575 lines
22 KiB
Python
575 lines
22 KiB
Python
# -*- coding: utf-8 -*-
|
|
# Part of Odoo, Flectra. See LICENSE file for full copyright and licensing details.
|
|
import base64
|
|
import io
|
|
import unittest
|
|
|
|
from flectra.tests.common import TransactionCase, can_import
|
|
from flectra.modules.module import get_module_resource
|
|
from flectra.tools import mute_logger, pycompat
|
|
|
|
ID_FIELD = {
|
|
'id': 'id',
|
|
'name': 'id',
|
|
'string': "External ID",
|
|
|
|
'required': False,
|
|
'fields': [],
|
|
'type': 'id',
|
|
}
|
|
|
|
|
|
def make_field(name='value', string='Value', required=False, fields=[], field_type='id'):
|
|
return [
|
|
ID_FIELD,
|
|
{'id': name, 'name': name, 'string': string, 'required': required, 'fields': fields, 'type': field_type},
|
|
]
|
|
|
|
|
|
def sorted_fields(fields):
|
|
""" recursively sort field lists to ease comparison """
|
|
recursed = [dict(field, fields=sorted_fields(field['fields'])) for field in fields]
|
|
return sorted(recursed, key=lambda field: field['id'])
|
|
|
|
|
|
class BaseImportCase(TransactionCase):
|
|
|
|
def assertEqualFields(self, fields1, fields2):
|
|
self.assertEqual(sorted_fields(fields1), sorted_fields(fields2))
|
|
|
|
|
|
class TestBasicFields(BaseImportCase):
|
|
|
|
def get_fields(self, field):
|
|
return self.env['base_import.import'].get_fields('base_import.tests.models.' + field)
|
|
|
|
def test_base(self):
|
|
""" A basic field is not required """
|
|
self.assertEqualFields(self.get_fields('char'), make_field(field_type='char'))
|
|
|
|
def test_required(self):
|
|
""" Required fields should be flagged (so they can be fill-required) """
|
|
self.assertEqualFields(self.get_fields('char.required'), make_field(required=True, field_type='char'))
|
|
|
|
def test_readonly(self):
|
|
""" Readonly fields should be filtered out"""
|
|
self.assertEqualFields(self.get_fields('char.readonly'), [ID_FIELD])
|
|
|
|
def test_readonly_states(self):
|
|
""" Readonly fields with states should not be filtered out"""
|
|
self.assertEqualFields(self.get_fields('char.states'), make_field(field_type='char'))
|
|
|
|
def test_readonly_states_noreadonly(self):
|
|
""" Readonly fields with states having nothing to do with
|
|
readonly should still be filtered out"""
|
|
self.assertEqualFields(self.get_fields('char.noreadonly'), [ID_FIELD])
|
|
|
|
def test_readonly_states_stillreadonly(self):
|
|
""" Readonly fields with readonly states leaving them readonly
|
|
always... filtered out"""
|
|
self.assertEqualFields(self.get_fields('char.stillreadonly'), [ID_FIELD])
|
|
|
|
def test_m2o(self):
|
|
""" M2O fields should allow import of themselves (name_get),
|
|
their id and their xid"""
|
|
self.assertEqualFields(self.get_fields('m2o'), make_field(field_type='many2one', fields=[
|
|
{'id': 'value', 'name': 'id', 'string': 'External ID', 'required': False, 'fields': [], 'type': 'id'},
|
|
{'id': 'value', 'name': '.id', 'string': 'Database ID', 'required': False, 'fields': [], 'type': 'id'},
|
|
]))
|
|
|
|
def test_m2o_required(self):
|
|
""" If an m2o field is required, its three sub-fields are
|
|
required as well (the client has to handle that: requiredness
|
|
is id-based)
|
|
"""
|
|
self.assertEqualFields(self.get_fields('m2o.required'), make_field(field_type='many2one', required=True, fields=[
|
|
{'id': 'value', 'name': 'id', 'string': 'External ID', 'required': True, 'fields': [], 'type': 'id'},
|
|
{'id': 'value', 'name': '.id', 'string': 'Database ID', 'required': True, 'fields': [], 'type': 'id'},
|
|
]))
|
|
|
|
|
|
class TestO2M(BaseImportCase):
|
|
|
|
def get_fields(self, field):
|
|
return self.env['base_import.import'].get_fields('base_import.tests.models.' + field)
|
|
|
|
def test_shallow(self):
|
|
self.assertEqualFields(self.get_fields('o2m'), make_field(field_type='one2many', fields=[
|
|
ID_FIELD,
|
|
# FIXME: should reverse field be ignored?
|
|
{'id': 'parent_id', 'name': 'parent_id', 'string': 'Parent', 'type': 'many2one', 'required': False, 'fields': [
|
|
{'id': 'parent_id', 'name': 'id', 'string': 'External ID', 'required': False, 'fields': [], 'type': 'id'},
|
|
{'id': 'parent_id', 'name': '.id', 'string': 'Database ID', 'required': False, 'fields': [], 'type': 'id'},
|
|
]},
|
|
{'id': 'value', 'name': 'value', 'string': 'Value', 'required': False, 'fields': [], 'type': 'integer'},
|
|
]))
|
|
|
|
|
|
class TestMatchHeadersSingle(TransactionCase):
|
|
|
|
def test_match_by_name(self):
|
|
match = self.env['base_import.import']._match_header('f0', [{'name': 'f0'}], {})
|
|
self.assertEqual(match, [{'name': 'f0'}])
|
|
|
|
def test_match_by_string(self):
|
|
match = self.env['base_import.import']._match_header('some field', [{'name': 'bob', 'string': "Some Field"}], {})
|
|
self.assertEqual(match, [{'name': 'bob', 'string': "Some Field"}])
|
|
|
|
def test_nomatch(self):
|
|
match = self.env['base_import.import']._match_header('should not be', [{'name': 'bob', 'string': "wheee"}], {})
|
|
self.assertEqual(match, [])
|
|
|
|
def test_recursive_match(self):
|
|
f = {
|
|
'name': 'f0',
|
|
'string': "My Field",
|
|
'fields': [
|
|
{'name': 'f0', 'string': "Sub field 0", 'fields': []},
|
|
{'name': 'f1', 'string': "Sub field 2", 'fields': []},
|
|
]
|
|
}
|
|
match = self.env['base_import.import']._match_header('f0/f1', [f], {})
|
|
self.assertEqual(match, [f, f['fields'][1]])
|
|
|
|
def test_recursive_nomatch(self):
|
|
""" Match first level, fail to match second level
|
|
"""
|
|
f = {
|
|
'name': 'f0',
|
|
'string': "My Field",
|
|
'fields': [
|
|
{'name': 'f0', 'string': "Sub field 0", 'fields': []},
|
|
{'name': 'f1', 'string': "Sub field 2", 'fields': []},
|
|
]
|
|
}
|
|
match = self.env['base_import.import']._match_header('f0/f2', [f], {})
|
|
self.assertEqual(match, [])
|
|
|
|
|
|
class TestMatchHeadersMultiple(TransactionCase):
|
|
|
|
def test_noheaders(self):
|
|
self.assertEqual(
|
|
self.env['base_import.import']._match_headers([], [], {}), ([], {})
|
|
)
|
|
|
|
def test_nomatch(self):
|
|
self.assertEqual(
|
|
self.env['base_import.import']._match_headers(
|
|
iter([
|
|
['foo', 'bar', 'baz', 'qux'],
|
|
['v1', 'v2', 'v3', 'v4'],
|
|
]),
|
|
[],
|
|
{'headers': True}),
|
|
(
|
|
['foo', 'bar', 'baz', 'qux'],
|
|
dict.fromkeys(range(4))
|
|
)
|
|
)
|
|
|
|
def test_mixed(self):
|
|
self.assertEqual(
|
|
self.env['base_import.import']._match_headers(
|
|
iter(['foo bar baz qux/corge'.split()]),
|
|
[
|
|
{'name': 'bar', 'string': 'Bar'},
|
|
{'name': 'bob', 'string': 'Baz'},
|
|
{'name': 'qux', 'string': 'Qux', 'fields': [
|
|
{'name': 'corge', 'fields': []},
|
|
]}
|
|
],
|
|
{'headers': True}),
|
|
(['foo', 'bar', 'baz', 'qux/corge'], {
|
|
0: None,
|
|
1: ['bar'],
|
|
2: ['bob'],
|
|
3: ['qux', 'corge'],
|
|
})
|
|
)
|
|
|
|
|
|
class TestPreview(TransactionCase):
|
|
|
|
def make_import(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'res.users',
|
|
'file': u"로그인,언어\nbob,1\n".encode('euc_kr'),
|
|
'file_type': 'text/csv',
|
|
'file_name': 'kr_data.csv',
|
|
})
|
|
return import_wizard
|
|
|
|
@mute_logger('flectra.addons.base_import.models.base_import')
|
|
def test_encoding(self):
|
|
import_wizard = self.make_import()
|
|
result = import_wizard.parse_preview({
|
|
'quoting': '"',
|
|
'separator': ',',
|
|
})
|
|
self.assertTrue('error' in result)
|
|
|
|
@mute_logger('flectra.addons.base_import.models.base_import')
|
|
def test_csv_errors(self):
|
|
import_wizard = self.make_import()
|
|
|
|
result = import_wizard.parse_preview({
|
|
'quoting': 'foo',
|
|
'separator': ',',
|
|
'encoding': 'euc_kr',
|
|
})
|
|
self.assertTrue('error' in result)
|
|
|
|
result = import_wizard.parse_preview({
|
|
'quoting': '"',
|
|
'separator': 'bob',
|
|
'encoding': 'euc_kr',
|
|
})
|
|
self.assertTrue('error' in result)
|
|
|
|
def test_csv_success(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n'
|
|
b'bar,3,4\n'
|
|
b'qux,5,6\n',
|
|
'file_type': 'text/csv'
|
|
})
|
|
|
|
result = import_wizard.parse_preview({
|
|
'quoting': '"',
|
|
'separator': ',',
|
|
'headers': True,
|
|
})
|
|
self.assertIsNone(result.get('error'))
|
|
self.assertEqual(result['matches'], {0: ['name'], 1: ['somevalue'], 2: None})
|
|
self.assertEqual(result['headers'], ['name', 'Some Value', 'Counter'])
|
|
# Order depends on iteration order of fields_get
|
|
self.assertItemsEqual(result['fields'], [
|
|
ID_FIELD,
|
|
{'id': 'name', 'name': 'name', 'string': 'Name', 'required': False, 'fields': [], 'type': 'char'},
|
|
{'id': 'somevalue', 'name': 'somevalue', 'string': 'Some Value', 'required': True, 'fields': [], 'type': 'integer'},
|
|
{'id': 'othervalue', 'name': 'othervalue', 'string': 'Other Variable', 'required': False, 'fields': [], 'type': 'integer'},
|
|
])
|
|
self.assertEqual(result['preview'], [
|
|
['foo', '1', '2'],
|
|
['bar', '3', '4'],
|
|
['qux', '5', '6'],
|
|
])
|
|
# Ensure we only have the response fields we expect
|
|
self.assertItemsEqual(list(result), ['matches', 'headers', 'fields', 'preview', 'headers_type', 'options', 'advanced_mode', 'debug'])
|
|
|
|
@unittest.skipUnless(can_import('xlrd'), "XLRD module not available")
|
|
def test_xls_success(self):
|
|
xls_file_path = get_module_resource('base_import', 'tests', 'test.xls')
|
|
file_content = open(xls_file_path, 'rb').read()
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': file_content,
|
|
'file_type': 'application/vnd.ms-excel'
|
|
})
|
|
|
|
result = import_wizard.parse_preview({
|
|
'headers': True,
|
|
})
|
|
self.assertIsNone(result.get('error'))
|
|
self.assertEqual(result['matches'], {0: ['name'], 1: ['somevalue'], 2: None})
|
|
self.assertEqual(result['headers'], ['name', 'Some Value', 'Counter'])
|
|
self.assertItemsEqual(result['fields'], [
|
|
ID_FIELD,
|
|
{'id': 'name', 'name': 'name', 'string': 'Name', 'required': False, 'fields': [], 'type': 'char'},
|
|
{'id': 'somevalue', 'name': 'somevalue', 'string': 'Some Value', 'required': True, 'fields': [], 'type': 'integer'},
|
|
{'id': 'othervalue', 'name': 'othervalue', 'string': 'Other Variable', 'required': False, 'fields': [], 'type': 'integer'},
|
|
])
|
|
self.assertEqual(result['preview'], [
|
|
['foo', '1', '2'],
|
|
['bar', '3', '4'],
|
|
['qux', '5', '6'],
|
|
])
|
|
# Ensure we only have the response fields we expect
|
|
self.assertItemsEqual(list(result), ['matches', 'headers', 'fields', 'preview', 'headers_type', 'options', 'advanced_mode', 'debug'])
|
|
|
|
@unittest.skipUnless(can_import('xlrd.xlsx'), "XLRD/XLSX not available")
|
|
def test_xlsx_success(self):
|
|
xlsx_file_path = get_module_resource('base_import', 'tests', 'test.xlsx')
|
|
file_content = open(xlsx_file_path, 'rb').read()
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': file_content,
|
|
'file_type': 'application/vnd.openxmlformats-officedocument.spreadsheetml.sheet'
|
|
})
|
|
|
|
result = import_wizard.parse_preview({
|
|
'headers': True,
|
|
})
|
|
self.assertIsNone(result.get('error'))
|
|
self.assertEqual(result['matches'], {0: ['name'], 1: ['somevalue'], 2: None})
|
|
self.assertEqual(result['headers'], ['name', 'Some Value', 'Counter'])
|
|
self.assertItemsEqual(result['fields'], [
|
|
ID_FIELD,
|
|
{'id': 'name', 'name': 'name', 'string': 'Name', 'required': False, 'fields': [], 'type': 'char'},
|
|
{'id': 'somevalue', 'name': 'somevalue', 'string': 'Some Value', 'required': True, 'fields': [], 'type': 'integer'},
|
|
{'id': 'othervalue', 'name': 'othervalue', 'string': 'Other Variable', 'required': False, 'fields': [], 'type': 'integer'},
|
|
])
|
|
self.assertEqual(result['preview'], [
|
|
['foo', '1', '2'],
|
|
['bar', '3', '4'],
|
|
['qux', '5', '6'],
|
|
])
|
|
# Ensure we only have the response fields we expect
|
|
self.assertItemsEqual(list(result), ['matches', 'headers', 'fields', 'preview', 'headers_type', 'options','advanced_mode', 'debug'])
|
|
|
|
@unittest.skipUnless(can_import('odf'), "ODFPY not available")
|
|
def test_ods_success(self):
|
|
ods_file_path = get_module_resource('base_import', 'tests', 'test.ods')
|
|
file_content = open(ods_file_path, 'rb').read()
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': file_content,
|
|
'file_type': 'application/vnd.oasis.opendocument.spreadsheet'
|
|
})
|
|
|
|
result = import_wizard.parse_preview({
|
|
'headers': True,
|
|
})
|
|
self.assertIsNone(result.get('error'))
|
|
self.assertEqual(result['matches'], {0: ['name'], 1: ['somevalue'], 2: None})
|
|
self.assertEqual(result['headers'], ['name', 'Some Value', 'Counter'])
|
|
self.assertItemsEqual(result['fields'], [
|
|
ID_FIELD,
|
|
{'id': 'name', 'name': 'name', 'string': 'Name', 'required': False, 'fields': [], 'type': 'char'},
|
|
{'id': 'somevalue', 'name': 'somevalue', 'string': 'Some Value', 'required': True, 'fields': [], 'type': 'integer'},
|
|
{'id': 'othervalue', 'name': 'othervalue', 'string': 'Other Variable', 'required': False, 'fields': [], 'type': 'integer'},
|
|
])
|
|
self.assertEqual(result['preview'], [
|
|
['foo', '1', '2'],
|
|
['bar', '3', '4'],
|
|
['aux', '5', '6'],
|
|
])
|
|
# Ensure we only have the response fields we expect
|
|
self.assertItemsEqual(list(result), ['matches', 'headers', 'fields', 'preview', 'headers_type', 'options', 'advanced_mode', 'debug'])
|
|
|
|
|
|
class test_convert_import_data(TransactionCase):
|
|
""" Tests conversion of base_import.import input into data which
|
|
can be fed to Model.load
|
|
"""
|
|
def test_all(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n'
|
|
b'bar,3,4\n'
|
|
b'qux,5,6\n',
|
|
'file_type': 'text/csv'
|
|
|
|
})
|
|
data, fields = import_wizard._convert_import_data(
|
|
['name', 'somevalue', 'othervalue'],
|
|
{'quoting': '"', 'separator': ',', 'headers': True}
|
|
)
|
|
|
|
self.assertItemsEqual(fields, ['name', 'somevalue', 'othervalue'])
|
|
self.assertItemsEqual(data, [
|
|
['foo', '1', '2'],
|
|
['bar', '3', '4'],
|
|
['qux', '5', '6'],
|
|
])
|
|
|
|
def test_date_fields(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'res.partner',
|
|
'file': u'name,date,create_date\n'
|
|
u'"foo","2013年07月18日","2016-10-12 06:06"\n'.encode('utf-8'),
|
|
'file_type': 'text/csv'
|
|
|
|
})
|
|
|
|
results = import_wizard.do(
|
|
['name', 'date', 'create_date'],
|
|
{
|
|
'date_format': '%Y年%m月%d日',
|
|
'datetime_format': '%Y-%m-%d %H:%M',
|
|
'quoting': '"',
|
|
'separator': ',',
|
|
'headers': True
|
|
}
|
|
)
|
|
|
|
# if results empty, no errors
|
|
self.assertItemsEqual(results, [])
|
|
|
|
def test_parse_relational_fields(self):
|
|
""" Ensure that relational fields float and date are correctly
|
|
parsed during the import call.
|
|
"""
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'res.partner',
|
|
'file': u'name,parent_id/id,parent_id/date,parent_id/credit_limit\n'
|
|
u'"foo","__export__.res_partner_1","2017年10月12日","5,69"\n'.encode('utf-8'),
|
|
'file_type': 'text/csv'
|
|
|
|
})
|
|
options = {
|
|
'date_format': '%Y年%m月%d日',
|
|
'quoting': '"',
|
|
'separator': ',',
|
|
'float_decimal_separator': ',',
|
|
'float_thousand_separator': '.',
|
|
'headers': True
|
|
}
|
|
data, import_fields = import_wizard._convert_import_data(
|
|
['name', 'parent_id/.id', 'parent_id/date', 'parent_id/credit_limit'],
|
|
options
|
|
)
|
|
result = import_wizard._parse_import_data(data, import_fields, options)
|
|
# Check if the data 5,69 as been correctly parsed.
|
|
self.assertEqual(float(result[0][-1]), 5.69)
|
|
self.assertEqual(str(result[0][-2]), '2017-10-12')
|
|
|
|
def test_filtered(self):
|
|
""" If ``False`` is provided as field mapping for a column,
|
|
that column should be removed from importable data
|
|
"""
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n'
|
|
b'bar,3,4\n'
|
|
b'qux,5,6\n',
|
|
'file_type': 'text/csv'
|
|
})
|
|
data, fields = import_wizard._convert_import_data(
|
|
['name', False, 'othervalue'],
|
|
{'quoting': '"', 'separator': ',', 'headers': True}
|
|
)
|
|
|
|
self.assertItemsEqual(fields, ['name', 'othervalue'])
|
|
self.assertItemsEqual(data, [
|
|
['foo', '2'],
|
|
['bar', '4'],
|
|
['qux', '6'],
|
|
])
|
|
|
|
def test_norow(self):
|
|
""" If a row is composed only of empty values (due to having
|
|
filtered out non-empty values from it), it should be removed
|
|
"""
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n'
|
|
b',3,\n'
|
|
b',5,6\n',
|
|
'file_type': 'text/csv'
|
|
})
|
|
data, fields = import_wizard._convert_import_data(
|
|
['name', False, 'othervalue'],
|
|
{'quoting': '"', 'separator': ',', 'headers': True}
|
|
)
|
|
|
|
self.assertItemsEqual(fields, ['name', 'othervalue'])
|
|
self.assertItemsEqual(data, [
|
|
['foo', '2'],
|
|
['', '6'],
|
|
])
|
|
|
|
def test_empty_rows(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value\n'
|
|
b'foo,1\n'
|
|
b'\n'
|
|
b'bar,2\n'
|
|
b' \n'
|
|
b'\t \n',
|
|
'file_type': 'text/csv'
|
|
})
|
|
data, fields = import_wizard._convert_import_data(
|
|
['name', 'somevalue'],
|
|
{'quoting': '"', 'separator': ',', 'headers': True}
|
|
)
|
|
|
|
self.assertItemsEqual(fields, ['name', 'somevalue'])
|
|
self.assertItemsEqual(data, [
|
|
['foo', '1'],
|
|
['bar', '2'],
|
|
])
|
|
|
|
def test_nofield(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n',
|
|
'file_type': 'text/csv'
|
|
|
|
})
|
|
self.assertRaises(ValueError, import_wizard._convert_import_data, [], {'quoting': '"', 'separator': ',', 'headers': True})
|
|
|
|
def test_falsefields(self):
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': b'name,Some Value,Counter\n'
|
|
b'foo,1,2\n',
|
|
'file_type': 'text/csv'
|
|
})
|
|
|
|
self.assertRaises(
|
|
ValueError,
|
|
import_wizard._convert_import_data,
|
|
[False, False, False],
|
|
{'quoting': '"', 'separator': ',', 'headers': True})
|
|
|
|
def test_newline_import(self):
|
|
"""
|
|
Ensure importing keep newlines
|
|
"""
|
|
output = io.BytesIO()
|
|
writer = pycompat.csv_writer(output, quoting=1)
|
|
|
|
data_row = [u"\tfoo\n\tbar", u" \"hello\" \n\n 'world' "]
|
|
|
|
writer.writerow([u"name", u"Some Value"])
|
|
writer.writerow(data_row)
|
|
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'base_import.tests.models.preview',
|
|
'file': output.getvalue(),
|
|
'file_type': 'text/csv',
|
|
})
|
|
data, _ = import_wizard._convert_import_data(
|
|
['name', 'somevalue'],
|
|
{'quoting': '"', 'separator': ',', 'headers': True}
|
|
)
|
|
|
|
self.assertItemsEqual(data, [data_row])
|
|
|
|
|
|
class test_failures(TransactionCase):
|
|
def test_big_attachments(self):
|
|
"""
|
|
Ensure big fields (e.g. b64-encoded image data) can be imported and
|
|
we're not hitting limits of the default CSV parser config
|
|
"""
|
|
from PIL import Image
|
|
|
|
im = Image.new('RGB', (1920, 1080))
|
|
fout = io.BytesIO()
|
|
|
|
writer = pycompat.csv_writer(fout, dialect=None)
|
|
writer.writerows([
|
|
[u'name', u'db_datas'],
|
|
[u'foo', base64.b64encode(im.tobytes()).decode('ascii')]
|
|
])
|
|
|
|
import_wizard = self.env['base_import.import'].create({
|
|
'res_model': 'ir.attachment',
|
|
'file': fout.getvalue(),
|
|
'file_type': 'text/csv'
|
|
})
|
|
results = import_wizard.do(
|
|
['name', 'db_datas'],
|
|
{'headers': True, 'separator': ',', 'quoting': '"'})
|
|
self.assertFalse(results, "results should be empty on successful import")
|