|
|
# Copyright 2019 Ecosoft Co., Ltd (http://ecosoft.co.th/) # License AGPL-3.0 or later (https://www.gnu.org/licenses/agpl.html)
import os import logging import base64 from io import BytesIO import time from datetime import date, datetime as dt from odoo.tools.float_utils import float_compare from odoo import models, fields, api, _ from odoo.tools.safe_eval import safe_eval from odoo.exceptions import ValidationError from . import common as co
_logger = logging.getLogger(__name__) try: from openpyxl import load_workbook from openpyxl.utils.exceptions import IllegalCharacterError except ImportError: _logger.debug( 'Cannot import "openpyxl". Please make sure it is installed.')
class XLSXExport(models.AbstractModel): _name = 'xlsx.export' _description = 'Excel Export AbstractModel'
@api.model def get_eval_context(self, model, record, value): eval_context = {'float_compare': float_compare, 'time': time, 'datetime': dt, 'date': date, 'value': value, 'object': record, 'model': self.env[model], 'env': self.env, 'context': self._context, } return eval_context
@api.model def _get_line_vals(self, record, line_field, fields): """ Get values of this field from record set and return as dict of vals
- record: main object - line_field: rows object, i.e., line_ids - fields: fields in line_ids, i.e., partner_id.display_name """
line_field, max_row = co.get_line_max(line_field) line_field = line_field.replace('_CONT_', '') # Remove _CONT_ if any lines = record[line_field] if max_row > 0 and len(lines) > max_row: raise Exception( _('Records in %s exceed max records allowed') % line_field) vals = dict([(field, []) for field in fields]) # value and do_style # Get field condition & aggre function field_cond_dict = {} aggre_func_dict = {} field_style_dict = {} style_cond_dict = {} pair_fields = [] # I.e., ('debit${value and . or .}@{sum}', 'debit') for field in fields: temp_field, eval_cond = co.get_field_condition(field) eval_cond = eval_cond or 'value or ""' temp_field, field_style = co.get_field_style(temp_field) temp_field, style_cond = co.get_field_style_cond(temp_field) raw_field, aggre_func = co.get_field_aggregation(temp_field) # Dict of all special conditions field_cond_dict.update({field: eval_cond}) aggre_func_dict.update({field: aggre_func}) field_style_dict.update({field: field_style}) style_cond_dict.update({field: style_cond}) # -- pair_fields.append((field, raw_field)) for line in lines: for field in pair_fields: # (field, raw_field) value = self._get_field_data(field[1], line) eval_cond = field_cond_dict[field[0]] eval_context = \ self.get_eval_context(line._name, line, value) if eval_cond: value = safe_eval(eval_cond, eval_context) # style w/Cond takes priority style_cond = style_cond_dict[field[0]] style = self._eval_style_cond(line._name, line, value, style_cond) if style is None: style = False # No style elif style is False: style = field_style_dict[field[0]] # Use default style vals[field[0]].append((value, style)) return (vals, aggre_func_dict,)
@api.model def _eval_style_cond(self, model, record, value, style_cond): eval_context = self.get_eval_context(model, record, value) field = style_cond = style_cond or '#??' styles = {} for i in range(style_cond.count('#{')): i += 1 field, style = co.get_field_style(field) styles.update({i: style}) style_cond = style_cond.replace('#{%s}' % style, str(i)) if not styles: return False res = safe_eval(style_cond, eval_context) if res is None or res is False: return res return styles[res]
@api.model def _fill_workbook_data(self, workbook, record, data_dict): """ Fill data from record with style in data_dict to workbook """ if not record or not data_dict: return try: for sheet_name in data_dict: ws = data_dict[sheet_name] st = False if isinstance(sheet_name, str): st = co.openpyxl_get_sheet_by_name(workbook, sheet_name) elif isinstance(sheet_name, int): if sheet_name > len(workbook.worksheets): raise Exception(_('Not enough worksheets')) st = workbook.worksheets[sheet_name - 1] if not st: raise ValidationError( _('Sheet %s not found') % sheet_name) # Fill data, header and rows self._fill_head(ws, st, record) self._fill_lines(ws, st, record) except KeyError as e: raise ValidationError(_('Key Error\n%s') % e) except IllegalCharacterError as e: raise ValidationError( _('IllegalCharacterError\n' 'Some exporting data contain special character\n%s') % e) except Exception as e: raise ValidationError( _('Error filling data into Excel sheets\n%s') % e)
@api.model def _get_field_data(self, _field, _line): """ Get field data, and convert data type if needed """ if not _field: return None line_copy = _line for f in _field.split('.'): line_copy = line_copy[f] if isinstance(line_copy, str): line_copy = line_copy.encode('utf-8') return line_copy
@api.model def _fill_head(self, ws, st, record): for rc, field in ws.get('_HEAD_', {}).items(): tmp_field, eval_cond = co.get_field_condition(field) eval_cond = eval_cond or 'value or ""' tmp_field, field_style = co.get_field_style(tmp_field) tmp_field, style_cond = co.get_field_style_cond(tmp_field) value = tmp_field and self._get_field_data(tmp_field, record) # Eval eval_context = self.get_eval_context(record._name, record, value) if eval_cond: value = safe_eval(eval_cond, eval_context) if value is not None: st[rc] = value fc = not style_cond and True or \ safe_eval(style_cond, eval_context) if field_style and fc: # has style and pass style_cond styles = self.env['xlsx.styles'].get_openpyxl_styles() co.fill_cell_style(st[rc], field_style, styles)
@api.model def _fill_lines(self, ws, st, record): line_fields = list(ws) if '_HEAD_' in line_fields: line_fields.remove('_HEAD_') cont_row = 0 # last data row to continue for line_field in line_fields: fields = ws.get(line_field, {}).values() vals, func = self._get_line_vals(record, line_field, fields) is_cont = '_CONT_' in line_field and True or False # continue row cont_set = 0 rows_inserted = False # flag to insert row for rc, field in ws.get(line_field, {}).items(): col, row = co.split_row_col(rc) # starting point # Case continue, start from the last data row if is_cont and not cont_set: # only once per line_field cont_set = cont_row + 1 if is_cont: row = cont_set rc = '%s%s' % (col, cont_set) i = 0 new_row = 0 new_rc = False row_count = len(vals[field]) # Insert rows to preserve total line if not rows_inserted: rows_inserted = True st.insert_rows(row+1, amount=row_count-1) # -- for (row_val, style) in vals[field]: new_row = row + i new_rc = '%s%s' % (col, new_row) row_val = co.adjust_cell_formula(row_val, i) if row_val not in ('None', None): st[new_rc] = co.str_to_number(row_val) if style: styles = self.env['xlsx.styles'].get_openpyxl_styles() co.fill_cell_style(st[new_rc], style, styles) i += 1 # Add footer line if at least one field have sum f = func.get(field, False) if f and new_row > 0: new_row += 1 f_rc = '%s%s' % (col, new_row) st[f_rc] = '=%s(%s:%s)' % (f, rc, new_rc) co.fill_cell_style(st[f_rc], style, styles) cont_row = cont_row < new_row and new_row or cont_row return
@api.model def export_xlsx(self, template, res_model, res_id): if template.res_model != res_model: raise ValidationError(_("Template's model mismatch")) data_dict = co.literal_eval(template.instruction.strip()) export_dict = data_dict.get('__EXPORT__', False) out_name = template.name if not export_dict: # If there is not __EXPORT__ formula, just export out_name = template.fname out_file = template.datas return (out_file, out_name) # Prepare temp file (from now, only xlsx file works for openpyxl) decoded_data = base64.decodestring(template.datas) ConfParam = self.env['ir.config_parameter'].sudo() ptemp = ConfParam.get_param('path_temp_file') or '/tmp' stamp = dt.utcnow().strftime('%H%M%S%f')[:-3] ftemp = '%s/temp%s.xlsx' % (ptemp, stamp) f = open(ftemp, 'wb') f.write(decoded_data) f.seek(0) f.close() # Workbook created, temp file removed wb = load_workbook(ftemp) os.remove(ftemp) # Start working with workbook record = res_model and self.env[res_model].browse(res_id) or False self._fill_workbook_data(wb, record, export_dict) # Return file as .xlsx content = BytesIO() wb.save(content) content.seek(0) # Set index to 0, and start reading out_file = base64.encodestring(content.read()) if record and 'name' in record and record.name: out_name = record.name.replace(' ', '').replace('/', '') else: fname = out_name.replace(' ', '').replace('/', '') ts = fields.Datetime.context_timestamp(self, dt.now()) out_name = '%s_%s' % (fname, ts.strftime('%Y%m%d_%H%M%S')) if not out_name or len(out_name) == 0: out_name = 'noname' out_ext = 'xlsx' # CSV (convert only on 1st sheet) if template.to_csv: delimiter = template.csv_delimiter out_file = co.csv_from_excel(out_file, delimiter, template.csv_quote) out_ext = template.csv_extension return (out_file, '%s.%s' % (out_name, out_ext))
|