Browse Source

[mrg] Add module base_partner_merge backport from trunk

Add module account_partner_merge needed to fix merging when account is installed
pull/2/head
Leonardo Pistone 11 years ago
parent
commit
b87ef03477
  1. 21
      account_partner_merge/__init__.py
  2. 34
      account_partner_merge/__openerp__.py
  3. 17
      account_partner_merge/account_partner_merge_view.xml
  4. 35
      account_partner_merge/partner_merge.py
  5. 1
      base_partner_merge/__init__.py
  6. 15
      base_partner_merge/__openerp__.py
  7. 897
      base_partner_merge/base_partner_merge.py
  8. 123
      base_partner_merge/base_partner_merge_view.xml
  9. 3
      base_partner_merge/security/ir.model.access.csv
  10. 123
      base_partner_merge/validate_email.py
  11. 21
      portal_partner_merge/__init__.py
  12. 36
      portal_partner_merge/__openerp__.py
  13. 26
      portal_partner_merge/wizard/__init__.py
  14. 39
      portal_partner_merge/wizard/portal_wizard.py

21
account_partner_merge/__init__.py

@ -0,0 +1,21 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# Author: Yannick Vaucher
# Copyright 2013 Camptocamp SA
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
from . import partner_merge

34
account_partner_merge/__openerp__.py

@ -0,0 +1,34 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# Author: Yannick Vaucher
# Copyright 2013 Camptocamp SA
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
{'name' : 'Account Partner Merge',
'version' : '1.0',
'category': 'Hidden',
'description': """Update invoice commercial_partner_id""",
'author' : 'Camptocamp',
'maintainer': 'Camptocamp',
'website': 'http://www.camptocamp.com/',
'depends' : ['account_report_company', 'base_partner_merge'],
'data': ['account_partner_merge_view.xml'],
'test': [],
'installable': True,
'auto_install': True,
'application': False,
}

17
account_partner_merge/account_partner_merge_view.xml

@ -0,0 +1,17 @@
<?xml version="1.0" encoding="UTF-8"?>
<openerp>
<data>
<record model='ir.ui.view' id='base_partner_merge_automatic_wizard_form'>
<field name='name'>account.partner.merge.automatic.wizard.form</field>
<field name='model'>base.partner.merge.automatic.wizard</field>
<field name='inherit_id' ref='base_partner_merge.base_partner_merge_automatic_wizard_form'/>
<field name='arch' type='xml'>
<xpath expr="//field[@name='partner_ids']/tree/field[@name='name']" position="replace">
<field name="display_name" />
</xpath>
</field>
</record>
</data>
</openerp>

35
account_partner_merge/partner_merge.py

@ -0,0 +1,35 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# Author: Yannick Vaucher
# Copyright 2013 Camptocamp SA
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
from openerp.osv import osv
class MergePartnerAutomatic(osv.TransientModel):
_inherit = 'base.partner.merge.automatic.wizard'
def _update_values(self, cr, uid, src_partners, dst_partner, context=None):
"""
Make sure we don't forget to update the stored value of invoice field commercial_partner_id
"""
super(MergePartnerAutomatic, self)._update_values(cr, uid, src_partners, dst_partner, context=context)
invoice_obj = self.pool.get('account.invoice')
invoice_ids = invoice_obj.search(cr, uid, [('partner_id', '=', dst_partner.id)], context=context)
# call write to refresh stored value
invoice_obj.write(cr, uid, invoice_ids, {}, context=context)

1
base_partner_merge/__init__.py

@ -0,0 +1 @@
import base_partner_merge

15
base_partner_merge/__openerp__.py

@ -0,0 +1,15 @@
{
'name': 'Base Partner Merge',
'author': 'OpenERP S.A.',
'category': 'Generic Modules/Base',
'version': '0.1',
'description': """backport module, to be removed when we switch to saas2 on the private servers""",
'depends': [
'base',
],
'data': [
'security/ir.model.access.csv',
'base_partner_merge_view.xml',
],
'installable': True,
}

897
base_partner_merge/base_partner_merge.py

@ -0,0 +1,897 @@
#!/usr/bin/env python
from __future__ import absolute_import
from email.utils import parseaddr
import functools
import htmlentitydefs
import itertools
import logging
import operator
import re
from ast import literal_eval
from openerp.tools import mute_logger
# Validation Library https://pypi.python.org/pypi/validate_email/1.1
from .validate_email import validate_email
import openerp
from openerp.osv import osv, orm
from openerp.osv import fields
from openerp.osv.orm import browse_record
from openerp.tools.translate import _
pattern = re.compile("&(\w+?);")
_logger = logging.getLogger('base.partner.merge')
# http://www.php2python.com/wiki/function.html-entity-decode/
def html_entity_decode_char(m, defs=htmlentitydefs.entitydefs):
try:
return defs[m.group(1)]
except KeyError:
return m.group(0)
def html_entity_decode(string):
return pattern.sub(html_entity_decode_char, string)
def sanitize_email(partner_email):
assert isinstance(partner_email, basestring) and partner_email
result = re.subn(r';|/|:', ',',
html_entity_decode(partner_email or ''))[0].split(',')
emails = [parseaddr(email)[1]
for item in result
for email in item.split()]
return [email.lower()
for email in emails
if validate_email(email)]
def is_integer_list(ids):
return all(isinstance(i, (int, long)) for i in ids)
class ResPartner(osv.Model):
_inherit = 'res.partner'
_columns = {
'id': fields.integer('Id', readonly=True),
'create_date': fields.datetime('Create Date', readonly=True),
}
class MergePartnerLine(osv.TransientModel):
_name = 'base.partner.merge.line'
_columns = {
'wizard_id': fields.many2one('base.partner.merge.automatic.wizard',
'Wizard'),
'min_id': fields.integer('MinID'),
'aggr_ids': fields.char('Ids', required=True),
}
_order = 'min_id asc'
class MergePartnerAutomatic(osv.TransientModel):
"""
The idea behind this wizard is to create a list of potential partners to
merge. We use two objects, the first one is the wizard for the end-user.
And the second will contain the partner list to merge.
"""
_name = 'base.partner.merge.automatic.wizard'
_columns = {
# Group by
'group_by_email': fields.boolean('Email'),
'group_by_name': fields.boolean('Name'),
'group_by_is_company': fields.boolean('Is Company'),
'group_by_vat': fields.boolean('VAT'),
'group_by_parent_id': fields.boolean('Parent Company'),
'state': fields.selection([('option', 'Option'),
('selection', 'Selection'),
('finished', 'Finished')],
'State',
readonly=True,
required=True),
'number_group': fields.integer("Group of Contacts", readonly=True),
'current_line_id': fields.many2one('base.partner.merge.line',
'Current Line'),
'line_ids': fields.one2many('base.partner.merge.line',
'wizard_id', 'Lines'),
'partner_ids': fields.many2many('res.partner', string='Contacts'),
'dst_partner_id': fields.many2one('res.partner',
string='Destination Contact'),
'exclude_contact': fields.boolean('A user associated to the contact'),
'exclude_journal_item': fields.boolean('Journal Items associated'
' to the contact'),
'maximum_group': fields.integer("Maximum of Group of Contacts"),
}
def default_get(self, cr, uid, fields, context=None):
if context is None:
context = {}
res = super(MergePartnerAutomatic, self
).default_get(cr, uid, fields, context)
if (context.get('active_model') == 'res.partner'
and context.get('active_ids')):
partner_ids = context['active_ids']
res['state'] = 'selection'
res['partner_ids'] = partner_ids
res['dst_partner_id'] = self._get_ordered_partner(cr, uid,
partner_ids,
context=context
)[-1].id
return res
_defaults = {
'state': 'option'
}
def get_fk_on(self, cr, table):
q = """ SELECT cl1.relname as table,
att1.attname as column
FROM pg_constraint as con, pg_class as cl1, pg_class as cl2,
pg_attribute as att1, pg_attribute as att2
WHERE con.conrelid = cl1.oid
AND con.confrelid = cl2.oid
AND array_lower(con.conkey, 1) = 1
AND con.conkey[1] = att1.attnum
AND att1.attrelid = cl1.oid
AND cl2.relname = %s
AND att2.attname = 'id'
AND array_lower(con.confkey, 1) = 1
AND con.confkey[1] = att2.attnum
AND att2.attrelid = cl2.oid
AND con.contype = 'f'
"""
return cr.execute(q, (table,))
def _update_foreign_keys(self, cr, uid, src_partners,
dst_partner, context=None):
_logger.debug('_update_foreign_keys for dst_partner: %s for '
'src_partners: %r',
dst_partner.id,
list(map(operator.attrgetter('id'), src_partners)))
# find the many2one relation to a partner
proxy = self.pool.get('res.partner')
self.get_fk_on(cr, 'res_partner')
# ignore two tables
for table, column in cr.fetchall():
if 'base_partner_merge_' in table:
continue
partner_ids = tuple(map(int, src_partners))
query = ("SELECT column_name FROM information_schema.columns"
" WHERE table_name LIKE '%s'") % (table)
cr.execute(query, ())
columns = []
for data in cr.fetchall():
if data[0] != column:
columns.append(data[0])
query_dic = {
'table': table,
'column': column,
'value': columns[0],
}
if len(columns) <= 1:
# unique key treated
query = """
UPDATE "%(table)s" as ___tu
SET %(column)s = %%s
WHERE
%(column)s = %%s AND
NOT EXISTS (
SELECT 1
FROM "%(table)s" as ___tw
WHERE
%(column)s = %%s AND
___tu.%(value)s = ___tw.%(value)s
)""" % query_dic
for partner_id in partner_ids:
cr.execute(query, (dst_partner.id, partner_id,
dst_partner.id))
else:
cr.execute("SAVEPOINT recursive_partner_savepoint")
try:
query = ('UPDATE "%(table)s" SET %(column)s = %%s WHERE '
'%(column)s IN %%s') % query_dic
cr.execute(query, (dst_partner.id, partner_ids,))
if (column == proxy._parent_name
and table == 'res_partner'):
query = """
WITH RECURSIVE cycle(id, parent_id) AS (
SELECT id, parent_id FROM res_partner
UNION
SELECT cycle.id, res_partner.parent_id
FROM res_partner, cycle
WHERE res_partner.id = cycle.parent_id
AND cycle.id != cycle.parent_id
)
SELECT id FROM cycle
WHERE id = parent_id AND id = %s
"""
cr.execute(query, (dst_partner.id,))
if cr.fetchall():
cr.execute("ROLLBACK TO SAVEPOINT "
"recursive_partner_savepoint")
finally:
cr.execute("RELEASE SAVEPOINT "
"recursive_partner_savepoint")
def _update_reference_fields(self, cr, uid, src_partners, dst_partner,
context=None):
_logger.debug('_update_reference_fields for dst_partner: %s for '
'src_partners: %r',
dst_partner.id,
list(map(operator.attrgetter('id'), src_partners)))
def update_records(model, src, field_model='model', field_id='res_id',
context=None):
proxy = self.pool.get(model)
if proxy is None:
return
domain = [(field_model, '=', 'res.partner'),
(field_id, '=', src.id)]
ids = proxy.search(cr, openerp.SUPERUSER_ID,
domain, context=context)
return proxy.write(cr, openerp.SUPERUSER_ID, ids,
{field_id: dst_partner.id}, context=context)
update_records = functools.partial(update_records, context=context)
for partner in src_partners:
update_records('base.calendar', src=partner,
field_model='model_id.model')
update_records('ir.attachment', src=partner,
field_model='res_model')
update_records('mail.followers', src=partner,
field_model='res_model')
update_records('mail.message', src=partner)
update_records('marketing.campaign.workitem', src=partner,
field_model='object_id.model')
update_records('ir.model.data', src=partner)
proxy = self.pool['ir.model.fields']
domain = [('ttype', '=', 'reference')]
record_ids = proxy.search(cr, openerp.SUPERUSER_ID, domain,
context=context)
for record in proxy.browse(cr, openerp.SUPERUSER_ID, record_ids,
context=context):
try:
proxy_model = self.pool[record.model]
except KeyError:
# ignore old tables
continue
if record.model == 'ir.property':
continue
field_type = proxy_model._columns.get(record.name).__class__._type
if field_type == 'function':
continue
for partner in src_partners:
domain = [
(record.name, '=', 'res.partner,%d' % partner.id)
]
model_ids = proxy_model.search(cr, openerp.SUPERUSER_ID,
domain, context=context)
values = {
record.name: 'res.partner,%d' % dst_partner.id,
}
proxy_model.write(cr, openerp.SUPERUSER_ID, model_ids, values,
context=context)
def _update_values(self, cr, uid, src_partners, dst_partner, context=None):
_logger.debug('_update_values for dst_partner: %s for src_partners: '
'%r',
dst_partner.id,
list(map(operator.attrgetter('id'), src_partners)))
columns = dst_partner._columns
def write_serializer(column, item):
if isinstance(item, browse_record):
return item.id
else:
return item
values = dict()
for column, field in columns.iteritems():
if (field._type not in ('many2many', 'one2many')
and not isinstance(field, fields.function)):
for item in itertools.chain(src_partners, [dst_partner]):
if item[column]:
values[column] = write_serializer(column,
item[column])
values.pop('id', None)
parent_id = values.pop('parent_id', None)
dst_partner.write(values)
if parent_id and parent_id != dst_partner.id:
try:
dst_partner.write({'parent_id': parent_id})
except (osv.except_osv, orm.except_orm):
_logger.info('Skip recursive partner hierarchies for '
'parent_id %s of partner: %s',
parent_id, dst_partner.id)
@mute_logger('openerp.osv.expression', 'openerp.osv.orm')
def _merge(self, cr, uid, partner_ids, dst_partner=None, context=None):
proxy = self.pool.get('res.partner')
partner_ids = proxy.exists(cr, uid, list(partner_ids),
context=context)
if len(partner_ids) < 2:
return
if len(partner_ids) > 3:
raise osv.except_osv(
_('Error'),
_("For safety reasons, you cannot merge more than 3 contacts "
"together. You can re-open the wizard several times if "
"needed."))
if (openerp.SUPERUSER_ID != uid
and len(set(partner.email for partner
in proxy.browse(cr, uid, partner_ids,
context=context))) > 1):
raise osv.except_osv(
_('Error'),
_("All contacts must have the same email. Only the "
"Administrator can merge contacts with different emails."))
if dst_partner and dst_partner.id in partner_ids:
src_partners = proxy.browse(cr, uid,
[id for id in partner_ids
if id != dst_partner.id],
context=context)
else:
ordered_partners = self._get_ordered_partner(cr, uid, partner_ids,
context)
dst_partner = ordered_partners[-1]
src_partners = ordered_partners[:-1]
_logger.info("dst_partner: %s", dst_partner.id)
if (openerp.SUPERUSER_ID != uid
and self._model_is_installed(cr, uid, 'account.move.line',
context=context)
and self.pool.get('account.move.line'
).search(cr, openerp.SUPERUSER_ID,
[('partner_id',
'in',
[partner.id for partner
in src_partners])],
context=context)):
raise osv.except_osv(
_('Error'),
_("Only the destination contact may be linked to existing "
"Journal Items. Please ask the Administrator if you need to"
" merge several contacts linked to existing Journal "
"Items."))
call_it = lambda function: function(cr, uid, src_partners,
dst_partner, context=context)
call_it(self._update_foreign_keys)
call_it(self._update_reference_fields)
call_it(self._update_values)
_logger.info('(uid = %s) merged the partners %r with %s',
uid,
list(map(operator.attrgetter('id'), src_partners)),
dst_partner.id)
dst_partner.message_post(
body='%s %s' % (
_("Merged with the following partners:"),
", ".join('%s<%s>(ID %s)' % (p.name, p.email or 'n/a', p.id)
for p in src_partners)))
for partner in src_partners:
partner.unlink()
def clean_emails(self, cr, uid, context=None):
"""
Clean the email address of the partner, if there is an email field
with a minimum of two addresses, the system will create a new partner,
with the information of the previous one and will copy the new cleaned
email into the email field.
"""
if context is None:
context = {}
proxy_model = self.pool['ir.model.fields']
field_ids = proxy_model.search(cr, uid,
[('model', '=', 'res.partner'),
('ttype', 'like', '%2many')],
context=context)
fields = proxy_model.read(cr, uid, field_ids, context=context)
reset_fields = dict((field['name'], []) for field in fields)
proxy_partner = self.pool['res.partner']
context['active_test'] = False
ids = proxy_partner.search(cr, uid, [], context=context)
fields = ['name', 'var' 'partner_id' 'is_company', 'email']
partners = proxy_partner.read(cr, uid, ids, fields, context=context)
partners.sort(key=operator.itemgetter('id'))
partners_len = len(partners)
_logger.info('partner_len: %r', partners_len)
for idx, partner in enumerate(partners):
if not partner['email']:
continue
percent = (idx / float(partners_len)) * 100.0
_logger.info('idx: %r', idx)
_logger.info('percent: %r', percent)
try:
emails = sanitize_email(partner['email'])
head, tail = emails[:1], emails[1:]
email = head[0] if head else False
proxy_partner.write(cr, uid, [partner['id']],
{'email': email}, context=context)
for email in tail:
values = dict(reset_fields, email=email)
proxy_partner.copy(cr, uid, partner['id'], values,
context=context)
except Exception:
_logger.exception("There is a problem with this partner: %r",
partner)
raise
return True
def close_cb(self, cr, uid, ids, context=None):
return {'type': 'ir.actions.act_window_close'}
def _generate_query(self, fields, maximum_group=100):
group_fields = ', '.join(fields)
filters = []
for field in fields:
if field in ['email', 'name']:
filters.append((field, 'IS NOT', 'NULL'))
criteria = ' AND '.join('%s %s %s' % (field, operator, value)
for field, operator, value in filters)
text = [
"SELECT min(id), array_agg(id)",
"FROM res_partner",
]
if criteria:
text.append('WHERE %s' % criteria)
text.extend([
"GROUP BY %s" % group_fields,
"HAVING COUNT(*) >= 2",
"ORDER BY min(id)",
])
if maximum_group:
text.extend([
"LIMIT %s" % maximum_group,
])
return ' '.join(text)
def _compute_selected_groupby(self, this):
group_by_str = 'group_by_'
group_by_len = len(group_by_str)
fields = [
key[group_by_len:]
for key in self._columns.keys()
if key.startswith(group_by_str)
]
groups = [
field
for field in fields
if getattr(this, '%s%s' % (group_by_str, field), False)
]
if not groups:
raise osv.except_osv(_('Error'),
_("You have to specify a filter for your "
"selection"))
return groups
def next_cb(self, cr, uid, ids, context=None):
"""
Don't compute any thing
"""
context = dict(context or {}, active_test=False)
this = self.browse(cr, uid, ids[0], context=context)
if this.current_line_id:
this.current_line_id.unlink()
return self._next_screen(cr, uid, this, context)
def _get_ordered_partner(self, cr, uid, partner_ids, context=None):
partners = self.pool.get('res.partner'
).browse(cr, uid,
list(partner_ids),
context=context)
ordered_partners = sorted(sorted(partners,
key=operator.attrgetter('create_date'),
reverse=True),
key=operator.attrgetter('active'),
reverse=True)
return ordered_partners
def _next_screen(self, cr, uid, this, context=None):
this.refresh()
values = {}
if this.line_ids:
# in this case, we try to find the next record.
current_line = this.line_ids[0]
current_partner_ids = literal_eval(current_line.aggr_ids)
values.update({
'current_line_id': current_line.id,
'partner_ids': [(6, 0, current_partner_ids)],
'dst_partner_id': self._get_ordered_partner(
cr, uid,
current_partner_ids,
context
)[-1].id,
'state': 'selection',
})
else:
values.update({
'current_line_id': False,
'partner_ids': [],
'state': 'finished',
})
this.write(values)
return {
'type': 'ir.actions.act_window',
'res_model': this._name,
'res_id': this.id,
'view_mode': 'form',
'target': 'new',
}
def _model_is_installed(self, cr, uid, model, context=None):
proxy = self.pool.get('ir.model')
domain = [('model', '=', model)]
return proxy.search_count(cr, uid, domain, context=context) > 0
def _partner_use_in(self, cr, uid, aggr_ids, models, context=None):
"""
Check if there is no occurence of this group of partner in the selected
model
"""
for model, field in models.iteritems():
proxy = self.pool.get(model)
domain = [(field, 'in', aggr_ids)]
if proxy.search_count(cr, uid, domain, context=context):
return True
return False
def compute_models(self, cr, uid, ids, context=None):
"""
Compute the different models needed by the system if you want to
exclude some partners.
"""
assert is_integer_list(ids)
this = self.browse(cr, uid, ids[0], context=context)
models = {}
if this.exclude_contact:
models['res.users'] = 'partner_id'
if (self._model_is_installed(cr, uid, 'account.move.line',
context=context)
and this.exclude_journal_item):
models['account.move.line'] = 'partner_id'
return models
def _process_query(self, cr, uid, ids, query, context=None):
"""
Execute the select request and write the result in this wizard
"""
proxy = self.pool.get('base.partner.merge.line')
this = self.browse(cr, uid, ids[0], context=context)
models = self.compute_models(cr, uid, ids, context=context)
cr.execute(query)
counter = 0
for min_id, aggr_ids in cr.fetchall():
if models and self._partner_use_in(cr, uid, aggr_ids, models,
context=context):
continue
values = {
'wizard_id': this.id,
'min_id': min_id,
'aggr_ids': aggr_ids,
}
proxy.create(cr, uid, values, context=context)
counter += 1
values = {
'state': 'selection',
'number_group': counter,
}
this.write(values)
_logger.info("counter: %s", counter)
def start_process_cb(self, cr, uid, ids, context=None):
"""
Start the process.
* Compute the selected groups (with duplication)
* If the user has selected the 'exclude_XXX' fields, avoid the
partners.
"""
assert is_integer_list(ids)
context = dict(context or {}, active_test=False)
this = self.browse(cr, uid, ids[0], context=context)
groups = self._compute_selected_groupby(this)
query = self._generate_query(groups, this.maximum_group)
self._process_query(cr, uid, ids, query, context=context)
return self._next_screen(cr, uid, this, context)
def automatic_process_cb(self, cr, uid, ids, context=None):
assert is_integer_list(ids)
this = self.browse(cr, uid, ids[0], context=context)
this.start_process_cb()
this.refresh()
for line in this.line_ids:
partner_ids = literal_eval(line.aggr_ids)
self._merge(cr, uid, partner_ids, context=context)
line.unlink()
cr.commit()
this.write({'state': 'finished'})
return {
'type': 'ir.actions.act_window',
'res_model': this._name,
'res_id': this.id,
'view_mode': 'form',
'target': 'new',
}
def parent_migration_process_cb(self, cr, uid, ids, context=None):
assert is_integer_list(ids)
context = dict(context or {}, active_test=False)
this = self.browse(cr, uid, ids[0], context=context)
query = """
SELECT
min(p1.id),
array_agg(DISTINCT p1.id)
FROM
res_partner as p1
INNER join
res_partner as p2
ON
p1.email = p2.email AND
p1.name = p2.name AND
(p1.parent_id = p2.id OR p1.id = p2.parent_id)
WHERE
p2.id IS NOT NULL
GROUP BY
p1.email,
p1.name,
CASE WHEN p1.parent_id = p2.id THEN p2.id
ELSE p1.id
END
HAVING COUNT(*) >= 2
ORDER BY
min(p1.id)
"""
self._process_query(cr, uid, ids, query, context=context)
for line in this.line_ids:
partner_ids = literal_eval(line.aggr_ids)
self._merge(cr, uid, partner_ids, context=context)
line.unlink()
cr.commit()
this.write({'state': 'finished'})
cr.execute("""
UPDATE
res_partner
SET
is_company = NULL,
parent_id = NULL
WHERE
parent_id = id
""")
return {
'type': 'ir.actions.act_window',
'res_model': this._name,
'res_id': this.id,
'view_mode': 'form',
'target': 'new',
}
def update_all_process_cb(self, cr, uid, ids, context=None):
assert is_integer_list(ids)
# WITH RECURSIVE cycle(id, parent_id) AS (
# SELECT id, parent_id FROM res_partner
# UNION
# SELECT cycle.id, res_partner.parent_id
# FROM res_partner, cycle
# WHERE res_partner.id = cycle.parent_id AND
# cycle.id != cycle.parent_id
# )
# UPDATE res_partner
# SET parent_id = NULL
# WHERE id in (SELECT id FROM cycle WHERE id = parent_id);
this = self.browse(cr, uid, ids[0], context=context)
self.parent_migration_process_cb(cr, uid, ids, context=None)
list_merge = [
{'group_by_vat': True,
'group_by_email': True,
'group_by_name': True},
# {'group_by_name': True,
# 'group_by_is_company': True,
# 'group_by_parent_id': True},
# {'group_by_email': True,
# 'group_by_is_company': True,
# 'group_by_parent_id': True},
# {'group_by_name': True,
# 'group_by_vat': True,
# 'group_by_is_company': True,
# 'exclude_journal_item': True},
# {'group_by_email': True,
# 'group_by_vat': True,
# 'group_by_is_company': True,
# 'exclude_journal_item': True},
# {'group_by_email': True,
# 'group_by_is_company': True,
# 'exclude_contact': True,
# 'exclude_journal_item': True},
# {'group_by_name': True,
# 'group_by_is_company': True,
# 'exclude_contact': True,
# 'exclude_journal_item': True}
]
for merge_value in list_merge:
id = self.create(cr, uid, merge_value, context=context)
self.automatic_process_cb(cr, uid, [id], context=context)
cr.execute("""
UPDATE
res_partner
SET
is_company = NULL
WHERE
parent_id IS NOT NULL AND
is_company IS NOT NULL
""")
# cr.execute("""
# UPDATE
# res_partner as p1
# SET
# is_company = NULL,
# parent_id = (
# SELECT p2.id
# FROM res_partner as p2
# WHERE p2.email = p1.email AND
# p2.parent_id != p2.id
# LIMIT 1
# )
# WHERE
# p1.parent_id = p1.id
# """)
return self._next_screen(cr, uid, this, context)
def merge_cb(self, cr, uid, ids, context=None):
assert is_integer_list(ids)
context = dict(context or {}, active_test=False)
this = self.browse(cr, uid, ids[0], context=context)
partner_ids = set(map(int, this.partner_ids))
if not partner_ids:
this.write({'state': 'finished'})
return {
'type': 'ir.actions.act_window',
'res_model': this._name,
'res_id': this.id,
'view_mode': 'form',
'target': 'new',
}
self._merge(cr, uid, partner_ids, this.dst_partner_id,
context=context)
if this.current_line_id:
this.current_line_id.unlink()
return self._next_screen(cr, uid, this, context)
def auto_set_parent_id(self, cr, uid, ids, context=None):
assert is_integer_list(ids)
# select partner who have one least invoice
partner_treated = ['@gmail.com']
cr.execute(""" SELECT p.id, p.email
FROM res_partner as p
LEFT JOIN account_invoice as a
ON p.id = a.partner_id AND a.state in ('open','paid')
WHERE p.grade_id is NOT NULL
GROUP BY p.id
ORDER BY COUNT(a.id) DESC
""")
re_email = re.compile(r".*@")
for id, email in cr.fetchall():
# check email domain
email = re_email.sub("@", email or "")
if not email or email in partner_treated:
continue
partner_treated.append(email)
# don't update the partners if they are more of one who have
# invoice
cr.execute("""
SELECT *
FROM res_partner as p
WHERE p.id != %s AND p.email LIKE '%%%s' AND
EXISTS (SELECT * FROM account_invoice as a
WHERE p.id = a.partner_id
AND a.state in ('open','paid'))
""" % (id, email))
if len(cr.fetchall()) > 1:
_logger.info("%s MORE OF ONE COMPANY", email)
continue
# to display changed values
cr.execute(""" SELECT id,email
FROM res_partner
WHERE parent_id != %s
AND id != %s AND email LIKE '%%%s'
""" % (id, id, email))
_logger.info("%r", cr.fetchall())
# upgrade
cr.execute(""" UPDATE res_partner
SET parent_id = %s
WHERE id != %s AND email LIKE '%%%s'
""" % (id, id, email))
return False

123
base_partner_merge/base_partner_merge_view.xml

@ -0,0 +1,123 @@
<?xml version="1.0" encoding="UTF-8"?>
<openerp>
<data>
<!-- the sequence of the configuration sub menu is 30 -->
<menuitem id='root_menu' name='Tools' parent='base.menu_base_partner' sequence="25"/>
<record model="ir.actions.act_window" id="base_partner_merge_automatic_act">
<field name="name">Deduplicate Contacts</field>
<field name="res_model">base.partner.merge.automatic.wizard</field>
<field name="view_type">form</field>
<field name="view_mode">form</field>
<field name="target">new</field>
<field name="context">{'active_test': False}</field>
</record>
<menuitem id='partner_merge_automatic_menu'
action='base_partner_merge_automatic_act'
groups='base.group_system'
parent='root_menu' />
<record model='ir.ui.view' id='base_partner_merge_automatic_wizard_form'>
<field name='name'>base.partner.merge.automatic.wizard.form</field>
<field name='model'>base.partner.merge.automatic.wizard</field>
<field name='arch' type='xml'>
<form string='Automatic Merge Wizard' version='7.0'>
<header>
<button name='merge_cb' string='Merge Selection'
class='oe_highlight'
type='object'
attrs="{'invisible': [('state', 'in', ('option', 'finished' ))]}"
/>
<button name='next_cb' string='Skip these contacts'
type='object' class='oe_link'
attrs="{'invisible': [('state', '!=', 'selection')]}" />
<button name='start_process_cb'
string='Merge with Manual Check'
type='object' class='oe_highlight'
attrs="{'invisible': [('state', '!=', 'option')]}" />
<button name='automatic_process_cb'
string='Merge Automatically'
type='object' class='oe_highlight'
confirm="Are you sure to execute the automatic merge of your contacts ?"
attrs="{'invisible': [('state', '!=', 'option')]}" />
<button name='update_all_process_cb'
string='Merge Automatically all process'
type='object'
confirm="Are you sure to execute the list of automatic merges of your contacts ?"
attrs="{'invisible': [('state', '!=', 'option')]}" />
<span class="or_cancel" attrs="{'invisible': [('state', '=', 'finished')]} ">or
<button name="close_cb" special="nosave" string="Cancel" type="object" class="oe_link oe_inline"/>
</span>
<span class="or_cancel" attrs="{'invisible': [('state', '!=', 'finished')]} ">
<button name="close_cb" special="nosave"
string="Close"
type="object"
class="oe_link oe_inline"/>
</span>
</header>
<sheet>
<group attrs="{'invisible': [('state', '!=', 'finished')]}" col="1">
<h2>There is no more contacts to merge for this request...</h2>
<button name="%(base_partner_merge_automatic_act)d" string="Deduplicate the other Contacts" class="oe_highlight"
type="action"/>
</group>
<p class="oe_grey" attrs="{'invisible': [('state', '!=', ('option'))]}">
Select the list of fields used to search for
duplicated records. If you select several fields,
OpenERP will propose you to merge only those having
all these fields in common. (not one of the fields).
</p>
<group attrs="{'invisible': ['|', ('state', 'not in', ('selection', 'finished')), ('number_group', '=', 0)]}">
<field name="state" invisible="1" />
<field name="number_group"/>
</group>
<group string="Search duplicates based on duplicated data in"
attrs="{'invisible': [('state', 'not in', ('option',))]}">
<field name='group_by_email' />
<field name='group_by_name' />
<field name='group_by_is_company' />
<field name='group_by_vat' />
<field name='group_by_parent_id' />
</group>
<group string="Exclude contacts having"
attrs="{'invisible': [('state', 'not in', ('option',))]}">
<field name='exclude_contact' />
<field name='exclude_journal_item' />
</group>
<separator string="Options" attrs="{'invisible': [('state', 'not in', ('option',))]}"/>
<group attrs="{'invisible': [('state', 'not in', ('option','finished'))]}">
<field name='maximum_group' attrs="{'readonly': [('state', 'in', ('finished'))]}"/>
</group>
<separator string="Merge the following contacts"
attrs="{'invisible': [('state', 'in', ('option', 'finished'))]}"/>
<group attrs="{'invisible': [('state', 'in', ('option', 'finished'))]}" col="1">
<p class="oe_grey">
The selected contacts will be merged together. All
documents linking to one of these contacts will be
redirected to the aggregated contact. You can remove
contacts from this list to avoid merging them.
</p>
<field name="dst_partner_id" domain="[('id', 'in', partner_ids and partner_ids[0] and partner_ids[0][2] or False)]" attrs="{'required': [('state', '=', 'selection')]}"/>
<field name="partner_ids" nolabel="1">
<tree string="Partners">
<field name="id" />
<field name="name" />
<field name="email" />
<field name="is_company" />
<field name="vat" />
<field name="country_id" />
</tree>
</field>
</group>
</sheet>
</form>
</field>
</record>
<act_window id="action_partner_merge" res_model="base.partner.merge.automatic.wizard" src_model="res.partner"
target="new" multi="True" key2="client_action_multi" view_mode="form" name="Automatic Merge"/>
</data>
</openerp>

3
base_partner_merge/security/ir.model.access.csv

@ -0,0 +1,3 @@
"id","name","model_id:id","group_id:id","perm_read","perm_write","perm_create","perm_unlink"
"access_base_partner_merge_line_manager","base_partner_merge_line.manager","model_base_partner_merge_line","base.group_system",1,1,1,1
"access_base_partner_merge_manager","base_partner_merge.manager","model_base_partner_merge_automatic_wizard","base.group_system",1,1,1,1

123
base_partner_merge/validate_email.py

@ -0,0 +1,123 @@
# RFC 2822 - style email validation for Python
# (c) 2012 Syrus Akbary <me@syrusakbary.com>
# Extended from (c) 2011 Noel Bush <noel@aitools.org>
# for support of mx and user check
# This code is made available to you under the GNU LGPL v3.
#
# This module provides a single method, valid_email_address(),
# which returns True or False to indicate whether a given address
# is valid according to the 'addr-spec' part of the specification
# given in RFC 2822. Ideally, we would like to find this
# in some other library, already thoroughly tested and well-
# maintained. The standard Python library email.utils
# contains a parse_addr() function, but it is not sufficient
# to detect many malformed addresses.
#
# This implementation aims to be faithful to the RFC, with the
# exception of a circular definition (see comments below), and
# with the omission of the pattern components marked as "obsolete".
import re
import smtplib
import socket
try:
import DNS
ServerError = DNS.ServerError
except:
DNS = None
class ServerError(Exception): pass
# All we are really doing is comparing the input string to one
# gigantic regular expression. But building that regexp, and
# ensuring its correctness, is made much easier by assembling it
# from the "tokens" defined by the RFC. Each of these tokens is
# tested in the accompanying unit test file.
#
# The section of RFC 2822 from which each pattern component is
# derived is given in an accompanying comment.
#
# (To make things simple, every string below is given as 'raw',
# even when it's not strictly necessary. This way we don't forget
# when it is necessary.)
#
WSP = r'[ \t]' # see 2.2.2. Structured Header Field Bodies
CRLF = r'(?:\r\n)' # see 2.2.3. Long Header Fields
NO_WS_CTL = r'\x01-\x08\x0b\x0c\x0f-\x1f\x7f' # see 3.2.1. Primitive Tokens
QUOTED_PAIR = r'(?:\\.)' # see 3.2.2. Quoted characters
FWS = r'(?:(?:' + WSP + r'*' + CRLF + r')?' + \
WSP + r'+)' # see 3.2.3. Folding white space and comments
CTEXT = r'[' + NO_WS_CTL + \
r'\x21-\x27\x2a-\x5b\x5d-\x7e]' # see 3.2.3
CCONTENT = r'(?:' + CTEXT + r'|' + \
QUOTED_PAIR + r')' # see 3.2.3 (NB: The RFC includes COMMENT here
# as well, but that would be circular.)
COMMENT = r'\((?:' + FWS + r'?' + CCONTENT + \
r')*' + FWS + r'?\)' # see 3.2.3
CFWS = r'(?:' + FWS + r'?' + COMMENT + ')*(?:' + \
FWS + '?' + COMMENT + '|' + FWS + ')' # see 3.2.3
ATEXT = r'[\w!#$%&\'\*\+\-/=\?\^`\{\|\}~]' # see 3.2.4. Atom
ATOM = CFWS + r'?' + ATEXT + r'+' + CFWS + r'?' # see 3.2.4
DOT_ATOM_TEXT = ATEXT + r'+(?:\.' + ATEXT + r'+)*' # see 3.2.4
DOT_ATOM = CFWS + r'?' + DOT_ATOM_TEXT + CFWS + r'?' # see 3.2.4
QTEXT = r'[' + NO_WS_CTL + \
r'\x21\x23-\x5b\x5d-\x7e]' # see 3.2.5. Quoted strings
QCONTENT = r'(?:' + QTEXT + r'|' + \
QUOTED_PAIR + r')' # see 3.2.5
QUOTED_STRING = CFWS + r'?' + r'"(?:' + FWS + \
r'?' + QCONTENT + r')*' + FWS + \
r'?' + r'"' + CFWS + r'?'
LOCAL_PART = r'(?:' + DOT_ATOM + r'|' + \
QUOTED_STRING + r')' # see 3.4.1. Addr-spec specification
DTEXT = r'[' + NO_WS_CTL + r'\x21-\x5a\x5e-\x7e]' # see 3.4.1
DCONTENT = r'(?:' + DTEXT + r'|' + \
QUOTED_PAIR + r')' # see 3.4.1
DOMAIN_LITERAL = CFWS + r'?' + r'\[' + \
r'(?:' + FWS + r'?' + DCONTENT + \
r')*' + FWS + r'?\]' + CFWS + r'?' # see 3.4.1
DOMAIN = r'(?:' + DOT_ATOM + r'|' + \
DOMAIN_LITERAL + r')' # see 3.4.1
ADDR_SPEC = LOCAL_PART + r'@' + DOMAIN # see 3.4.1
# A valid address will match exactly the 3.4.1 addr-spec.
VALID_ADDRESS_REGEXP = '^' + ADDR_SPEC + '$'
def validate_email(email, check_mx=False,verify=False):
"""Indicate whether the given string is a valid email address
according to the 'addr-spec' portion of RFC 2822 (see section
3.4.1). Parts of the spec that are marked obsolete are *not*
included in this test, and certain arcane constructions that
depend on circular definitions in the spec may not pass, but in
general this should correctly identify any email address likely
to be in use as of 2011."""
try:
assert re.match(VALID_ADDRESS_REGEXP, email) is not None
check_mx |= verify
if check_mx:
if not DNS: raise Exception('For check the mx records or check if the email exists you must have installed pyDNS python package')
DNS.DiscoverNameServers()
hostname = email[email.find('@')+1:]
mx_hosts = DNS.mxlookup(hostname)
for mx in mx_hosts:
try:
smtp = smtplib.SMTP()
smtp.connect(mx[1])
if not verify: return True
status, _ = smtp.helo()
if status != 250: continue
smtp.mail('')
status, _ = smtp.rcpt(email)
if status != 250: return False
break
except smtplib.SMTPServerDisconnected: #Server not permits verify user
break
except smtplib.SMTPConnectError:
continue
except (AssertionError, ServerError):
return False
return True
# import sys
# sys.modules[__name__],sys.modules['validate_email_module'] = validate_email,sys.modules[__name__]
# from validate_email_module import *

21
portal_partner_merge/__init__.py

@ -0,0 +1,21 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# Author: Yannick Vaucher
# Copyright 2013 Camptocamp SA
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
from . import wizard

36
portal_partner_merge/__openerp__.py

@ -0,0 +1,36 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# Author: Yannick Vaucher
# Copyright 2013 Camptocamp SA
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
{'name' : 'Portal Partner Merge',
'version' : '1.0',
'category': 'Hidden',
'description': """
Link module for base_partner_merge which extract portal dependency
""",
'author' : 'Camptocamp',
'maintainer': 'Camptocamp',
'website': 'http://www.camptocamp.com/',
'depends' : ['portal', 'base_partner_merge'],
'data': [],
'test': [],
'installable': True,
'auto_install': True,
'application': False,
}

26
portal_partner_merge/wizard/__init__.py

@ -0,0 +1,26 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# OpenERP, Open Source Business Applications
# Copyright (c) 2013 OpenERP S.A. <http://openerp.com>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
from . import portal_wizard
# vim:expandtab:smartindent:tabstop=4:softtabstop=4:shiftwidth=4:

39
portal_partner_merge/wizard/portal_wizard.py

@ -0,0 +1,39 @@
# -*- coding: utf-8 -*-
##############################################################################
#
# OpenERP, Open Source Business Applications
# Copyright (c) 2013 OpenERP S.A. <http://openerp.com>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU Affero General Public License as
# published by the Free Software Foundation, either version 3 of the
# License, or (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU Affero General Public License for more details.
#
# You should have received a copy of the GNU Affero General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.
#
##############################################################################
from openerp.osv import osv
from openerp.tools.translate import _
class wizard_user(osv.TransientModel):
_inherit = 'portal.wizard.user'
def get_error_messages(self, cr, uid, ids, context=None):
error_msg = super(wizard_user, self
).get_error_messages(cr, uid, ids, context=context)
if error_msg:
error_msg[-1] = '%s\n%s' % (
error_msg[-1],
_("- Merge existing contacts together using the Automatic "
"Merge wizard, available in the More menu after selecting "
"several contacts in the Customers list"))
return error_msg
# vim:expandtab:smartindent:tabstop=4:softtabstop=4:shiftwidth=4:
Loading…
Cancel
Save