1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798 |
- from django.contrib.auth.models import User
- from django.core.management.base import BaseCommand
- import json
- from pygal.models import Item, Tag
- import os
-
- from ._private import KEY_USERDATA_VERSION
- from ._private import KEY_USERDATA_FAVOURITE
- from ._private import KEY_USERDATA_TAGS
-
-
- class Command(BaseCommand):
- help = 'Import userdata from a JSON-File.'
-
- def add_arguments(self, parser):
- parser.add_argument('filename(s)', nargs='+', type=str)
-
- def error_skipped_file(self, path, reason):
- self.stderr.write(' Skipped file %s caused by an error (%s) !' % (path, reason))
-
- def handle(self, *args, **options):
- for fn in options['filename(s)']:
- fn = os.path.abspath(fn)
- self.stdout.write('Parsing %s:' % fn)
- try:
- with open(fn, 'r') as fh:
- data = json.load(fh)
- try:
- version = data.pop(KEY_USERDATA_VERSION)
- except KeyError:
- version = 1
- except PermissionError:
- self.error_skipped_file(fn, 'file permission')
- except json.decoder.JSONDecodeError:
- self.error_skipped_file(fn, 'json decoding')
- except FileNotFoundError:
- self.error_skipped_file(fn, 'file not found')
- else:
- for section in data:
- self.stdout.write(' Importing "%s":' % section)
- cnt_all_datasets = 0
- cnt_success = 0
- cnt_already_exist = 0
- cnt_user_missing = 0
- cnt_item_missing = 0
- if section == KEY_USERDATA_FAVOURITE:
- for username in data[section]:
- try:
- user = User.objects.get(username=username)
- except User.DoesNotExist:
- user = None
- for rel_path in data[section][username]:
- cnt_all_datasets += 1
- if user is None:
- cnt_user_missing += 1
- try:
- i = Item.objects.get(rel_path=rel_path)
- except Item.DoesNotExist:
- cnt_item_missing += 1
- else:
- if user in i.favourite_of.all():
- cnt_already_exist += 1
- else:
- cnt_success += 1
- i.favourite_of.add(user)
- elif section == KEY_USERDATA_TAGS:
- for rel_path in data[section]:
- cnt_all_datasets += len(data[section][rel_path])
- try:
- i = Item.objects.get(rel_path=rel_path)
- except Item.DoesNotExist:
- cnt_item_missing += len(data[section][rel_path])
- else:
- for tag_data in data[section][rel_path]:
- if len(Tag.objects.filter(item=i, text=tag_data['text'])) > 0:
- cnt_already_exist += 1
- else:
- tag_data['item'] = i
- Tag(**tag_data).save()
- cnt_success += 1
- else:
- self.stdout.write(self.style.ERROR(' Section unknown!'))
- #
- # REPORT
- #
- if cnt_all_datasets > 0:
- if (cnt_success + cnt_already_exist) < cnt_all_datasets:
- style = self.style.WARNING
- self.stdout.write(style(' NOT all data imported.'))
- else:
- style = self.style.SUCCESS
- self.stdout.write(style(' All data imported.'))
- self.stdout.write(style(' %5d %s(s) successfully added.' % (cnt_success, section)))
- self.stdout.write(style(' %5d %s(s) already set correctly.' % (cnt_already_exist, section)))
- if cnt_user_missing > 0:
- self.stdout.write(self.style.ERROR(' %5d %s(s) not added, because the target user does not exist.' % (cnt_user_missing, section)))
- if cnt_item_missing > 0:
- self.stdout.write(self.style.ERROR(' %5d %s(s) not added, because the target item does not exist.' % (cnt_item_missing, section)))
|