You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

255 lines
9.6 KiB

# Copyright 2014-2016 Akretion (Alexis de Lattre
# <alexis.delattre@akretion.com>)
# Copyright 2014 Lorenzo Battistini <lorenzo.battistini@agilebg.com>
# Copyright 2017 ForgeFlow, S.L. <.com>
# Copyright 2018 Aitor Bouzas <aitor.bouzas@adaptivecity.com>
# Copyright 2016-2020 Tecnativa - Pedro M. Baeza
# License AGPL-3.0 or later (https://www.gnu.org/licenses/agpl.html).
import csv
import io
import logging
import os
import tempfile
import zipfile
import requests
from odoo import _, api, fields, models
from odoo.exceptions import UserError
logger = logging.getLogger(__name__)
class CityZipGeonamesImport(models.TransientModel):
_name = "city.zip.geonames.import"
_description = "Import City Zips from Geonames"
country_ids = fields.Many2many("res.country", string="Countries")
letter_case = fields.Selection(
[("unchanged", "Unchanged"), ("title", "Title Case"), ("upper", "Upper Case")],
string="Letter Case",
default="unchanged",
help="Converts retreived city and state names to Title Case "
"(upper case on each first letter of a word) or Upper Case "
"(all letters upper case).",
)
@api.model
def transform_city_name(self, city, country):
"""Override it for transforming city name (if needed)
:param city: Original city name
:param country: Country record
:return: Transformed city name
"""
res = city
if self.letter_case == "title":
res = city.title()
elif self.letter_case == "upper":
res = city.upper()
return res
@api.model
def _domain_search_city_zip(self, row, city_id=False):
domain = [("name", "=", row[1])]
if city_id:
domain += [("city_id", "=", city_id)]
return domain
@api.model
def select_state(self, row, country):
code = row[country.geonames_state_code_column or 4]
return self.env["res.country.state"].search(
[("country_id", "=", country.id), ("code", "=", code)], limit=1
)
@api.model
def select_city(self, row, country, state_id):
# This has to be done by SQL for performance reasons avoiding
# left join with ir_translation on the translatable field "name"
self.env.cr.execute(
"SELECT id, name FROM res_city "
"WHERE name = %s AND country_id = %s AND state_id = %s LIMIT 1",
(self.transform_city_name(row[2], country), country.id, state_id),
)
row_city = self.env.cr.fetchone()
return (row_city[0], row_city[1]) if row_city else (False, False)
@api.model
def select_zip(self, row, country, state_id):
city_id, _ = self.select_city(row, country, state_id)
return self.env["res.city.zip"].search(
self._domain_search_city_zip(row, city_id)
)
@api.model
def prepare_state(self, row, country):
return {
"name": row[country.geonames_state_name_column or 3],
"code": row[country.geonames_state_code_column or 4],
"country_id": country.id,
}
@api.model
def prepare_city(self, row, country, state_id):
vals = {
"name": self.transform_city_name(row[2], country),
"state_id": state_id,
"country_id": country.id,
}
return vals
@api.model
def prepare_zip(self, row, city_id):
vals = {"name": row[1], "city_id": city_id}
return vals
@api.model
def get_and_parse_csv(self, country):
country_code = country.code
config_url = self.env["ir.config_parameter"].get_param(
"geonames.url", default="http://download.geonames.org/export/zip/%s.zip"
)
url = config_url % country_code
logger.info("Starting to download %s" % url)
res_request = requests.get(url)
if res_request.status_code != requests.codes.ok:
raise UserError(
_("Got an error %d when trying to download the file %s.")
% (res_request.status_code, url)
)
f_geonames = zipfile.ZipFile(io.BytesIO(res_request.content))
tempdir = tempfile.mkdtemp(prefix="odoo")
f_geonames.extract("%s.txt" % country_code, tempdir)
data_file = open(
os.path.join(tempdir, "%s.txt" % country_code), "r", encoding="utf-8"
)
data_file.seek(0)
reader = csv.reader(data_file, delimiter=" ")
parsed_csv = [row for i, row in enumerate(reader)]
data_file.close()
logger.info("The geonames zipfile has been decompressed")
return parsed_csv
def _create_states(self, parsed_csv, search_states, max_import, country):
# States
state_vals_list = []
state_dict = {}
for i, row in enumerate(parsed_csv):
if max_import and i == max_import:
break
state = self.select_state(row, country) if search_states else False
if not state:
state_vals = self.prepare_state(row, country)
if state_vals not in state_vals_list:
state_vals_list.append(state_vals)
else:
state_dict[state.code] = state.id
created_states = self.env["res.country.state"].create(state_vals_list)
for i, vals in enumerate(state_vals_list):
state_dict[vals["code"]] = created_states[i].id
return state_dict
def _create_cities(
self, parsed_csv, search_cities, max_import, state_dict, country
):
# Cities
city_vals_list = []
city_dict = {}
for i, row in enumerate(parsed_csv):
if max_import and i == max_import:
break
state_id = state_dict[row[country.geonames_state_code_column or 4]]
city_id, city_name = (
self.select_city(row, country, state_id)
if search_cities
else (False, False)
)
if not city_id:
city_vals = self.prepare_city(row, country, state_id)
if city_vals not in city_vals_list:
city_vals_list.append(city_vals)
else:
city_dict[(city_name, state_id)] = city_id
ctx = dict(self.env.context)
ctx.pop("lang", None) # make sure no translation is added
created_cities = self.env["res.city"].with_context(ctx).create(city_vals_list)
for i, vals in enumerate(city_vals_list):
city_dict[(vals["name"], vals["state_id"])] = created_cities[i].id
return city_dict
def run_import(self):
for country in self.country_ids:
parsed_csv = self.get_and_parse_csv(country)
self._process_csv(parsed_csv, country)
return True
def _action_remove_old_records(self, model_name, old_records, country):
model = self.env[model_name]
items = model.browse(list(old_records))
try:
logger.info("removing %s entries" % model._name)
items.unlink()
logger.info(
"%d entries deleted for country %s" % (len(old_records), country.name)
)
except Exception:
for item in items:
try:
item.unlink()
except Exception:
logger.info(_("%d could not be deleted %") % item.name)
def _process_csv(self, parsed_csv, country):
state_model = self.env["res.country.state"]
zip_model = self.env["res.city.zip"]
res_city_model = self.env["res.city"]
# Store current record list
old_zips = set(zip_model.search([("city_id.country_id", "=", country.id)]).ids)
search_zips = len(old_zips) > 0
old_cities = set(res_city_model.search([("country_id", "=", country.id)]).ids)
search_cities = len(old_cities) > 0
current_states = state_model.search([("country_id", "=", country.id)])
search_states = len(current_states) > 0
max_import = self.env.context.get("max_import", 0)
logger.info("Starting to create the cities and/or city zip entries")
# Pre-create states and cities
state_dict = self._create_states(parsed_csv, search_states, max_import, country)
city_dict = self._create_cities(
parsed_csv, search_cities, max_import, state_dict, country
)
# Zips
zip_vals_list = []
for i, row in enumerate(parsed_csv):
if max_import and i == max_import:
break
# Don't search if there aren't any records
zip_code = False
state_id = state_dict[row[country.geonames_state_code_column or 4]]
if search_zips:
zip_code = self.select_zip(row, country, state_id)
if not zip_code:
city_id = city_dict[
(self.transform_city_name(row[2], country), state_id)
]
zip_vals = self.prepare_zip(row, city_id)
if zip_vals not in zip_vals_list:
zip_vals_list.append(zip_vals)
else:
old_zips.discard(zip_code.id)
zip_model.create(zip_vals_list)
if not max_import:
if old_zips:
self._action_remove_old_records("res.city.zip", old_zips, country)
old_cities -= set(city_dict.values())
if old_cities:
self._action_remove_old_records("res.city", old_cities, country)
logger.info(
"The wizard to create cities and/or city zip entries from "
"geonames has been successfully completed."
)
return True