337 lines
19 KiB
Python
337 lines
19 KiB
Python
import csv
|
|
from datetime import datetime, timedelta
|
|
from io import BytesIO
|
|
from zipfile import ZipFile
|
|
|
|
import requests
|
|
from django.core.management import BaseCommand
|
|
|
|
from sncfgtfs.models import Agency, Calendar, CalendarDate, FeedInfo, Route, Stop, StopTime, Transfer, Trip
|
|
|
|
|
|
class Command(BaseCommand):
|
|
help = "Update the SNCF GTFS database."
|
|
|
|
GTFS_FEEDS = {
|
|
"TGV": "https://eu.ftp.opendatasoft.com/sncf/gtfs/export_gtfs_voyages.zip",
|
|
"IC": "https://eu.ftp.opendatasoft.com/sncf/gtfs/export-intercites-gtfs-last.zip",
|
|
"TER": "https://eu.ftp.opendatasoft.com/sncf/gtfs/export-ter-gtfs-last.zip",
|
|
"TN": "https://eu.ftp.opendatasoft.com/sncf/gtfs/transilien-gtfs.zip",
|
|
}
|
|
|
|
def add_arguments(self, parser):
|
|
parser.add_argument('--bulk_size', type=int, default=1000, help='Number of objects to create in bulk.')
|
|
|
|
def handle(self, *args, **options):
|
|
bulk_size = options['bulk_size']
|
|
|
|
if not FeedInfo.objects.exists():
|
|
last_update_date = "1970-01-01"
|
|
else:
|
|
last_update_date = FeedInfo.objects.get().version
|
|
|
|
for url in self.GTFS_FEEDS.values():
|
|
last_modified = requests.head(url).headers["Last-Modified"]
|
|
last_modified = datetime.strptime(last_modified, "%a, %d %b %Y %H:%M:%S %Z")
|
|
if last_modified.date().isoformat() > last_update_date:
|
|
break
|
|
else:
|
|
self.stdout.write(self.style.WARNING("Database already up-to-date."))
|
|
return
|
|
|
|
self.stdout.write("Updating database...")
|
|
|
|
for transport_type, feed_url in self.GTFS_FEEDS.items():
|
|
self.stdout.write(f"Downloading {transport_type} GTFS feed...")
|
|
with ZipFile(BytesIO(requests.get(feed_url).content)) as zipfile:
|
|
agencies = []
|
|
for agency_dict in csv.DictReader(zipfile.read("agency.txt").decode().splitlines()):
|
|
agency_dict: dict
|
|
agency = Agency(
|
|
id=agency_dict['agency_id'],
|
|
name=agency_dict['agency_name'],
|
|
url=agency_dict['agency_url'],
|
|
timezone=agency_dict['agency_timezone'],
|
|
lang=agency_dict['agency_lang'],
|
|
phone=agency_dict.get('agency_phone', ""),
|
|
email=agency_dict.get('agency_email', ""),
|
|
)
|
|
agencies.append(agency)
|
|
if agencies:
|
|
Agency.objects.bulk_create(agencies,
|
|
update_conflicts=True,
|
|
update_fields=['name', 'url', 'timezone', 'lang', 'phone', 'email'],
|
|
unique_fields=['id'])
|
|
agencies.clear()
|
|
|
|
stops = []
|
|
for stop_dict in csv.DictReader(zipfile.read("stops.txt").decode().splitlines()):
|
|
stop_dict: dict
|
|
stop = Stop(
|
|
id=stop_dict["stop_id"],
|
|
name=stop_dict['stop_name'],
|
|
desc=stop_dict['stop_desc'],
|
|
lat=stop_dict['stop_lat'],
|
|
lon=stop_dict['stop_lon'],
|
|
zone_id=stop_dict['zone_id'],
|
|
url=stop_dict['stop_url'],
|
|
location_type=stop_dict['location_type'],
|
|
parent_station_id=stop_dict['parent_station'] or None
|
|
if last_update_date != "1970-01-01" or transport_type != "TN" else None,
|
|
timezone=stop_dict.get('stop_timezone', ""),
|
|
wheelchair_boarding=stop_dict.get('wheelchair_boarding', 0),
|
|
level_id=stop_dict.get('level_id', ""),
|
|
platform_code=stop_dict.get('platform_code', ""),
|
|
)
|
|
stops.append(stop)
|
|
|
|
if len(stops) >= bulk_size:
|
|
Stop.objects.bulk_create(stops,
|
|
update_conflicts=True,
|
|
update_fields=['name', 'desc', 'lat', 'lon', 'zone_id', 'url',
|
|
'location_type', 'parent_station_id', 'timezone',
|
|
'wheelchair_boarding', 'level_id', 'platform_code'],
|
|
unique_fields=['id'])
|
|
stops.clear()
|
|
if stops:
|
|
Stop.objects.bulk_create(stops,
|
|
update_conflicts=True,
|
|
update_fields=['name', 'desc', 'lat', 'lon', 'zone_id', 'url',
|
|
'location_type', 'parent_station_id', 'timezone',
|
|
'wheelchair_boarding', 'level_id', 'platform_code'],
|
|
unique_fields=['id'])
|
|
stops.clear()
|
|
|
|
routes = []
|
|
for route_dict in csv.DictReader(zipfile.read("routes.txt").decode().splitlines()):
|
|
route_dict: dict
|
|
route = Route(
|
|
id=route_dict['route_id'],
|
|
agency_id=route_dict['agency_id'],
|
|
short_name=route_dict['route_short_name'],
|
|
long_name=route_dict['route_long_name'],
|
|
desc=route_dict['route_desc'],
|
|
type=route_dict['route_type'],
|
|
url=route_dict['route_url'],
|
|
color=route_dict['route_color'],
|
|
text_color=route_dict['route_text_color'],
|
|
)
|
|
routes.append(route)
|
|
|
|
if len(routes) >= bulk_size:
|
|
Route.objects.bulk_create(routes,
|
|
update_conflicts=True,
|
|
update_fields=['agency_id', 'short_name', 'long_name', 'desc',
|
|
'type', 'url', 'color', 'text_color'],
|
|
unique_fields=['id'])
|
|
routes.clear()
|
|
if routes:
|
|
Route.objects.bulk_create(routes,
|
|
update_conflicts=True,
|
|
update_fields=['agency_id', 'short_name', 'long_name', 'desc',
|
|
'type', 'url', 'color', 'text_color'],
|
|
unique_fields=['id'])
|
|
routes.clear()
|
|
|
|
calendar_ids = []
|
|
if "calendar.txt" in zipfile.namelist():
|
|
calendars = []
|
|
for calendar_dict in csv.DictReader(zipfile.read("calendar.txt").decode().splitlines()):
|
|
calendar_dict: dict
|
|
calendar = Calendar(
|
|
id=f"{transport_type}-{calendar_dict['service_id']}",
|
|
monday=calendar_dict['monday'],
|
|
tuesday=calendar_dict['tuesday'],
|
|
wednesday=calendar_dict['wednesday'],
|
|
thursday=calendar_dict['thursday'],
|
|
friday=calendar_dict['friday'],
|
|
saturday=calendar_dict['saturday'],
|
|
sunday=calendar_dict['sunday'],
|
|
start_date=calendar_dict['start_date'],
|
|
end_date=calendar_dict['end_date'],
|
|
transport_type=transport_type,
|
|
)
|
|
calendars.append(calendar)
|
|
calendar_ids.append(calendar.id)
|
|
|
|
if len(calendars) >= bulk_size:
|
|
Calendar.objects.bulk_create(calendars,
|
|
update_conflicts=True,
|
|
update_fields=['monday', 'tuesday', 'wednesday', 'thursday',
|
|
'friday', 'saturday', 'sunday', 'start_date',
|
|
'end_date', 'transport_type'],
|
|
unique_fields=['id'])
|
|
calendars.clear()
|
|
if calendars:
|
|
Calendar.objects.bulk_create(calendars, update_conflicts=True,
|
|
update_fields=['monday', 'tuesday', 'wednesday', 'thursday',
|
|
'friday', 'saturday', 'sunday', 'start_date',
|
|
'end_date', 'transport_type'],
|
|
unique_fields=['id'])
|
|
calendars.clear()
|
|
|
|
calendars = []
|
|
calendar_dates = []
|
|
for calendar_date_dict in csv.DictReader(zipfile.read("calendar_dates.txt").decode().splitlines()):
|
|
calendar_date_dict: dict
|
|
calendar_date = CalendarDate(
|
|
id=f"{transport_type}-{calendar_date_dict['service_id']}-{calendar_date_dict['date']}",
|
|
service_id=f"{transport_type}-{calendar_date_dict['service_id']}",
|
|
date=calendar_date_dict['date'],
|
|
exception_type=calendar_date_dict['exception_type'],
|
|
)
|
|
calendar_dates.append(calendar_date)
|
|
|
|
if calendar_date.service_id not in calendar_ids:
|
|
calendar = Calendar(
|
|
id=f"{transport_type}-{calendar_date_dict['service_id']}",
|
|
monday=False,
|
|
tuesday=False,
|
|
wednesday=False,
|
|
thursday=False,
|
|
friday=False,
|
|
saturday=False,
|
|
sunday=False,
|
|
start_date=calendar_date_dict['date'],
|
|
end_date=calendar_date_dict['date'],
|
|
transport_type=transport_type,
|
|
)
|
|
calendars.append(calendar)
|
|
|
|
if len(calendar_dates) >= bulk_size:
|
|
Calendar.objects.bulk_create(calendars,
|
|
update_conflicts=True,
|
|
update_fields=['end_date'],
|
|
unique_fields=['id'])
|
|
CalendarDate.objects.bulk_create(calendar_dates,
|
|
update_conflicts=True,
|
|
update_fields=['service_id', 'date', 'exception_type'],
|
|
unique_fields=['id'])
|
|
calendars.clear()
|
|
calendar_dates.clear()
|
|
|
|
if calendar_dates:
|
|
Calendar.objects.bulk_create(calendars,
|
|
update_conflicts=True,
|
|
update_fields=['end_date'],
|
|
unique_fields=['id'])
|
|
CalendarDate.objects.bulk_create(calendar_dates,
|
|
update_conflicts=True,
|
|
update_fields=['service_id', 'date', 'exception_type'],
|
|
unique_fields=['id'])
|
|
calendars.clear()
|
|
calendar_dates.clear()
|
|
|
|
trips = []
|
|
for trip_dict in csv.DictReader(zipfile.read("trips.txt").decode().splitlines()):
|
|
trip_dict: dict
|
|
trip = Trip(
|
|
id=trip_dict['trip_id'],
|
|
route_id=trip_dict['route_id'],
|
|
service_id=f"{transport_type}-{trip_dict['service_id']}",
|
|
headsign=trip_dict['trip_headsign'],
|
|
short_name=trip_dict.get('trip_short_name', ""),
|
|
direction_id=trip_dict['direction_id'] or None,
|
|
block_id=trip_dict['block_id'],
|
|
shape_id=trip_dict['shape_id'],
|
|
wheelchair_accessible=trip_dict.get('wheelchair_accessible', None),
|
|
bikes_allowed=trip_dict.get('bikes_allowed', None),
|
|
)
|
|
trips.append(trip)
|
|
|
|
if len(trips) >= bulk_size:
|
|
Trip.objects.bulk_create(trips,
|
|
update_conflicts=True,
|
|
update_fields=['route_id', 'service_id', 'headsign', 'short_name',
|
|
'direction_id', 'block_id', 'shape_id',
|
|
'wheelchair_accessible', 'bikes_allowed'],
|
|
unique_fields=['id'])
|
|
trips.clear()
|
|
if trips:
|
|
Trip.objects.bulk_create(trips,
|
|
update_conflicts=True,
|
|
update_fields=['route_id', 'service_id', 'headsign', 'short_name',
|
|
'direction_id', 'block_id', 'shape_id',
|
|
'wheelchair_accessible', 'bikes_allowed'],
|
|
unique_fields=['id'])
|
|
trips.clear()
|
|
|
|
stop_times = []
|
|
for stop_time_dict in csv.DictReader(zipfile.read("stop_times.txt").decode().splitlines()):
|
|
stop_time_dict: dict
|
|
|
|
arr_time = stop_time_dict['arrival_time']
|
|
arr_time = int(arr_time[:2]) * 3600 + int(arr_time[3:5]) * 60 + int(arr_time[6:])
|
|
dep_time = stop_time_dict['departure_time']
|
|
dep_time = int(dep_time[:2]) * 3600 + int(dep_time[3:5]) * 60 + int(dep_time[6:])
|
|
st = StopTime(
|
|
id=f"{stop_time_dict['trip_id']}-{stop_time_dict['stop_id']}",
|
|
trip_id=stop_time_dict['trip_id'],
|
|
arrival_time=timedelta(seconds=arr_time),
|
|
departure_time=timedelta(seconds=dep_time),
|
|
stop_id=stop_time_dict['stop_id'],
|
|
stop_sequence=stop_time_dict['stop_sequence'],
|
|
stop_headsign=stop_time_dict['stop_headsign'],
|
|
pickup_type=stop_time_dict['pickup_type'],
|
|
drop_off_type=stop_time_dict['drop_off_type'],
|
|
timepoint=stop_time_dict.get('timepoint', None),
|
|
)
|
|
stop_times.append(st)
|
|
|
|
if len(stop_times) >= bulk_size:
|
|
StopTime.objects.bulk_create(stop_times,
|
|
update_conflicts=True,
|
|
update_fields=['stop_id', 'arrival_time', 'departure_time',
|
|
'stop_headsign', 'pickup_type',
|
|
'drop_off_type', 'timepoint'],
|
|
unique_fields=['id'])
|
|
stop_times.clear()
|
|
if stop_times:
|
|
StopTime.objects.bulk_create(stop_times,
|
|
update_conflicts=True,
|
|
update_fields=['stop_id', 'arrival_time', 'departure_time',
|
|
'stop_headsign', 'pickup_type',
|
|
'drop_off_type', 'timepoint'],
|
|
unique_fields=['id'])
|
|
stop_times.clear()
|
|
|
|
transfers = []
|
|
for transfer_dict in csv.DictReader(zipfile.read("transfers.txt").decode().splitlines()):
|
|
transfer_dict: dict
|
|
transfer = Transfer(
|
|
id=f"{transfer_dict['from_stop_id']}-{transfer_dict['to_stop_id']}",
|
|
from_stop_id=transfer_dict['from_stop_id'],
|
|
to_stop_id=transfer_dict['to_stop_id'],
|
|
transfer_type=transfer_dict['transfer_type'],
|
|
min_transfer_time=transfer_dict['min_transfer_time'],
|
|
)
|
|
transfers.append(transfer)
|
|
|
|
if len(transfers) >= bulk_size:
|
|
Transfer.objects.bulk_create(transfers,
|
|
update_conflicts=True,
|
|
update_fields=['transfer_type', 'min_transfer_time'],
|
|
unique_fields=['id'])
|
|
transfers.clear()
|
|
|
|
if transfers:
|
|
Transfer.objects.bulk_create(transfers,
|
|
update_conflicts=True,
|
|
update_fields=['transfer_type', 'min_transfer_time'],
|
|
unique_fields=['id'])
|
|
transfers.clear()
|
|
|
|
if "feed_info.txt" in zipfile.namelist():
|
|
for feed_info_dict in csv.DictReader(zipfile.read("feed_info.txt").decode().splitlines()):
|
|
feed_info_dict: dict
|
|
FeedInfo.objects.update_or_create(
|
|
publisher_name=feed_info_dict['feed_publisher_name'],
|
|
defaults=dict(
|
|
publisher_url=feed_info_dict['feed_publisher_url'],
|
|
lang=feed_info_dict['feed_lang'],
|
|
start_date=feed_info_dict['feed_start_date'],
|
|
end_date=feed_info_dict['feed_end_date'],
|
|
version=feed_info_dict['feed_version'],
|
|
)
|
|
)
|