Files
hudozka/utils/Sync/sync/schedule.py
T

98 lines
2.2 KiB
Python

from toml import loads as toml_str
from yaml import load as yaml
from sync import untouched
from sync.core.schedule import SyncSchedule
from sync.data import get_data
from utils.fn import lmap, constant
dir_documents = '/Users/tmshv/Dropbox/Dev/Hud school/Schedules'
def sync_schedules(provider, collection, update=True, delete=True):
"""
:param collection:
:param provider:
:param update:
:param delete:
:return:
"""
return main(
SyncSchedule(
collection,
provider
),
update_documents=update,
delete_documents=delete
)
def combine(fns):
def call(value):
data = value.read()
for f in fns:
try:
return f(data)
except Exception as e:
pass
return None
return call
def toml(bytes_value):
s = bytes_value.decode('utf-8')
return toml_str(s)
def main(sync, update_documents=False, delete_documents=False):
# COLLECT YAML/MD/JPG/PDF MANIFEST FILES
documents = sync.provider.type_filter('.', '.yaml')
documents += sync.provider.type_filter('.', '.toml')
# READ MANIFEST
documents = lmap(
lambda i: {
'file': i,
**get_data(sync.provider, i, combine([yaml, toml, constant({})]))
},
documents
)
# CREATE DOCUMENT IDENTITY
documents = lmap(sync.create_id, documents)
# CREATE DOCUMENT HASH
documents = lmap(sync.create_hash, documents)
# CREATE SCOPE OF CURRENT SESSION
# ALL DOCUMENTS IDS FOUNDED
scope_documents_ids = lmap(
lambda i: i['id'],
documents
)
# SKIP UNTOUCHED DOCUMENTS
documents = untouched(documents, sync)
# DO HEAVY PROCESS WITH DOCUMENTS
# SYNC SCHEDULE
if update_documents:
documents = lmap(
sync.update,
documents
)
documents_to_remove = sync.query({'id': {'$nin': scope_documents_ids}})
if delete_documents:
documents_to_remove = lmap(
sync.delete,
map(
lambda i: {'_id': i['_id']},
documents_to_remove
)
)
return documents, documents_to_remove