|
|
# Copyright 2019 Ecosoft Co., Ltd (http://ecosoft.co.th/) # License AGPL-3.0 or later (https://www.gnu.org/licenses/agpl.html)
import base64 import uuid import xlrd import xlwt import time from io import BytesIO from . import common as co from ast import literal_eval from datetime import date, datetime as dt from odoo.tools.float_utils import float_compare from odoo import models, api, _ from odoo.exceptions import ValidationError from odoo.tools.safe_eval import safe_eval
class XLSXImport(models.AbstractModel): _name = 'xlsx.import' _description = 'Excel Import AbstractModel'
@api.model def get_eval_context(self, model=False, value=False): eval_context = {'float_compare': float_compare, 'time': time, 'datetime': dt, 'date': date, 'env': self.env, 'context': self._context, 'value': False, 'model': False, } if model: eval_context.update({'model': self.env[model]}) if value: if isinstance(value, str): # Remove non Ord 128 character value = ''.join([i if ord(i) < 128 else ' ' for i in value]) eval_context.update({'value': value}) return eval_context
@api.model def get_external_id(self, record): """ Get external ID of the record, if not already exists create one """ ModelData = self.env['ir.model.data'] xml_id = record.get_external_id() if not xml_id or (record.id in xml_id and xml_id[record.id] == ''): ModelData.create({'name': '%s_%s' % (record._table, record.id), 'module': 'excel_import_export', 'model': record._name, 'res_id': record.id, }) xml_id = record.get_external_id() return xml_id[record.id]
@api.model def _get_field_type(self, model, field): try: record = self.env[model].new() for f in field.split('/'): field_type = record._fields[f].type if field_type in ('one2many', 'many2many'): record = record[f] else: return field_type except Exception: raise ValidationError( _('Invalid declaration, %s has no valid field type') % field)
@api.model def _delete_record_data(self, record, data_dict): """ If no _NODEL_, delete existing lines before importing """ if not record or not data_dict: return try: for sheet_name in data_dict: worksheet = data_dict[sheet_name] line_fields = filter(lambda x: x != '_HEAD_', worksheet) for line_field in line_fields: if '_NODEL_' not in line_field: if line_field in record and record[line_field]: record[line_field].unlink() # Remove _NODEL_ from dict for s, sv in data_dict.items(): for f, fv in data_dict[s].items(): if '_NODEL_' in f: new_fv = data_dict[s].pop(f) data_dict[s][f.replace('_NODEL_', '')] = new_fv except Exception as e: raise ValidationError(_('Error deleting data\n%s') % e)
@api.model def _get_line_vals(self, st, worksheet, model, line_field): """ Get values of this field from excel sheet """ vals = {} for rc, columns in worksheet.get(line_field, {}).items(): if not isinstance(columns, list): columns = [columns] for field in columns: rc, key_eval_cond = co.get_field_condition(rc) x_field, val_eval_cond = co.get_field_condition(field) row, col = co.pos2idx(rc) out_field = '%s/%s' % (line_field, x_field) field_type = self._get_field_type(model, out_field) vals.update({out_field: []}) for idx in range(row, st.nrows): value = co._get_cell_value(st.cell(idx, col), field_type=field_type) eval_context = self.get_eval_context(model=model, value=value) if key_eval_cond: value = safe_eval(key_eval_cond, eval_context) if val_eval_cond: value = safe_eval(val_eval_cond, eval_context) vals[out_field].append(value) if not filter(lambda x: x != '', vals[out_field]): vals.pop(out_field) return vals
@api.model def _import_record_data(self, import_file, record, data_dict): """ From complex excel, create temp simple excel and do import """ if not data_dict: return try: header_fields = [] decoded_data = base64.decodestring(import_file) wb = xlrd.open_workbook(file_contents=decoded_data) col_idx = 0 out_wb = xlwt.Workbook() out_st = out_wb.add_sheet("Sheet 1") xml_id = record and self.get_external_id(record) or \ '%s.%s' % ('xls', uuid.uuid4()) out_st.write(0, 0, 'id') # id and xml_id on first column out_st.write(1, 0, xml_id) header_fields.append('id') col_idx += 1 model = record._name for sheet_name in data_dict: # For each Sheet worksheet = data_dict[sheet_name] st = False if isinstance(sheet_name, str): st = co.xlrd_get_sheet_by_name(wb, sheet_name) elif isinstance(sheet_name, int): st = wb.sheet_by_index(sheet_name - 1) if not st: raise ValidationError( _('Sheet %s not found') % sheet_name) # HEAD updates for rc, field in worksheet.get('_HEAD_', {}).items(): rc, key_eval_cond = co.get_field_condition(rc) field, val_eval_cond = co.get_field_condition(field) field_type = self._get_field_type(model, field) value = False try: row, col = co.pos2idx(rc) value = co._get_cell_value(st.cell(row, col), field_type=field_type) except Exception: pass eval_context = self.get_eval_context(model=model, value=value) if key_eval_cond: value = str(safe_eval(key_eval_cond, eval_context)) if val_eval_cond: value = str(safe_eval(val_eval_cond, eval_context)) out_st.write(0, col_idx, field) # Next Column out_st.write(1, col_idx, value) # Next Value header_fields.append(field) col_idx += 1 # Line Items line_fields = filter(lambda x: x != '_HEAD_', worksheet) for line_field in line_fields: vals = self._get_line_vals(st, worksheet, model, line_field) for field in vals: # Columns, i.e., line_ids/field_id out_st.write(0, col_idx, field) header_fields.append(field) # Data i = 1 for value in vals[field]: out_st.write(i, col_idx, value) i += 1 col_idx += 1 content = BytesIO() out_wb.save(content) content.seek(0) # Set index to 0, and start reading xls_file = content.read() # Do the import Import = self.env['base_import.import'] imp = Import.create({ 'res_model': model, 'file': xls_file, 'file_type': 'application/vnd.ms-excel', 'file_name': 'temp.xls', }) errors = imp.do( header_fields, header_fields, {'headers': True, 'advanced': True, 'keep_matches': False, 'encoding': '', 'separator': '', 'quoting': '"', 'date_format': '%Y-%m-%d', 'datetime_format': '%Y-%m-%d %H:%M:%S', 'float_thousand_separator': ',', 'float_decimal_separator': '.', 'fields': []}) if errors.get('messages'): message = _('Error importing data') messages = errors['messages'] if isinstance(messages, dict): message = messages['message'] if isinstance(messages, list): message = ', '.join([x['message'] for x in messages]) raise ValidationError(message.encode('utf-8')) return self.env.ref(xml_id) except xlrd.XLRDError: raise ValidationError( _('Invalid file style, only .xls or .xlsx file allowed')) except Exception as e: raise ValidationError(_('Error importing data\n%s') % e)
@api.model def _post_import_operation(self, record, operation): """ Run python code after import """ if not record or not operation: return try: if '${' in operation: code = (operation.split('${'))[1].split('}')[0] eval_context = {'object': record} safe_eval(code, eval_context) except Exception as e: raise ValidationError(_('Post import operation error\n%s') % e)
@api.model def import_xlsx(self, import_file, template, res_model=False, res_id=False): """
- If res_id = False, we want to create new document first - Delete fields' data according to data_dict['__IMPORT__'] - Import data from excel according to data_dict['__IMPORT__'] """
self = self.sudo() if res_model and template.res_model != res_model: raise ValidationError(_("Template's model mismatch")) record = self.env[template.res_model].browse(res_id) data_dict = literal_eval(template.instruction.strip()) if not data_dict.get('__IMPORT__'): raise ValidationError( _("No data_dict['__IMPORT__'] in template %s") % template.name) if record: # Delete existing data first self._delete_record_data(record, data_dict['__IMPORT__']) # Fill up record with data from excel sheets record = self._import_record_data(import_file, record, data_dict['__IMPORT__']) # Post Import Operation, i.e., cleanup some data if data_dict.get('__POST_IMPORT__', False): self._post_import_operation(record, data_dict['__POST_IMPORT__']) return record
|