Hacked By AnonymousFox

Current Path : /proc/thread-self/root/proc/thread-self/root/proc/self/root/opt/alt/python34/lib64/python3.4/ensurepip/rewheel/
Upload File :
Current File : //proc/thread-self/root/proc/thread-self/root/proc/self/root/opt/alt/python34/lib64/python3.4/ensurepip/rewheel/__init__.py

import argparse
import codecs
import csv
import email.parser
import os
import io
import re
import site
import subprocess
import sys
import zipfile

def run():
    parser = argparse.ArgumentParser(description='Recreate wheel of package with given RECORD.')
    parser.add_argument('record_path',
                        help='Path to RECORD file')
    parser.add_argument('-o', '--output-dir',
                        help='Dir where to place the wheel, defaults to current working dir.',
                        dest='outdir',
                        default=os.path.curdir)

    ns = parser.parse_args()
    retcode = 0
    try:
        print(rewheel_from_record(**vars(ns)))
    except BaseException as e:
        print('Failed: {}'.format(e))
        retcode = 1
    sys.exit(1)

def find_system_records(projects):
    """Return list of paths to RECORD files for system-installed projects.

    If a project is not installed, the resulting list contains None instead
    of a path to its RECORD
    """
    records = []
    # get system site-packages dirs
    sys_sitepack = site.getsitepackages([sys.base_prefix, sys.base_exec_prefix])
    sys_sitepack = [sp for sp in sys_sitepack if os.path.exists(sp)]
    # try to find all projects in all system site-packages
    for project in projects:
        path = None
        for sp in sys_sitepack:
            dist_info_re = os.path.join(sp, project) + '-[^\{0}]+\.dist-info'.format(os.sep)
            candidates = [os.path.join(sp, p) for p in os.listdir(sp)]
            # filter out candidate dirs based on the above regexp
            filtered = [c for c in candidates if re.match(dist_info_re, c)]
            # if we have 0 or 2 or more dirs, something is wrong...
            if len(filtered) == 1:
                path = filtered[0]
        if path is not None:
            records.append(os.path.join(path, 'RECORD'))
        else:
            records.append(None)
    return records

def rewheel_from_record(record_path, outdir):
    """Recreates a whee of package with given record_path and returns path
    to the newly created wheel."""
    site_dir = os.path.dirname(os.path.dirname(record_path))
    record_relpath = record_path[len(site_dir):].strip(os.path.sep)
    to_write, to_omit = get_records_to_pack(site_dir, record_relpath)
    new_wheel_name = get_wheel_name(record_path)
    new_wheel_path = os.path.join(outdir, new_wheel_name + '.whl')

    new_wheel = zipfile.ZipFile(new_wheel_path, mode='w', compression=zipfile.ZIP_DEFLATED)
    # we need to write a new record with just the files that we will write,
    # e.g. not binaries and *.pyc/*.pyo files
    new_record = io.StringIO()
    writer = csv.writer(new_record)

    # handle files that we can write straight away
    for f, sha_hash, size in to_write:
        new_wheel.write(os.path.join(site_dir, f), arcname=f)
        writer.writerow([f, sha_hash,size])

    # rewrite the old wheel file with a new computed one
    writer.writerow([record_relpath, '', ''])
    new_wheel.writestr(record_relpath, new_record.getvalue())

    new_wheel.close()

    return new_wheel.filename

def get_wheel_name(record_path):
    """Return proper name of the wheel, without .whl."""

    wheel_info_path = os.path.join(os.path.dirname(record_path), 'WHEEL')
    with codecs.open(wheel_info_path, encoding='utf-8') as wheel_info_file:
        wheel_info = email.parser.Parser().parsestr(wheel_info_file.read())

    metadata_path = os.path.join(os.path.dirname(record_path), 'METADATA')
    with codecs.open(metadata_path, encoding='utf-8') as metadata_file:
        metadata = email.parser.Parser().parsestr(metadata_file.read())

    # construct name parts according to wheel spec
    distribution = metadata.get('Name')
    version = metadata.get('Version')
    build_tag = '' # nothing for now
    lang_tag = []
    for t in wheel_info.get_all('Tag'):
        lang_tag.append(t.split('-')[0])
    lang_tag = '.'.join(lang_tag)
    abi_tag, plat_tag = wheel_info.get('Tag').split('-')[1:3]
    # leave out build tag, if it is empty
    to_join = filter(None, [distribution, version, build_tag, lang_tag, abi_tag, plat_tag])
    return '-'.join(list(to_join))

def get_records_to_pack(site_dir, record_relpath):
    """Accepts path of sitedir and path of RECORD file relative to it.
    Returns two lists:
    - list of files that can be written to new RECORD straight away
    - list of files that shouldn't be written or need some processing
      (pyc and pyo files, scripts)
    """
    record_file_path = os.path.join(site_dir, record_relpath)
    with codecs.open(record_file_path, encoding='utf-8') as record_file:
        record_contents = record_file.read()
    # temporary fix for https://github.com/pypa/pip/issues/1376
    # we need to ignore files under ".data" directory
    data_dir = os.path.dirname(record_relpath).strip(os.path.sep)
    data_dir = data_dir[:-len('dist-info')] + 'data'

    to_write = []
    to_omit = []
    for l in record_contents.splitlines():
        spl = l.split(',')
        if len(spl) == 3:
            # new record will omit (or write differently):
            # - abs paths, paths with ".." (entry points),
            # - pyc+pyo files
            # - the old RECORD file
            # TODO: is there any better way to recognize an entry point?
            if os.path.isabs(spl[0]) or spl[0].startswith('..') or \
               spl[0].endswith('.pyc') or spl[0].endswith('.pyo') or \
               spl[0] == record_relpath or spl[0].startswith(data_dir):
                to_omit.append(spl)
            else:
                to_write.append(spl)
        else:
            pass # bad RECORD or empty line
    return to_write, to_omit

Hacked By AnonymousFox1.0, Coded By AnonymousFox