xadmin导出文件部分源码
如下
--- 在项目中全局搜索FileResponse可找见
class ExportPlugin(BaseAdminPlugin): export_mimes = {'xlsx': 'application/vnd.ms-excel', 'xls': 'application/vnd.ms-excel', 'csv': 'text/csv', 'xml': 'application/xhtml+xml', 'json': 'application/json'} def init_request(self, *args, **kwargs): return self.request.GET.get('_do_') == 'export' def _format_value(self, o): if (o.field is None and getattr(o.attr, 'boolean', False)) or \ (o.field and isinstance(o.field, (BooleanField, NullBooleanField))): value = o.value elif str(o.text).startswith("<span class='text-muted'>"): value = escape(str(o.text)[25:-7]) else: value = escape(str(o.text)) return value def _get_objects(self, context): headers = [c for c in context['result_headers'].cells if c.export] rows = context['results'] return [dict([ (force_text(headers[i].text), self._format_value(o)) for i, o in enumerate(filter(lambda c:getattr(c, 'export', False), r.cells))]) for r in rows] def _get_datas(self, context): rows = context['results'] new_rows = [[self._format_value(o) for o in filter(lambda c:getattr(c, 'export', False), r.cells)] for r in rows] new_rows.insert(0, [force_text(c.text) for c in context['result_headers'].cells if c.export]) return new_rows def get_xlsx_export(self, context): datas = self._get_datas(context) output = io.BytesIO() export_header = ( self.request.GET.get('export_xlsx_header', 'off') == 'on') model_name = self.opts.verbose_name book = xlsxwriter.Workbook(output) sheet = book.add_worksheet( u"%s %s" % (_(u'Sheet'), force_text(model_name))) styles = {'datetime': book.add_format({'num_format': 'yyyy-mm-dd hh:mm:ss'}), 'date': book.add_format({'num_format': 'yyyy-mm-dd'}), 'time': book.add_format({'num_format': 'hh:mm:ss'}), 'header': book.add_format({'font': 'name Times New Roman', 'color': 'red', 'bold': 'on', 'num_format': '#,##0.00'}), 'default': book.add_format()} if not export_header: datas = datas[1:] for rowx, row in enumerate(datas): for colx, value in enumerate(row): if export_header and rowx == 0: cell_style = styles['header'] else: if isinstance(value, datetime.datetime): cell_style = styles['datetime'] elif isinstance(value, datetime.date): cell_style = styles['date'] elif isinstance(value, datetime.time): cell_style = styles['time'] else: cell_style = styles['default'] sheet.write(rowx, colx, value, cell_style) book.close() output.seek(0) return output.getvalue() def get_xls_export(self, context): datas = self._get_datas(context) output = io.BytesIO() export_header = ( self.request.GET.get('export_xls_header', 'off') == 'on') model_name = self.opts.verbose_name book = xlwt.Workbook(encoding='utf8') sheet = book.add_sheet( u"%s %s" % (_(u'Sheet'), force_text(model_name))) styles = {'datetime': xlwt.easyxf(num_format_str='yyyy-mm-dd hh:mm:ss'), 'date': xlwt.easyxf(num_format_str='yyyy-mm-dd'), 'time': xlwt.easyxf(num_format_str='hh:mm:ss'), 'header': xlwt.easyxf('font: name Times New Roman, color-index red, bold on', num_format_str='#,##0.00'), 'default': xlwt.Style.default_style} if not export_header: datas = datas[1:] for rowx, row in enumerate(datas): for colx, value in enumerate(row): if export_header and rowx == 0: cell_style = styles['header'] else: if isinstance(value, datetime.datetime): cell_style = styles['datetime'] elif isinstance(value, datetime.date): cell_style = styles['date'] elif isinstance(value, datetime.time): cell_style = styles['time'] else: cell_style = styles['default'] sheet.write(rowx, colx, value, style=cell_style) book.save(output) output.seek(0) return output.getvalue() def _format_csv_text(self, t): if isinstance(t, bool): return _('Yes') if t else _('No') t = t.replace('"', '""').replace(',', '\,') cls_str = str if six.PY3 else basestring if isinstance(t, cls_str): t = '"%s"' % t return t def get_csv_export(self, context): datas = self._get_datas(context) stream = [] if self.request.GET.get('export_csv_header', 'off') != 'on': datas = datas[1:] for row in datas: stream.append(','.join(map(self._format_csv_text, row))) return '\r\n'.join(stream) def _to_xml(self, xml, data): if isinstance(data, (list, tuple)): for item in data: xml.startElement("row", {}) self._to_xml(xml, item) xml.endElement("row") elif isinstance(data, dict): for key, value in iteritems(data): key = key.replace(' ', '_') xml.startElement(key, {}) self._to_xml(xml, value) xml.endElement(key) else: xml.characters(smart_text(data)) def get_xml_export(self, context): results = self._get_objects(context) stream = io.StringIO() xml = SimplerXMLGenerator(stream, "utf-8") xml.startDocument() xml.startElement("objects", {}) self._to_xml(xml, results) xml.endElement("objects") xml.endDocument() return stream.getvalue().split('\n')[1] def get_json_export(self, context): results = self._get_objects(context) return json.dumps({'objects': results}, ensure_ascii=False, indent=(self.request.GET.get('export_json_format', 'off') == 'on') and 4 or None) def get_response(self, response, context, *args, **kwargs): file_type = self.request.GET.get('export_type', 'csv') response = HttpResponse( content_type="%s; charset=UTF-8" % self.export_mimes[file_type]) file_name = self.opts.verbose_name.replace(' ', '_') response['Content-Disposition'] = ('attachment; filename=%s.%s' % ( file_name, file_type)).encode('utf-8') response.write(getattr(self, 'get_%s_export' % file_type)(context)) return response # View Methods def get_result_list(self, __): if self.request.GET.get('all', 'off') == 'on': self.admin_view.list_per_page = sys.maxsize return __() def result_header(self, item, field_name, row): item.export = not item.attr or field_name == '__str__' or getattr(item.attr, 'allow_export', True) return item def result_item(self, item, obj, field_name, row): item.export = item.field or field_name == '__str__' or getattr(item.attr, 'allow_export', True) return item