#!/usr/bin/env python
# -*- coding: utf-8 -*-
###
### Author: Chris Iatrou (ichrispa@core-vector.net)
### Version: rev 13
###
### This program was created for educational purposes and has been
### contributed to the open62541 project by the author. All licensing
### terms for this source is inherited by the terms and conditions
### specified for by the open62541 project (see the projects readme
### file for more information on the LGPL terms and restrictions).
###
### This program is not meant to be used in a production environment. The
### author is not liable for any complications arising due to the use of
### this program.
###
from __future__ import print_function
import sys
import xml.dom.minidom as dom
import logging
import codecs
import re
from datatypes import *
from nodes import *
from opaque_type_mapping import opaque_type_mapping
__all__ = ['NodeSet', 'getSubTypesOf']
logger = logging.getLogger(__name__)
if sys.version_info[0] >= 3:
# strings are already parsed to unicode
def unicode(s):
return s
string_types = str
else:
string_types = basestring
####################
# Helper Functions #
####################
hassubtype = NodeId("ns=0;i=45")
def getSubTypesOf(nodeset, node, skipNodes=[]):
if node in skipNodes:
return []
re = set()
re.add(node)
for ref in node.references:
if (ref.referenceType == hassubtype):
skipAll = set()
skipAll.update(skipNodes)
skipAll.update(re)
if (ref.source == node.id and ref.isForward):
re.update(getSubTypesOf(nodeset, nodeset.nodes[ref.target], skipNodes=skipAll))
elif (ref.target == node.id and not ref.isForward):
re.update(getSubTypesOf(nodeset, nodeset.nodes[ref.source], skipNodes=skipAll))
return re
def extractNamespaces(xmlfile):
# Extract a list of namespaces used. The first namespace is always
# "http://opcfoundation.org/UA/". minidom gobbles up
# elements, without a decent way to reliably
# access this dom2 elements (only attribute xmlns= are accessible
# using minidom). We need them for dereferencing though... This function
# attempts to do just that.
namespaces = ["http://opcfoundation.org/UA/"]
infile = codecs.open(xmlfile.name, encoding='utf-8')
foundURIs = False
nsline = ""
for line in infile:
if "" in line.lower():
foundURIs = True
elif "" in line.lower():
nsline = nsline + line
break
if foundURIs:
nsline = nsline + line
if len(nsline) > 0:
ns = dom.parseString(nsline).getElementsByTagName("NamespaceUris")
for uri in ns[0].childNodes:
if uri.nodeType != uri.ELEMENT_NODE:
continue
if uri.firstChild.data in namespaces:
continue
namespaces.append(uri.firstChild.data)
infile.close()
return namespaces
def buildAliasList(xmlelement):
"""Parses the XML Element present in must XML NodeSet definitions.
Contents the Alias element are stored in a dictionary for further
dereferencing during pointer linkage (see linkOpenPointer())."""
aliases = {}
for al in xmlelement.childNodes:
if al.nodeType == al.ELEMENT_NODE:
if al.hasAttribute("Alias"):
aliasst = al.getAttribute("Alias")
aliasnd = unicode(al.firstChild.data)
aliases[aliasst] = aliasnd
return aliases
class NodeSet(object):
""" This class handles parsing XML description of namespaces, instantiating
nodes, linking references, graphing the namespace and compiling a binary
representation.
Note that nodes assigned to this class are not restricted to having a
single namespace ID. This class represents the entire physical address
space of the binary representation and all nodes that are to be included
in that segment of memory.
"""
def __init__(self):
self.nodes = {}
self.aliases = {}
self.namespaces = ["http://opcfoundation.org/UA/"]
self.namespaceMapping = {};
def sanitize(self):
for n in self.nodes.values():
if n.sanitize() == False:
raise Exception("Failed to sanitize node " + str(n))
# Sanitize reference consistency
for n in self.nodes.values():
for ref in n.references:
if not ref.source == n.id:
raise Exception("Reference " + str(ref) + " has an invalid source")
if not ref.referenceType in self.nodes:
raise Exception("Reference " + str(ref) + " has an unknown reference type")
if not ref.target in self.nodes:
raise Exception("Reference " + str(ref) + " has an unknown target")
def addNamespace(self, nsURL):
if not nsURL in self.namespaces:
self.namespaces.append(nsURL)
def createNamespaceMapping(self, orig_namespaces):
"""Creates a dict that maps from the nsindex in the original nodeset to the
nsindex in the combined nodeset"""
m = {}
for index, name in enumerate(orig_namespaces):
m[index] = self.namespaces.index(name)
return m
def getNodeByBrowseName(self, idstring):
return next((n for n in self.nodes.values() if idstring == n.browseName.name), None)
def getNodeById(self, namespace, id):
nodeId = NodeId()
nodeId.ns = namespace
nodeId.i = id
return self.nodes[nodeId]
def getRoot(self):
return self.getNodeByBrowseName("Root")
def createNode(self, xmlelement, modelUri, hidden=False):
ndtype = xmlelement.localName.lower()
if ndtype[:2] == "ua":
ndtype = ndtype[2:]
node = None
if ndtype == 'variable':
node = VariableNode(xmlelement)
if ndtype == 'object':
node = ObjectNode(xmlelement)
if ndtype == 'method':
node = MethodNode(xmlelement)
if ndtype == 'objecttype':
node = ObjectTypeNode(xmlelement)
if ndtype == 'variabletype':
node = VariableTypeNode(xmlelement)
if ndtype == 'methodtype':
node = MethodNode(xmlelement)
if ndtype == 'datatype':
node = DataTypeNode(xmlelement)
if ndtype == 'referencetype':
node = ReferenceTypeNode(xmlelement)
if node is None:
return None
node.modelUri = modelUri
node.hidden = hidden
return node
def hide_node(self, nodeId, hidden=True):
if not nodeId in self.nodes:
return False
node = self.nodes[nodeId]
node.hidden = hidden
return True
def merge_dicts(self, *dict_args):
"""
Given any number of dicts, shallow copy and merge into a new dict,
precedence goes to key value pairs in latter dicts.
"""
result = {}
for dictionary in dict_args:
result.update(dictionary)
return result
def addNodeSet(self, xmlfile, hidden=False, typesArray="UA_TYPES"):
# Extract NodeSet DOM
fileContent = xmlfile.read()
# Remove BOM since the dom parser cannot handle it on python 3 windows
if fileContent.startswith( codecs.BOM_UTF8 ):
fileContent = fileContent.lstrip( codecs.BOM_UTF8 )
if (sys.version_info >= (3, 0)):
fileContent = fileContent.decode("utf-8")
# Remove the uax namespace from tags. UaModeler adds this namespace to some elements
fileContent = re.sub(r"<([/]?)uax:(.+?)([/]?)>", "<\g<1>\g<2>\g<3>>", fileContent)
nodesets = dom.parseString(fileContent).getElementsByTagName("UANodeSet")
if len(nodesets) == 0 or len(nodesets) > 1:
raise Exception(self, self.originXML + " contains no or more then 1 nodeset")
nodeset = nodesets[0]
# Extract the modelUri
try:
modelTag = nodeset.getElementsByTagName("Models")[0].getElementsByTagName("Model")[0]
modelUri = modelTag.attributes["ModelUri"].nodeValue
except Exception:
# Ignore exception and try to use namespace array
modelUri = None
# Create the namespace mapping
orig_namespaces = extractNamespaces(xmlfile) # List of namespaces used in the xml file
if modelUri is None and len(orig_namespaces) > 1:
modelUri = orig_namespaces[1]
if modelUri is None:
raise Exception(self, self.originXML + " does not define the nodeset URI in Models/Model/ModelUri or NamespaceUris array.")
for ns in orig_namespaces:
self.addNamespace(ns)
self.namespaceMapping[modelUri] = self.createNamespaceMapping(orig_namespaces)
# Extract the aliases
for nd in nodeset.childNodes:
if nd.nodeType != nd.ELEMENT_NODE:
continue
ndtype = nd.localName.lower()
if 'aliases' in ndtype:
self.aliases = self.merge_dicts(self.aliases, buildAliasList(nd))
# Instantiate nodes
newnodes = []
for nd in nodeset.childNodes:
if nd.nodeType != nd.ELEMENT_NODE:
continue
node = self.createNode(nd, modelUri, hidden)
if not node:
continue
node.replaceAliases(self.aliases)
node.replaceNamespaces(self.namespaceMapping[modelUri])
node.typesArray = typesArray
# Add the node the the global dict
if node.id in self.nodes:
raise Exception("XMLElement with duplicate ID " + str(node.id))
self.nodes[node.id] = node
newnodes.append(node)
def getBinaryEncodingIdForNode(self, nodeId):
"""
The node should have a 'HasEncoding' forward reference which points to the encoding ids.
These can be XML Encoding or Binary Encoding. Therefore we also need to check if the SymbolicName
of the target node is "DefaultBinary"
"""
node = self.nodes[nodeId]
for ref in node.references:
if ref.referenceType.ns == 0 and ref.referenceType.i == 38:
refNode = self.nodes[ref.target]
if refNode.symbolicName.value == "DefaultBinary":
return ref.target
raise Exception("No DefaultBinary encoding defined for node " + str(nodeId))
def buildEncodingRules(self):
""" Calls buildEncoding() for all DataType nodes (opcua_node_dataType_t).
No return value
"""
stat = {True: 0, False: 0}
for n in self.nodes.values():
if isinstance(n, DataTypeNode):
n.buildEncoding(self)
stat[n.isEncodable()] = stat[n.isEncodable()] + 1
logger.debug("Type definitions built/passed: " + str(stat))
def allocateVariables(self):
for n in self.nodes.values():
if isinstance(n, VariableNode):
n.allocateValue(self)
def getBaseDataType(self, node):
if node is None:
return None
if node.browseName.name not in opaque_type_mapping:
return node
for ref in node.references:
if ref.isForward:
continue
if ref.referenceType.i == 45:
return self.getBaseDataType(self.nodes[ref.target])
return node
def getNodeTypeDefinition(self, node):
for ref in node.references:
# 40 = HasTypeDefinition
if ref.referenceType.i == 40 and ref.isForward:
return self.nodes[ref.target]
return None
def getDataTypeNode(self, dataType):
if isinstance(dataType, string_types):
if not valueIsInternalType(dataType):
logger.error("Not a valid dataType string: " + dataType)
return None
return self.nodes[NodeId(self.aliases[dataType])]
if isinstance(dataType, NodeId):
if dataType.i == 0:
return None
dataTypeNode = self.nodes[dataType]
if not isinstance(dataTypeNode, DataTypeNode):
logger.error("Node id " + str(dataType) + " is not reference a valid dataType.")
return None
return dataTypeNode
return None
def getRelevantOrderingReferences(self):
relevant_types = set()
relevant_types.update(getSubTypesOf(self, self.getNodeByBrowseName("HierarchicalReferences"), []))
relevant_types.update(getSubTypesOf(self, self.getNodeByBrowseName("HasEncoding"), []))
relevant_types.update(getSubTypesOf(self, self.getNodeByBrowseName("HasTypeDefinition"), []))
return list(map(lambda x: x.id, relevant_types))
def addInverseReferences(self):
# Ensure that every reference has an inverse reference in the target
for u in self.nodes.values():
for ref in u.references:
back = Reference(ref.target, ref.referenceType, ref.source, not ref.isForward)
self.nodes[ref.target].references.add(back) # ref set does not make a duplicate entry
def setNodeParent(self):
parentreftypes = getSubTypesOf(self, self.getNodeByBrowseName("HierarchicalReferences"))
parentreftypes = list(map(lambda x: x.id, parentreftypes))
for node in self.nodes.values():
parentref = node.getParentReference(parentreftypes)
if parentref is not None:
node.parent = self.nodes[parentref.target]
node.parentReference = self.nodes[parentref.referenceType]