# -*- coding: utf-8 -*- import copy import openerp from openerp.exceptions import AccessError from openerp.osv import osv from lxml import etree, html from openerp import api class view(osv.osv): _inherit = 'ir.ui.view' @api.cr_uid_ids_context def render(self, cr, uid, id_or_xml_id, values=None, engine='ir.qweb', context=None): if not values: values = {} if values.get('editable'): try: if not isinstance(id_or_xml_id, (int, long)): if '.' not in id_or_xml_id: raise ValueError('Invalid template id: %r' % (id_or_xml_id,)) id_or_xml_id = self.get_view_id(cr, uid, id_or_xml_id, context=context) self.check_access_rule(cr, uid, [id_or_xml_id], 'write', context=context) except AccessError: values['editable'] = False return super(view, self).render(cr, uid, id_or_xml_id, values=values, engine=engine, context=context) #------------------------------------------------------ # Save from html #------------------------------------------------------ def extract_embedded_fields(self, cr, uid, arch, context=None): return arch.xpath('//*[@data-oe-model != "ir.ui.view"]') def save_embedded_field(self, cr, uid, el, context=None): Model = self.pool[el.get('data-oe-model')] field = el.get('data-oe-field') converter = self.pool['ir.qweb'].get_converter_for(el.get('data-oe-type')) value = converter.from_html(cr, uid, Model, Model._fields[field], el) if value is not None: # TODO: batch writes? Model.write(cr, uid, [int(el.get('data-oe-id'))], { field: value }, context=context) def _pretty_arch(self, arch): # remove_blank_string does not seem to work on HTMLParser, and # pretty-printing with lxml more or less requires stripping # whitespace: http://lxml.de/FAQ.html#why-doesn-t-the-pretty-print-option-reformat-my-xml-output # so serialize to XML, parse as XML (remove whitespace) then serialize # as XML (pretty print) arch_no_whitespace = etree.fromstring( etree.tostring(arch, encoding='utf-8'), parser=etree.XMLParser(encoding='utf-8', remove_blank_text=True)) return etree.tostring( arch_no_whitespace, encoding='unicode', pretty_print=True) def replace_arch_section(self, cr, uid, view_id, section_xpath, replacement, context=None): # the root of the arch section shouldn't actually be replaced as it's # not really editable itself, only the content truly is editable. [view] = self.browse(cr, uid, [view_id], context=context) arch = etree.fromstring(view.arch.encode('utf-8')) # => get the replacement root if not section_xpath: root = arch else: # ensure there's only one match [root] = arch.xpath(section_xpath) root.text = replacement.text root.tail = replacement.tail # replace all children del root[:] for child in replacement: root.append(copy.deepcopy(child)) return arch def to_field_ref(self, cr, uid, el, context=None): # filter out meta-information inserted in the document attributes = dict((k, v) for k, v in el.items() if not k.startswith('data-oe-')) attributes['t-field'] = el.get('data-oe-expression') out = html.html_parser.makeelement(el.tag, attrib=attributes) out.tail = el.tail return out def save(self, cr, uid, res_id, value, xpath=None, context=None): """ Update a view section. The view section may embed fields to write :param str model: :param int res_id: :param str xpath: valid xpath to the tag to replace """ res_id = int(res_id) arch_section = html.fromstring( value, parser=html.HTMLParser(encoding='utf-8')) if xpath is None: # value is an embedded field on its own, not a view section self.save_embedded_field(cr, uid, arch_section, context=context) return for el in self.extract_embedded_fields(cr, uid, arch_section, context=context): self.save_embedded_field(cr, uid, el, context=context) # transform embedded field back to t-field el.getparent().replace(el, self.to_field_ref(cr, uid, el, context=context)) arch = self.replace_arch_section(cr, uid, res_id, xpath, arch_section, context=context) self.write(cr, uid, res_id, { 'arch': self._pretty_arch(arch) }, context=context) view = self.browse(cr, openerp.SUPERUSER_ID, res_id, context=context) if view.model_data_id: view.model_data_id.write({'noupdate': True})