments. If omitted, sys.argv[1:] is used. Returns a suggested exit status (0, 1, 2). """ # Set up option parser parser = optparse.OptionParser(usage="futurize [options] file|dir ...") parser.add_option("-V", "--version", action="store_true", help="Report the version number of futurize") parser.add_option("-a", "--all-imports", action="store_true", help="Add all __future__ and future imports to each module") parser.add_option("-1", "--stage1", action="store_true", help="Modernize Python 2 code only; no compatibility with Python 3 (or dependency on ``future``)") parser.add_option("-2", "--stage2", action="store_true", help="Take modernized (stage1) code and add a dependency on ``future`` to provide Py3 compatibility.") parser.add_option("-0", "--both-stages", action="store_true", help="Apply both stages 1 and 2") parser.add_option("-u", "--unicode-literals", action="store_true", help="Add ``from __future__ import unicode_literals`` to implicitly convert all unadorned string literals '' into unicode strings") parser.add_option("-f", "--fix", action="append", default=[], help="Each FIX specifies a transformation; default: all.\nEither use '-f division -f metaclass' etc. or use the fully-qualified module name: '-f lib2to3.fixes.fix_types -f libfuturize.fixes.fix_unicode_keep_u'") parser.add_option("-j", "--processes", action="store", default=1, type="int", help="Run 2to3 concurrently") parser.add_option("-x", "--nofix", action="append", default=[], help="Prevent a fixer from being run.") parser.add_option("-l", "--list-fixes", action="store_true", help="List available transformations") parser.add_option("-p", "--print-function", action="store_true", help="Modify the grammar so that print() is a function") parser.add_option("-v", "--verbose", action="store_true", help="More verbose logging") parser.add_option("--no-diffs", action="store_true", help="Don't show diffs of the refactoring") parser.add_option("-w", "--write", action="store_true", help="Write back modified files") parser.add_option("-n", "--nobackups", action="store_true", default=False, help="Don't write backups for modified files.") parser.add_option("-o", "--output-dir", action="store", type="str", default="", help="Put output files in this directory " "instead of overwriting the input files. Requires -n. " "For Python >= 2.7 only.") parser.add_option("-W", "--write-unchanged-files", action="store_true", help="Also write files even if no changes were required" " (useful with --output-dir); implies -w.") parser.add_option("--add-suffix", action="store", type="str", default="", help="Append this string to all output filenames." " Requires -n if non-empty. For Python >= 2.7 only." "ex: --add-suffix='3' will generate .py3 files.") # Parse command line arguments flags = {} refactor_stdin = False options, args = parser.parse_args(args) if options.write_unchanged_files: flags["write_unchanged_files"] = True if not options.write: warn("--write-unchanged-files/-W implies -w.") options.write = True # If we allowed these, the original files would be renamed to backup names # but not replaced. if options.output_dir and not options.nobackups: parser.error("Can't use --output-dir/-o without -n.") if options.add_suffix and not options.nobackups: parser.error("Can't use --add-suffix without -n.") if not options.write and options.no_diffs: warn("not writing files and not printing diffs; that's not very useful") if not options.write and options.nobackups: parser.error("Can't use -n without -w") if "-" in args: refactor_stdin = True if options.write: print("Can't write to stdin.", file=sys.stderr) return 2 # Is this ever necessary? if options.print_function: flags["print_function"] = True # Set up logging handler level = logging.DEBUG if options.verbose else logging.INFO logging.basicConfig(format='%(name)s: %(message)s', level=level) logger = logging.getLogger('libfuturize.main') if options.stage1 or options.stage2: assert options.both_stages is None options.both_stages = False else: options.both_stages = True avail_fixes = set() if options.stage1 or options.both_stages: avail_fixes.update(lib2to3_fix_names_stage1) avail_fixes.update(libfuturize_fix_names_stage1) if options.stage2 or options.both_stages: avail_fixes.update(lib2to3_fix_names_stage2) avail_fixes.update(libfuturize_fix_names_stage2) if options.unicode_literals: avail_fixes.add('libfuturize.fixes.fix_unicode_literals_import') if options.version: print(__version__) return 0 if options.list_fixes: print("Available transformations for the -f/--fix option:") # for fixname in sorted(refactor.get_all_fix_names(fixer_pkg)): for fixname in sorted(avail_fixes): print(fixname) if not args: return 0 if not args: print("At least one file or directory argument required.", file=sys.stderr) print("Use --help to show usage.", file=sys.stderr) return 2 unwanted_fixes = set() for fix in options.nofix: if ".fix_" in fix: unwanted_fixes.add(fix) else: # Infer the full module name for the fixer. # First ensure that no names clash (e.g. # lib2to3.fixes.fix_blah and libfuturize.fixes.fix_blah): found = [f for f in avail_fixes if f.endswith('fix_{0}'.format(fix))] if len(found) > 1: print("Ambiguous fixer name. Choose a fully qualified " "module name instead from these:\n" + "\n".join(" " + myf for myf in found), file=sys.stderr) return 2 elif len(found) == 0: print("Unknown fixer. Use --list-fixes or -l for a list.", file=sys.stderr) return 2 unwanted_fixes.add(found[0]) extra_fixes = set() if options.all_imports: if options.stage1: prefix = 'libfuturize.fixes.' extra_fixes.add(prefix + 'fix_add__future__imports_except_unicode_literals') else: # In case the user hasn't run stage1 for some reason: prefix = 'libpasteurize.fixes.' extra_fixes.add(prefix + 'fix_add_all__future__imports') extra_fixes.add(prefix + 'fix_add_future_standard_library_import') extra_fixes.add(prefix + 'fix_add_all_future_builtins') explicit = set() if options.fix: all_present = False for fix in options.fix: if fix == 'all': all_present = True else: if ".fix_" in fix: explicit.add(fix) else: # Infer the full module name for the fixer. # First ensure that no names clash (e.g. # lib2to3.fixes.fix_blah and libfuturize.fixes.fix_blah): found = [f for f in avail_fixes if f.endswith('fix_{0}'.format(fix))] if len(found) > 1: print("Ambiguous fixer name. Choose a fully qualified " "module name instead from these:\n" + "\n".join(" " + myf for myf in found), file=sys.stderr) return 2 elif len(found) == 0: print("Unknown fixer. Use --list-fixes or -l for a list.", file=sys.stderr) return 2 explicit.add(found[0]) if len(explicit & unwanted_fixes) > 0: print("Conflicting usage: the following fixers have been " "simultaneously requested and disallowed:\n" + "\n".join(" " + myf for myf in (explicit & unwanted_fixes)), file=sys.stderr) return 2 requested = avail_fixes.union(explicit) if all_present else explicit else: requested = avail_fixes.union(explicit) fixer_names = (requested | extra_fixes) - unwanted_fixes input_base_dir = os.path.commonprefix(args) if (input_base_dir and not input_base_dir.endswith(os.sep) and not os.path.isdir(input_base_dir)): # One or more similar names were passed, their directory is the base. # os.path.commonprefix() is ignorant of path elements, this corrects # for that weird API. input_base_dir = os.path.dirname(input_base_dir) if options.output_dir: input_base_dir = input_base_dir.rstrip(os.sep) logger.info('Output in %r will mirror the input directory %r layout.', options.output_dir, input_base_dir) # Initialize the refactoring tool if future.utils.PY26: extra_kwargs = {} else: extra_kwargs = { 'append_suffix': options.add_suffix, 'output_dir': options.output_dir, 'input_base_dir': input_base_dir, } rt = StdoutRefactoringTool( sorted(fixer_names), flags, sorted(explicit), options.nobackups, not options.no_diffs, **extra_kwargs) # Refactor all files and directories passed as arguments if not rt.errors: if refactor_stdin: rt.refactor_stdin() else: try: rt.refactor(args, options.write, None, options.processes) except refactor.MultiprocessingUnsupported: assert options.processes > 1 print("Sorry, -j isn't " \ "supported on this platform.", file=sys.stderr) return 1 rt.summarize() # Return error status (0 if rt.errors is zero) return int(bool(rt.errors))