flectra/addons/l10n_lu/scripts/tax2csv.py

184 lines
7.6 KiB
Python
Raw Permalink Normal View History

from collections import OrderedDict
import xlrd
2018-01-16 11:34:37 +01:00
from flectra.tools import pycompat
def _is_true(s):
return s not in ('F', 'False', 0, '', None, False)
class LuxTaxGenerator:
def __init__(self, filename):
self.workbook = xlrd.open_workbook('tax.xls')
self.sheet_info = \
self.workbook.sheet_by_name('INFO')
self.sheet_taxes = \
self.workbook.sheet_by_name('TAXES')
self.sheet_tax_codes = \
self.workbook.sheet_by_name('TAX.CODES')
self.sheet_fiscal_pos_map = \
self.workbook.sheet_by_name('FISCAL.POSITION.MAPPINGS')
self.suffix = self.sheet_info.cell_value(4, 2)
def iter_tax_codes(self):
keys = [c.value for c in self.sheet_tax_codes.row(0)]
yield keys
for i in range(1, self.sheet_tax_codes.nrows):
row = (c.value for c in self.sheet_tax_codes.row(i))
d = OrderedDict(pycompat.izip(keys, row))
d['sign'] = int(d['sign'])
d['sequence'] = int(d['sequence'])
yield d
def iter_taxes(self):
keys = [c.value for c in self.sheet_taxes.row(0)]
yield keys
for i in range(1, self.sheet_taxes.nrows):
row = (c.value for c in self.sheet_taxes.row(i))
yield OrderedDict(pycompat.izip(keys, row))
def iter_fiscal_pos_map(self):
keys = [c.value for c in self.sheet_fiscal_pos_map.row(0)]
yield keys
for i in range(1, self.sheet_fiscal_pos_map.nrows):
row = (c.value for c in self.sheet_fiscal_pos_map.row(i))
yield OrderedDict(pycompat.izip(keys, row))
def tax_codes_to_csv(self):
writer = pycompat.csv_writer(open('account.tax.code.template-%s.csv' %
self.suffix, 'wb'))
tax_codes_iterator = self.iter_tax_codes()
keys = next(tax_codes_iterator)
writer.writerow(keys)
# write structure tax codes
tax_codes = {} # code: id
for row in tax_codes_iterator:
tax_code = row['code']
if tax_code in tax_codes:
raise RuntimeError('duplicate tax code %s' % tax_code)
tax_codes[tax_code] = row['id']
writer.writerow([pycompat.to_text(v) for v in row.values()])
# read taxes and add leaf tax codes
new_tax_codes = {} # id: parent_code
def add_new_tax_code(tax_code_id, new_name, new_parent_code):
if not tax_code_id:
return
name, parent_code = new_tax_codes.get(tax_code_id, (None, None))
if parent_code and parent_code != new_parent_code:
raise RuntimeError('tax code "%s" already exist with '
'parent %s while trying to add it with '
'parent %s' %
(tax_code_id, parent_code, new_parent_code))
else:
new_tax_codes[tax_code_id] = (new_name, new_parent_code)
taxes_iterator = self.iter_taxes()
next(taxes_iterator)
for row in taxes_iterator:
if not _is_true(row['active']):
continue
if row['child_depend'] and row['amount'] != 1:
raise RuntimeError('amount must be one if child_depend '
'for %s' % row['id'])
# base parent
base_code = row['BASE_CODE']
if not base_code or base_code == '/':
base_code = 'NA'
if base_code not in tax_codes:
raise RuntimeError('undefined tax code %s' % base_code)
if base_code != 'NA':
if row['child_depend']:
raise RuntimeError('base code specified '
'with child_depend for %s' % row['id'])
if not row['child_depend']:
# ... in lux, we have the same code for invoice and refund
if base_code != 'NA':
assert row['base_code_id:id'], 'missing base_code_id for %s' % row['id']
assert row['ref_base_code_id:id'] == row['base_code_id:id']
add_new_tax_code(row['base_code_id:id'],
'Base - ' + row['name'],
base_code)
# tax parent
tax_code = row['TAX_CODE']
if not tax_code or tax_code == '/':
tax_code = 'NA'
if tax_code not in tax_codes:
raise RuntimeError('undefined tax code %s' % tax_code)
if tax_code == 'NA':
if row['amount'] and not row['child_depend']:
raise RuntimeError('TAX_CODE not specified '
'for non-zero tax %s' % row['id'])
if row['tax_code_id:id']:
raise RuntimeError('tax_code_id specified '
'for tax %s' % row['id'])
else:
if row['child_depend']:
raise RuntimeError('TAX_CODE specified '
'with child_depend for %s' % row['id'])
if not row['amount']:
raise RuntimeError('TAX_CODE specified '
'for zero tax %s' % row['id'])
if not row['tax_code_id:id']:
raise RuntimeError('tax_code_id not specified '
'for tax %s' % row['id'])
if not row['child_depend'] and row['amount']:
# ... in lux, we have the same code for invoice and refund
assert row['tax_code_id:id'], 'missing tax_code_id for %s' % row['id']
assert row['ref_tax_code_id:id'] == row['tax_code_id:id']
add_new_tax_code(row['tax_code_id:id'],
'Taxe - ' + row['name'],
tax_code)
for tax_code_id in sorted(new_tax_codes):
name, parent_code = new_tax_codes[tax_code_id]
writer.writerow([
tax_code_id,
u'lu_tct_m' + parent_code,
tax_code_id.replace('lu_tax_code_template_', u''),
u'1',
u'',
pycompat.to_text(name),
u''
])
def taxes_to_csv(self):
writer = pycompat.csv_writer(open('account.tax.template-%s.csv' %
self.suffix, 'wb'))
taxes_iterator = self.iter_taxes()
keys = next(taxes_iterator)
writer.writerow(keys[3:] + ['sequence'])
seq = 100
for row in sorted(taxes_iterator, key=lambda r: r['description']):
if not _is_true(row['active']):
continue
seq += 1
if row['parent_id:id']:
cur_seq = seq + 1000
else:
cur_seq = seq
writer.writerow([
pycompat.to_text(v)
for v in list(row.values())[3:]
] + [cur_seq])
def fiscal_pos_map_to_csv(self):
writer = pycompat.csv_writer(open('account.fiscal.'
'position.tax.template-%s.csv' %
self.suffix, 'wb'))
fiscal_pos_map_iterator = self.iter_fiscal_pos_map()
keys = next(fiscal_pos_map_iterator)
writer.writerow(keys)
for row in fiscal_pos_map_iterator:
writer.writerow([pycompat.to_text(s) for s in row.values()])
if __name__ == '__main__':
o = LuxTaxGenerator('tax.xls')
o.tax_codes_to_csv()
o.taxes_to_csv()
o.fiscal_pos_map_to_csv()