def __len__(self):
return self.size()
+ def __nonzero__(self):
+ """ Ensure that a node_descriptor will never equal False
+
+ Since we do define __len__ and __iter__ for us, we must avoid
+ being regarded as non-true objects.
+ """
+ return True
+
def next(self, str):
raise NotImplementedError
for dfld in dirr.dctx_ids:
try:
self.dctx['dctx_' + dfld.field] = safe_eval(dfld.expr,dc2)
- except Exception:
+ except Exception,e:
print "Cannot eval %s" % dfld.expr
print e
pass
where.append(('id','=',self.resm_id))
if name:
- where.append((self.namefield,'=',name))
+ # The =like character will match underscores against any characters
+ # including the special ones that couldn't exist in a FTP/DAV request
+ where.append((self.namefield,'=like',name.replace('\\','\\\\')))
is_allowed = self.check_perms(1)
else:
is_allowed = self.check_perms(5)
for bo in obj.browse(cr, uid, resids, context=ctx):
if not bo:
continue
- name = getattr(bo, self.namefield)
- if not name:
+ res_name = getattr(bo, self.namefield)
+ if not res_name:
continue
# Yes! we can't do better but skip nameless records.
+
+ # Escape the name for characters not supported in filenames
+ res_name = res_name.replace('/','_') # any other weird char?
+
+ if name and (res_name != name):
+ # we have matched _ to any character, but we only meant to match
+ # the special ones.
+ # Eg. 'a_c' will find 'abc', 'a/c', 'a_c', may only
+ # return 'a/c' and 'a_c'
+ continue
- res.append(self.res_obj_class(name, self.dir_id, self, self.context, self.res_model, bo))
+ res.append(self.res_obj_class(res_name, self.dir_id, self, self.context, self.res_model, bo))
return res
def _get_ttag(self,cr):
self.domain = parent.domain
self.displayname = path
self.dctx_dict = parent.dctx_dict
- self.res_find_all = parent.res_find_all
+ if isinstance(parent, node_res_dir):
+ self.res_find_all = parent.res_find_all
+ else:
+ self.res_find_all = False
if res_bo:
self.res_id = res_bo.id
dc2 = self.context.context.copy()
for fld,expr in self.dctx_dict.items():
try:
self.dctx[fld] = safe_eval(expr, dc2)
- except Exception:
+ except Exception,e:
print "Cannot eval %s for %s" % (expr, fld)
print e
pass
# Directory Structure display in tree structure
if self.res_id and directory.ressource_tree:
where1 = []
+ if name:
+ where1.append(('name','=like',name.replace('\\','\\\\')))
if obj._parent_name in obj.fields_get(cr, uid):
- where1 = where + [(obj._parent_name, '=', self.res_id)]
+ where1.append((obj._parent_name, '=', self.res_id))
namefield = directory.resource_field.name or 'name'
resids = obj.search(cr, uid, where1, context=ctx)
for bo in obj.browse(cr, uid, resids, context=ctx):
res_name = getattr(bo, namefield)
if not res_name:
continue
+ res_name = res_name.replace('/', '_')
+ if name and (res_name != name):
+ continue
# TODO Revise
klass = directory.get_node_class(directory, dynamic=True, context=ctx)
- res.append(klass(res_name, dir_id=self.dir_id, parent=self, context=self.context, res_model=self.res_model, res_bo=bo))
+ rnode = klass(res_name, dir_id=self.dir_id, parent=self, context=self.context,
+ res_model=self.res_model, res_bo=bo)
+ rnode.res_find_all = self.res_find_all
+ res.append(rnode)
where2 = where + [('parent_id','=',self.dir_id) ]
ids = dirobj.search(cr, uid, where2, context=ctx)
+ bo = obj.browse(cr, uid, self.res_id, context=ctx)
+
for dirr in dirobj.browse(cr, uid, ids, context=ctx):
+ if name and (name != dirr.name):
+ continue
if dirr.type == 'directory':
klass = dirr.get_node_class(dirr, dynamic=True, context=ctx)
- res.append(klass(dirr.name, dirr.id, self, self.context, self.res_model, res_bo = None, res_id = self.res_id))
+ res.append(klass(dirr.name, dirr.id, self, self.context, self.res_model, res_bo = bo, res_id = self.res_id))
elif dirr.type == 'ressource':
# child resources can be controlled by properly set dctx
klass = dirr.get_node_class(dirr, context=ctx)
- res.append(klass(dirr.name,self,self.context, dirr, {'active_id': self.res_id}))
+ res.append(klass(dirr.name,self,self.context, dirr, {'active_id': self.res_id})) # bo?
fil_obj = dirobj.pool.get('ir.attachment')
if self.res_find_all:
where2 = where
- where3 = where2 + [('res_model', '=', self.res_model), ('res_id','=',self.res_id)]
- # print "where clause for dir_obj", where2
+ where3 = where2 + [('res_model', '=', self.res_model), ('res_id','=',self.res_id)]
+ # print "where clause for dir_obj", where3
ids = fil_obj.search(cr, uid, where3, context=ctx)
if ids:
for fil in fil_obj.browse(cr, uid, ids, context=ctx):
# Get Child Ressource Directories
if directory.ressource_type_id and directory.ressource_type_id.id:
where4 = where + [('ressource_parent_type_id','=',directory.ressource_type_id.id)]
- where5 = where4 + [('ressource_id','=',0)]
+ where5 = where4 + ['|', ('ressource_id','=',0), ('ressource_id','=',self.res_id)]
dirids = dirobj.search(cr,uid, where5)
- where5 = where4 + [('ressource_id','=',self.res_id)]
- dirids = dirids + dirobj.search(cr,uid, where5)
for dirr in dirobj.browse(cr, uid, dirids, context=ctx):
if dirr.type == 'directory' and not dirr.parent_id:
klass = dirr.get_node_class(dirr, dynamic=True, context=ctx)
- res.append(klass(dirr.name, dirr.id, self, self.context, self.res_model, res_bo = None, res_id = self.res_id))
+ rnode = klass(dirr.name, dirr.id, self, self.context, self.res_model, res_bo = bo, res_id = self.res_id)
+ rnode.res_find_all = dirr.resource_find_all
+ res.append(rnode)
if dirr.type == 'ressource':
klass = dirr.get_node_class(dirr, context=ctx)
- res.append(klass(dirr.name, self, self.context, dirr, {'active_id': self.res_id}))
+ rnode = klass(dirr.name, self, self.context, dirr, {'active_id': self.res_id})
+ rnode.res_find_all = dirr.resource_find_all
+ res.append(rnode)
return res
def create_child_collection(self, cr, objname):
'name': objname,
'ressource_parent_type_id': obj and obj.ressource_type_id.id or False,
'ressource_id': object2 and object2.id or False,
- 'parent_id' : False
+ 'parent_id' : False,
+ 'resource_find_all': False,
}
if (obj and (obj.type in ('directory'))) or not object2:
val['parent_id'] = obj and obj.id or False