# test for xml.dom.minidom
import copy
import pickle
from test.support import run_unittest, findfile
import unittest
import xml.dom.minidom
from xml.dom.minidom import parse, Node, Document, parseString
from xml.dom.minidom import getDOMImplementation
tstfile = findfile("test.xml", subdir="xmltestdata")
sample = ("\n"
"\n"
" \n"
"]> text\n"
" ")
# The tests of DocumentType importing use these helpers to construct
# the documents to work with, since not all DOM builders actually
# create the DocumentType nodes.
def create_doc_without_doctype(doctype=None):
return getDOMImplementation().createDocument(None, "doc", doctype)
def create_nonempty_doctype():
doctype = getDOMImplementation().createDocumentType("doc", None, None)
doctype.entities._seq = []
doctype.notations._seq = []
notation = xml.dom.minidom.Notation("my-notation", None,
"http://xml.python.org/notations/my")
doctype.notations._seq.append(notation)
entity = xml.dom.minidom.Entity("my-entity", None,
"http://xml.python.org/entities/my",
"my-notation")
entity.version = "1.0"
entity.encoding = "utf-8"
entity.actualEncoding = "us-ascii"
doctype.entities._seq.append(entity)
return doctype
def create_doc_with_doctype():
doctype = create_nonempty_doctype()
doc = create_doc_without_doctype(doctype)
doctype.entities.item(0).ownerDocument = doc
doctype.notations.item(0).ownerDocument = doc
return doc
class MinidomTest(unittest.TestCase):
def confirm(self, test, testname = "Test"):
self.assertTrue(test, testname)
def checkWholeText(self, node, s):
t = node.wholeText
self.confirm(t == s, "looking for %r, found %r" % (s, t))
def testParseFromBinaryFile(self):
with open(tstfile, 'rb') as file:
dom = parse(file)
dom.unlink()
self.confirm(isinstance(dom, Document))
def testParseFromTextFile(self):
with open(tstfile, 'r', encoding='iso-8859-1') as file:
dom = parse(file)
dom.unlink()
self.confirm(isinstance(dom, Document))
def testGetElementsByTagName(self):
dom = parse(tstfile)
self.confirm(dom.getElementsByTagName("LI") == \
dom.documentElement.getElementsByTagName("LI"))
dom.unlink()
def testInsertBefore(self):
dom = parseString("")
root = dom.documentElement
elem = root.childNodes[0]
nelem = dom.createElement("element")
root.insertBefore(nelem, elem)
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2
and root.childNodes[0] is nelem
and root.childNodes.item(0) is nelem
and root.childNodes[1] is elem
and root.childNodes.item(1) is elem
and root.firstChild is nelem
and root.lastChild is elem
and root.toxml() == ""
, "testInsertBefore -- node properly placed in tree")
nelem = dom.createElement("element")
root.insertBefore(nelem, None)
self.confirm(len(root.childNodes) == 3
and root.childNodes.length == 3
and root.childNodes[1] is elem
and root.childNodes.item(1) is elem
and root.childNodes[2] is nelem
and root.childNodes.item(2) is nelem
and root.lastChild is nelem
and nelem.previousSibling is elem
and root.toxml() == ""
, "testInsertBefore -- node properly placed in tree")
nelem2 = dom.createElement("bar")
root.insertBefore(nelem2, nelem)
self.confirm(len(root.childNodes) == 4
and root.childNodes.length == 4
and root.childNodes[2] is nelem2
and root.childNodes.item(2) is nelem2
and root.childNodes[3] is nelem
and root.childNodes.item(3) is nelem
and nelem2.nextSibling is nelem
and nelem.previousSibling is nelem2
and root.toxml() ==
""
, "testInsertBefore -- node properly placed in tree")
dom.unlink()
def _create_fragment_test_nodes(self):
dom = parseString("")
orig = dom.createTextNode("original")
c1 = dom.createTextNode("foo")
c2 = dom.createTextNode("bar")
c3 = dom.createTextNode("bat")
dom.documentElement.appendChild(orig)
frag = dom.createDocumentFragment()
frag.appendChild(c1)
frag.appendChild(c2)
frag.appendChild(c3)
return dom, orig, c1, c2, c3, frag
def testInsertBeforeFragment(self):
dom, orig, c1, c2, c3, frag = self._create_fragment_test_nodes()
dom.documentElement.insertBefore(frag, None)
self.confirm(tuple(dom.documentElement.childNodes) ==
(orig, c1, c2, c3),
"insertBefore(, None)")
frag.unlink()
dom.unlink()
dom, orig, c1, c2, c3, frag = self._create_fragment_test_nodes()
dom.documentElement.insertBefore(frag, orig)
self.confirm(tuple(dom.documentElement.childNodes) ==
(c1, c2, c3, orig),
"insertBefore(, orig)")
frag.unlink()
dom.unlink()
def testAppendChild(self):
dom = parse(tstfile)
dom.documentElement.appendChild(dom.createComment("Hello"))
self.confirm(dom.documentElement.childNodes[-1].nodeName == "#comment")
self.confirm(dom.documentElement.childNodes[-1].data == "Hello")
dom.unlink()
def testAppendChildFragment(self):
dom, orig, c1, c2, c3, frag = self._create_fragment_test_nodes()
dom.documentElement.appendChild(frag)
self.confirm(tuple(dom.documentElement.childNodes) ==
(orig, c1, c2, c3),
"appendChild()")
frag.unlink()
dom.unlink()
def testReplaceChildFragment(self):
dom, orig, c1, c2, c3, frag = self._create_fragment_test_nodes()
dom.documentElement.replaceChild(frag, orig)
orig.unlink()
self.confirm(tuple(dom.documentElement.childNodes) == (c1, c2, c3),
"replaceChild()")
frag.unlink()
dom.unlink()
def testLegalChildren(self):
dom = Document()
elem = dom.createElement('element')
text = dom.createTextNode('text')
self.assertRaises(xml.dom.HierarchyRequestErr, dom.appendChild, text)
dom.appendChild(elem)
self.assertRaises(xml.dom.HierarchyRequestErr, dom.insertBefore, text,
elem)
self.assertRaises(xml.dom.HierarchyRequestErr, dom.replaceChild, text,
elem)
nodemap = elem.attributes
self.assertRaises(xml.dom.HierarchyRequestErr, nodemap.setNamedItem,
text)
self.assertRaises(xml.dom.HierarchyRequestErr, nodemap.setNamedItemNS,
text)
elem.appendChild(text)
dom.unlink()
def testNamedNodeMapSetItem(self):
dom = Document()
elem = dom.createElement('element')
attrs = elem.attributes
attrs["foo"] = "bar"
a = attrs.item(0)
self.confirm(a.ownerDocument is dom,
"NamedNodeMap.__setitem__() sets ownerDocument")
self.confirm(a.ownerElement is elem,
"NamedNodeMap.__setitem__() sets ownerElement")
self.confirm(a.value == "bar",
"NamedNodeMap.__setitem__() sets value")
self.confirm(a.nodeValue == "bar",
"NamedNodeMap.__setitem__() sets nodeValue")
elem.unlink()
dom.unlink()
def testNonZero(self):
dom = parse(tstfile)
self.confirm(dom)# should not be zero
dom.appendChild(dom.createComment("foo"))
self.confirm(not dom.childNodes[-1].childNodes)
dom.unlink()
def testUnlink(self):
dom = parse(tstfile)
self.assertTrue(dom.childNodes)
dom.unlink()
self.assertFalse(dom.childNodes)
def testContext(self):
with parse(tstfile) as dom:
self.assertTrue(dom.childNodes)
self.assertFalse(dom.childNodes)
def testElement(self):
dom = Document()
dom.appendChild(dom.createElement("abc"))
self.confirm(dom.documentElement)
dom.unlink()
def testAAA(self):
dom = parseString("")
el = dom.documentElement
el.setAttribute("spam", "jam2")
self.confirm(el.toxml() == '', "testAAA")
a = el.getAttributeNode("spam")
self.confirm(a.ownerDocument is dom,
"setAttribute() sets ownerDocument")
self.confirm(a.ownerElement is dom.documentElement,
"setAttribute() sets ownerElement")
dom.unlink()
def testAAB(self):
dom = parseString("")
el = dom.documentElement
el.setAttribute("spam", "jam")
el.setAttribute("spam", "jam2")
self.confirm(el.toxml() == '', "testAAB")
dom.unlink()
def testAddAttr(self):
dom = Document()
child = dom.appendChild(dom.createElement("abc"))
child.setAttribute("def", "ghi")
self.confirm(child.getAttribute("def") == "ghi")
self.confirm(child.attributes["def"].value == "ghi")
child.setAttribute("jkl", "mno")
self.confirm(child.getAttribute("jkl") == "mno")
self.confirm(child.attributes["jkl"].value == "mno")
self.confirm(len(child.attributes) == 2)
child.setAttribute("def", "newval")
self.confirm(child.getAttribute("def") == "newval")
self.confirm(child.attributes["def"].value == "newval")
self.confirm(len(child.attributes) == 2)
dom.unlink()
def testDeleteAttr(self):
dom = Document()
child = dom.appendChild(dom.createElement("abc"))
self.confirm(len(child.attributes) == 0)
child.setAttribute("def", "ghi")
self.confirm(len(child.attributes) == 1)
del child.attributes["def"]
self.confirm(len(child.attributes) == 0)
dom.unlink()
def testRemoveAttr(self):
dom = Document()
child = dom.appendChild(dom.createElement("abc"))
child.setAttribute("def", "ghi")
self.confirm(len(child.attributes) == 1)
self.assertRaises(xml.dom.NotFoundErr, child.removeAttribute, "foo")
child.removeAttribute("def")
self.confirm(len(child.attributes) == 0)
dom.unlink()
def testRemoveAttrNS(self):
dom = Document()
child = dom.appendChild(
dom.createElementNS("http://www.python.org", "python:abc"))
child.setAttributeNS("http://www.w3.org", "xmlns:python",
"http://www.python.org")
child.setAttributeNS("http://www.python.org", "python:abcattr", "foo")
self.assertRaises(xml.dom.NotFoundErr, child.removeAttributeNS,
"foo", "http://www.python.org")
self.confirm(len(child.attributes) == 2)
child.removeAttributeNS("http://www.python.org", "abcattr")
self.confirm(len(child.attributes) == 1)
dom.unlink()
def testRemoveAttributeNode(self):
dom = Document()
child = dom.appendChild(dom.createElement("foo"))
child.setAttribute("spam", "jam")
self.confirm(len(child.attributes) == 1)
node = child.getAttributeNode("spam")
self.assertRaises(xml.dom.NotFoundErr, child.removeAttributeNode,
None)
child.removeAttributeNode(node)
self.confirm(len(child.attributes) == 0
and child.getAttributeNode("spam") is None)
dom2 = Document()
child2 = dom2.appendChild(dom2.createElement("foo"))
node2 = child2.getAttributeNode("spam")
self.assertRaises(xml.dom.NotFoundErr, child2.removeAttributeNode,
node2)
dom.unlink()
def testHasAttribute(self):
dom = Document()
child = dom.appendChild(dom.createElement("foo"))
child.setAttribute("spam", "jam")
self.confirm(child.hasAttribute("spam"))
def testChangeAttr(self):
dom = parseString("")
el = dom.documentElement
el.setAttribute("spam", "jam")
self.confirm(len(el.attributes) == 1)
el.setAttribute("spam", "bam")
# Set this attribute to be an ID and make sure that doesn't change
# when changing the value:
el.setIdAttribute("spam")
self.confirm(len(el.attributes) == 1
and el.attributes["spam"].value == "bam"
and el.attributes["spam"].nodeValue == "bam"
and el.getAttribute("spam") == "bam"
and el.getAttributeNode("spam").isId)
el.attributes["spam"] = "ham"
self.confirm(len(el.attributes) == 1
and el.attributes["spam"].value == "ham"
and el.attributes["spam"].nodeValue == "ham"
and el.getAttribute("spam") == "ham"
and el.attributes["spam"].isId)
el.setAttribute("spam2", "bam")
self.confirm(len(el.attributes) == 2
and el.attributes["spam"].value == "ham"
and el.attributes["spam"].nodeValue == "ham"
and el.getAttribute("spam") == "ham"
and el.attributes["spam2"].value == "bam"
and el.attributes["spam2"].nodeValue == "bam"
and el.getAttribute("spam2") == "bam")
el.attributes["spam2"] = "bam2"
self.confirm(len(el.attributes) == 2
and el.attributes["spam"].value == "ham"
and el.attributes["spam"].nodeValue == "ham"
and el.getAttribute("spam") == "ham"
and el.attributes["spam2"].value == "bam2"
and el.attributes["spam2"].nodeValue == "bam2"
and el.getAttribute("spam2") == "bam2")
dom.unlink()
def testGetAttrList(self):
pass
def testGetAttrValues(self):
pass
def testGetAttrLength(self):
pass
def testGetAttribute(self):
dom = Document()
child = dom.appendChild(
dom.createElementNS("http://www.python.org", "python:abc"))
self.assertEqual(child.getAttribute('missing'), '')
def testGetAttributeNS(self):
dom = Document()
child = dom.appendChild(
dom.createElementNS("http://www.python.org", "python:abc"))
child.setAttributeNS("http://www.w3.org", "xmlns:python",
"http://www.python.org")
self.assertEqual(child.getAttributeNS("http://www.w3.org", "python"),
'http://www.python.org')
self.assertEqual(child.getAttributeNS("http://www.w3.org", "other"),
'')
child2 = child.appendChild(dom.createElement('abc'))
self.assertEqual(child2.getAttributeNS("http://www.python.org", "missing"),
'')
def testGetAttributeNode(self): pass
def testGetElementsByTagNameNS(self):
d="""
"""
dom = parseString(d)
elems = dom.getElementsByTagNameNS("http://pyxml.sf.net/minidom",
"myelem")
self.confirm(len(elems) == 1
and elems[0].namespaceURI == "http://pyxml.sf.net/minidom"
and elems[0].localName == "myelem"
and elems[0].prefix == "minidom"
and elems[0].tagName == "minidom:myelem"
and elems[0].nodeName == "minidom:myelem")
dom.unlink()
def get_empty_nodelist_from_elements_by_tagName_ns_helper(self, doc, nsuri,
lname):
nodelist = doc.getElementsByTagNameNS(nsuri, lname)
self.confirm(len(nodelist) == 0)
def testGetEmptyNodeListFromElementsByTagNameNS(self):
doc = parseString('')
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, 'http://xml.python.org/namespaces/a', 'localname')
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, '*', 'splat')
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, 'http://xml.python.org/namespaces/a', '*')
doc = parseString('')
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, "http://xml.python.org/splat", "not-there")
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, "*", "not-there")
self.get_empty_nodelist_from_elements_by_tagName_ns_helper(
doc, "http://somewhere.else.net/not-there", "e")
def testElementReprAndStr(self):
dom = Document()
el = dom.appendChild(dom.createElement("abc"))
string1 = repr(el)
string2 = str(el)
self.confirm(string1 == string2)
dom.unlink()
def testElementReprAndStrUnicode(self):
dom = Document()
el = dom.appendChild(dom.createElement("abc"))
string1 = repr(el)
string2 = str(el)
self.confirm(string1 == string2)
dom.unlink()
def testElementReprAndStrUnicodeNS(self):
dom = Document()
el = dom.appendChild(
dom.createElementNS("http://www.slashdot.org", "slash:abc"))
string1 = repr(el)
string2 = str(el)
self.confirm(string1 == string2)
self.confirm("slash:abc" in string1)
dom.unlink()
def testAttributeRepr(self):
dom = Document()
el = dom.appendChild(dom.createElement("abc"))
node = el.setAttribute("abc", "def")
self.confirm(str(node) == repr(node))
dom.unlink()
def testTextNodeRepr(self): pass
def testWriteXML(self):
str = ''
dom = parseString(str)
domstr = dom.toxml()
dom.unlink()
self.confirm(str == domstr)
def testAltNewline(self):
str = '\n\n'
dom = parseString(str)
domstr = dom.toprettyxml(newl="\r\n")
dom.unlink()
self.confirm(domstr == str.replace("\n", "\r\n"))
def test_toprettyxml_with_text_nodes(self):
# see issue #4147, text nodes are not indented
decl = '\n'
self.assertEqual(parseString('A').toprettyxml(),
decl + 'A\n')
self.assertEqual(parseString('AA').toprettyxml(),
decl + '\n\tA\n\tA\n\n')
self.assertEqual(parseString('AA').toprettyxml(),
decl + '\n\tA\n\tA\n\n')
self.assertEqual(parseString('AA').toprettyxml(),
decl + '\n\tA\n\tA\n\n')
self.assertEqual(parseString('AAA').toprettyxml(),
decl + '\n\tA\n\tA\n\tA\n\n')
def test_toprettyxml_with_adjacent_text_nodes(self):
# see issue #4147, adjacent text nodes are indented normally
dom = Document()
elem = dom.createElement('elem')
elem.appendChild(dom.createTextNode('TEXT'))
elem.appendChild(dom.createTextNode('TEXT'))
dom.appendChild(elem)
decl = '\n'
self.assertEqual(dom.toprettyxml(),
decl + '\n\tTEXT\n\tTEXT\n\n')
def test_toprettyxml_preserves_content_of_text_node(self):
# see issue #4147
for str in ('A', 'C'):
dom = parseString(str)
dom2 = parseString(dom.toprettyxml())
self.assertEqual(
dom.getElementsByTagName('B')[0].childNodes[0].toxml(),
dom2.getElementsByTagName('B')[0].childNodes[0].toxml())
def testProcessingInstruction(self):
dom = parseString('')
pi = dom.documentElement.firstChild
self.confirm(pi.target == "mypi"
and pi.data == "data \t\n "
and pi.nodeName == "mypi"
and pi.nodeType == Node.PROCESSING_INSTRUCTION_NODE
and pi.attributes is None
and not pi.hasChildNodes()
and len(pi.childNodes) == 0
and pi.firstChild is None
and pi.lastChild is None
and pi.localName is None
and pi.namespaceURI == xml.dom.EMPTY_NAMESPACE)
def testProcessingInstructionRepr(self): pass
def testTextRepr(self): pass
def testWriteText(self): pass
def testDocumentElement(self): pass
def testTooManyDocumentElements(self):
doc = parseString("")
elem = doc.createElement("extra")
# Should raise an exception when adding an extra document element.
self.assertRaises(xml.dom.HierarchyRequestErr, doc.appendChild, elem)
elem.unlink()
doc.unlink()
def testCreateElementNS(self): pass
def testCreateAttributeNS(self): pass
def testParse(self): pass
def testParseString(self): pass
def testComment(self): pass
def testAttrListItem(self): pass
def testAttrListItems(self): pass
def testAttrListItemNS(self): pass
def testAttrListKeys(self): pass
def testAttrListKeysNS(self): pass
def testRemoveNamedItem(self):
doc = parseString("")
e = doc.documentElement
attrs = e.attributes
a1 = e.getAttributeNode("a")
a2 = attrs.removeNamedItem("a")
self.confirm(a1.isSameNode(a2))
self.assertRaises(xml.dom.NotFoundErr, attrs.removeNamedItem, "a")
def testRemoveNamedItemNS(self):
doc = parseString("")
e = doc.documentElement
attrs = e.attributes
a1 = e.getAttributeNodeNS("http://xml.python.org/", "b")
a2 = attrs.removeNamedItemNS("http://xml.python.org/", "b")
self.confirm(a1.isSameNode(a2))
self.assertRaises(xml.dom.NotFoundErr, attrs.removeNamedItemNS,
"http://xml.python.org/", "b")
def testAttrListValues(self): pass
def testAttrListLength(self): pass
def testAttrList__getitem__(self): pass
def testAttrList__setitem__(self): pass
def testSetAttrValueandNodeValue(self): pass
def testParseElement(self): pass
def testParseAttributes(self): pass
def testParseElementNamespaces(self): pass
def testParseAttributeNamespaces(self): pass
def testParseProcessingInstructions(self): pass
def testChildNodes(self): pass
def testFirstChild(self): pass
def testHasChildNodes(self):
dom = parseString("")
doc = dom.documentElement
self.assertTrue(doc.hasChildNodes())
dom2 = parseString("")
doc2 = dom2.documentElement
self.assertFalse(doc2.hasChildNodes())
def _testCloneElementCopiesAttributes(self, e1, e2, test):
attrs1 = e1.attributes
attrs2 = e2.attributes
keys1 = list(attrs1.keys())
keys2 = list(attrs2.keys())
keys1.sort()
keys2.sort()
self.confirm(keys1 == keys2, "clone of element has same attribute keys")
for i in range(len(keys1)):
a1 = attrs1.item(i)
a2 = attrs2.item(i)
self.confirm(a1 is not a2
and a1.value == a2.value
and a1.nodeValue == a2.nodeValue
and a1.namespaceURI == a2.namespaceURI
and a1.localName == a2.localName
, "clone of attribute node has proper attribute values")
self.confirm(a2.ownerElement is e2,
"clone of attribute node correctly owned")
def _setupCloneElement(self, deep):
dom = parseString("")
root = dom.documentElement
clone = root.cloneNode(deep)
self._testCloneElementCopiesAttributes(
root, clone, "testCloneElement" + (deep and "Deep" or "Shallow"))
# mutilate the original so shared data is detected
root.tagName = root.nodeName = "MODIFIED"
root.setAttribute("attr", "NEW VALUE")
root.setAttribute("added", "VALUE")
return dom, clone
def testCloneElementShallow(self):
dom, clone = self._setupCloneElement(0)
self.confirm(len(clone.childNodes) == 0
and clone.childNodes.length == 0
and clone.parentNode is None
and clone.toxml() == ''
, "testCloneElementShallow")
dom.unlink()
def testCloneElementDeep(self):
dom, clone = self._setupCloneElement(1)
self.confirm(len(clone.childNodes) == 1
and clone.childNodes.length == 1
and clone.parentNode is None
and clone.toxml() == ''
, "testCloneElementDeep")
dom.unlink()
def testCloneDocumentShallow(self):
doc = parseString("\n"
""
"\n"
"]>\n"
"")
doc2 = doc.cloneNode(0)
self.confirm(doc2 is None,
"testCloneDocumentShallow:"
" shallow cloning of documents makes no sense!")
def testCloneDocumentDeep(self):
doc = parseString("\n"
""
"\n"
"]>\n"
"")
doc2 = doc.cloneNode(1)
self.confirm(not (doc.isSameNode(doc2) or doc2.isSameNode(doc)),
"testCloneDocumentDeep: document objects not distinct")
self.confirm(len(doc.childNodes) == len(doc2.childNodes),
"testCloneDocumentDeep: wrong number of Document children")
self.confirm(doc2.documentElement.nodeType == Node.ELEMENT_NODE,
"testCloneDocumentDeep: documentElement not an ELEMENT_NODE")
self.confirm(doc2.documentElement.ownerDocument.isSameNode(doc2),
"testCloneDocumentDeep: documentElement owner is not new document")
self.confirm(not doc.documentElement.isSameNode(doc2.documentElement),
"testCloneDocumentDeep: documentElement should not be shared")
if doc.doctype is not None:
# check the doctype iff the original DOM maintained it
self.confirm(doc2.doctype.nodeType == Node.DOCUMENT_TYPE_NODE,
"testCloneDocumentDeep: doctype not a DOCUMENT_TYPE_NODE")
self.confirm(doc2.doctype.ownerDocument.isSameNode(doc2))
self.confirm(not doc.doctype.isSameNode(doc2.doctype))
def testCloneDocumentTypeDeepOk(self):
doctype = create_nonempty_doctype()
clone = doctype.cloneNode(1)
self.confirm(clone is not None
and clone.nodeName == doctype.nodeName
and clone.name == doctype.name
and clone.publicId == doctype.publicId
and clone.systemId == doctype.systemId
and len(clone.entities) == len(doctype.entities)
and clone.entities.item(len(clone.entities)) is None
and len(clone.notations) == len(doctype.notations)
and clone.notations.item(len(clone.notations)) is None
and len(clone.childNodes) == 0)
for i in range(len(doctype.entities)):
se = doctype.entities.item(i)
ce = clone.entities.item(i)
self.confirm((not se.isSameNode(ce))
and (not ce.isSameNode(se))
and ce.nodeName == se.nodeName
and ce.notationName == se.notationName
and ce.publicId == se.publicId
and ce.systemId == se.systemId
and ce.encoding == se.encoding
and ce.actualEncoding == se.actualEncoding
and ce.version == se.version)
for i in range(len(doctype.notations)):
sn = doctype.notations.item(i)
cn = clone.notations.item(i)
self.confirm((not sn.isSameNode(cn))
and (not cn.isSameNode(sn))
and cn.nodeName == sn.nodeName
and cn.publicId == sn.publicId
and cn.systemId == sn.systemId)
def testCloneDocumentTypeDeepNotOk(self):
doc = create_doc_with_doctype()
clone = doc.doctype.cloneNode(1)
self.confirm(clone is None, "testCloneDocumentTypeDeepNotOk")
def testCloneDocumentTypeShallowOk(self):
doctype = create_nonempty_doctype()
clone = doctype.cloneNode(0)
self.confirm(clone is not None
and clone.nodeName == doctype.nodeName
and clone.name == doctype.name
and clone.publicId == doctype.publicId
and clone.systemId == doctype.systemId
and len(clone.entities) == 0
and clone.entities.item(0) is None
and len(clone.notations) == 0
and clone.notations.item(0) is None
and len(clone.childNodes) == 0)
def testCloneDocumentTypeShallowNotOk(self):
doc = create_doc_with_doctype()
clone = doc.doctype.cloneNode(0)
self.confirm(clone is None, "testCloneDocumentTypeShallowNotOk")
def check_import_document(self, deep, testName):
doc1 = parseString("")
doc2 = parseString("")
self.assertRaises(xml.dom.NotSupportedErr, doc1.importNode, doc2, deep)
def testImportDocumentShallow(self):
self.check_import_document(0, "testImportDocumentShallow")
def testImportDocumentDeep(self):
self.check_import_document(1, "testImportDocumentDeep")
def testImportDocumentTypeShallow(self):
src = create_doc_with_doctype()
target = create_doc_without_doctype()
self.assertRaises(xml.dom.NotSupportedErr, target.importNode,
src.doctype, 0)
def testImportDocumentTypeDeep(self):
src = create_doc_with_doctype()
target = create_doc_without_doctype()
self.assertRaises(xml.dom.NotSupportedErr, target.importNode,
src.doctype, 1)
# Testing attribute clones uses a helper, and should always be deep,
# even if the argument to cloneNode is false.
def check_clone_attribute(self, deep, testName):
doc = parseString("")
attr = doc.documentElement.getAttributeNode("attr")
self.assertNotEqual(attr, None)
clone = attr.cloneNode(deep)
self.confirm(not clone.isSameNode(attr))
self.confirm(not attr.isSameNode(clone))
self.confirm(clone.ownerElement is None,
testName + ": ownerElement should be None")
self.confirm(clone.ownerDocument.isSameNode(attr.ownerDocument),
testName + ": ownerDocument does not match")
self.confirm(clone.specified,
testName + ": cloned attribute must have specified == True")
def testCloneAttributeShallow(self):
self.check_clone_attribute(0, "testCloneAttributeShallow")
def testCloneAttributeDeep(self):
self.check_clone_attribute(1, "testCloneAttributeDeep")
def check_clone_pi(self, deep, testName):
doc = parseString("")
pi = doc.firstChild
self.assertEqual(pi.nodeType, Node.PROCESSING_INSTRUCTION_NODE)
clone = pi.cloneNode(deep)
self.confirm(clone.target == pi.target
and clone.data == pi.data)
def testClonePIShallow(self):
self.check_clone_pi(0, "testClonePIShallow")
def testClonePIDeep(self):
self.check_clone_pi(1, "testClonePIDeep")
def testNormalize(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode("first"))
root.appendChild(doc.createTextNode("second"))
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2,
"testNormalize -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 1
and root.childNodes.length == 1
and root.firstChild is root.lastChild
and root.firstChild.data == "firstsecond"
, "testNormalize -- result")
doc.unlink()
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode(""))
doc.normalize()
self.confirm(len(root.childNodes) == 0
and root.childNodes.length == 0,
"testNormalize -- single empty node removed")
doc.unlink()
def testNormalizeCombineAndNextSibling(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode("first"))
root.appendChild(doc.createTextNode("second"))
root.appendChild(doc.createElement("i"))
self.confirm(len(root.childNodes) == 3
and root.childNodes.length == 3,
"testNormalizeCombineAndNextSibling -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2
and root.firstChild.data == "firstsecond"
and root.firstChild is not root.lastChild
and root.firstChild.nextSibling is root.lastChild
and root.firstChild.previousSibling is None
and root.lastChild.previousSibling is root.firstChild
and root.lastChild.nextSibling is None
, "testNormalizeCombinedAndNextSibling -- result")
doc.unlink()
def testNormalizeDeleteWithPrevSibling(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode("first"))
root.appendChild(doc.createTextNode(""))
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2,
"testNormalizeDeleteWithPrevSibling -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 1
and root.childNodes.length == 1
and root.firstChild.data == "first"
and root.firstChild is root.lastChild
and root.firstChild.nextSibling is None
and root.firstChild.previousSibling is None
, "testNormalizeDeleteWithPrevSibling -- result")
doc.unlink()
def testNormalizeDeleteWithNextSibling(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode(""))
root.appendChild(doc.createTextNode("second"))
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2,
"testNormalizeDeleteWithNextSibling -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 1
and root.childNodes.length == 1
and root.firstChild.data == "second"
and root.firstChild is root.lastChild
and root.firstChild.nextSibling is None
and root.firstChild.previousSibling is None
, "testNormalizeDeleteWithNextSibling -- result")
doc.unlink()
def testNormalizeDeleteWithTwoNonTextSiblings(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createElement("i"))
root.appendChild(doc.createTextNode(""))
root.appendChild(doc.createElement("i"))
self.confirm(len(root.childNodes) == 3
and root.childNodes.length == 3,
"testNormalizeDeleteWithTwoSiblings -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2
and root.firstChild is not root.lastChild
and root.firstChild.nextSibling is root.lastChild
and root.firstChild.previousSibling is None
and root.lastChild.previousSibling is root.firstChild
and root.lastChild.nextSibling is None
, "testNormalizeDeleteWithTwoSiblings -- result")
doc.unlink()
def testNormalizeDeleteAndCombine(self):
doc = parseString("")
root = doc.documentElement
root.appendChild(doc.createTextNode(""))
root.appendChild(doc.createTextNode("second"))
root.appendChild(doc.createTextNode(""))
root.appendChild(doc.createTextNode("fourth"))
root.appendChild(doc.createTextNode(""))
self.confirm(len(root.childNodes) == 5
and root.childNodes.length == 5,
"testNormalizeDeleteAndCombine -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 1
and root.childNodes.length == 1
and root.firstChild is root.lastChild
and root.firstChild.data == "secondfourth"
and root.firstChild.previousSibling is None
and root.firstChild.nextSibling is None
, "testNormalizeDeleteAndCombine -- result")
doc.unlink()
def testNormalizeRecursion(self):
doc = parseString(""
""
""
"t"
#
#x
""
""
""
"t2"
#x2
""
"t3"
#x3
""
#
"")
root = doc.documentElement
root.childNodes[0].appendChild(doc.createTextNode(""))
root.childNodes[0].appendChild(doc.createTextNode("x"))
root.childNodes[1].childNodes[0].appendChild(doc.createTextNode("x2"))
root.childNodes[1].appendChild(doc.createTextNode("x3"))
root.appendChild(doc.createTextNode(""))
self.confirm(len(root.childNodes) == 3
and root.childNodes.length == 3
and len(root.childNodes[0].childNodes) == 4
and root.childNodes[0].childNodes.length == 4
and len(root.childNodes[1].childNodes) == 3
and root.childNodes[1].childNodes.length == 3
and len(root.childNodes[1].childNodes[0].childNodes) == 2
and root.childNodes[1].childNodes[0].childNodes.length == 2
, "testNormalize2 -- preparation")
doc.normalize()
self.confirm(len(root.childNodes) == 2
and root.childNodes.length == 2
and len(root.childNodes[0].childNodes) == 2
and root.childNodes[0].childNodes.length == 2
and len(root.childNodes[1].childNodes) == 2
and root.childNodes[1].childNodes.length == 2
and len(root.childNodes[1].childNodes[0].childNodes) == 1
and root.childNodes[1].childNodes[0].childNodes.length == 1
, "testNormalize2 -- childNodes lengths")
self.confirm(root.childNodes[0].childNodes[1].data == "tx"
and root.childNodes[1].childNodes[0].childNodes[0].data == "t2x2"
and root.childNodes[1].childNodes[1].data == "t3x3"
, "testNormalize2 -- joined text fields")
self.confirm(root.childNodes[0].childNodes[1].nextSibling is None
and root.childNodes[0].childNodes[1].previousSibling
is root.childNodes[0].childNodes[0]
and root.childNodes[0].childNodes[0].previousSibling is None
and root.childNodes[0].childNodes[0].nextSibling
is root.childNodes[0].childNodes[1]
and root.childNodes[1].childNodes[1].nextSibling is None
and root.childNodes[1].childNodes[1].previousSibling
is root.childNodes[1].childNodes[0]
and root.childNodes[1].childNodes[0].previousSibling is None
and root.childNodes[1].childNodes[0].nextSibling
is root.childNodes[1].childNodes[1]
, "testNormalize2 -- sibling pointers")
doc.unlink()
def testBug0777884(self):
doc = parseString("text")
text = doc.documentElement.childNodes[0]
self.assertEqual(text.nodeType, Node.TEXT_NODE)
# Should run quietly, doing nothing.
text.normalize()
doc.unlink()
def testBug1433694(self):
doc = parseString("t")
node = doc.documentElement
node.childNodes[1].nodeValue = ""
node.normalize()
self.confirm(node.childNodes[-1].nextSibling is None,
"Final child's .nextSibling should be None")
def testSiblings(self):
doc = parseString("text?")
root = doc.documentElement
(pi, text, elm) = root.childNodes
self.confirm(pi.nextSibling is text and
pi.previousSibling is None and
text.nextSibling is elm and
text.previousSibling is pi and
elm.nextSibling is None and
elm.previousSibling is text, "testSiblings")
doc.unlink()
def testParents(self):
doc = parseString(
"")
root = doc.documentElement
elm1 = root.childNodes[0]
(elm2a, elm2b) = elm1.childNodes
elm3 = elm2b.childNodes[0]
self.confirm(root.parentNode is doc and
elm1.parentNode is root and
elm2a.parentNode is elm1 and
elm2b.parentNode is elm1 and
elm3.parentNode is elm2b, "testParents")
doc.unlink()
def testNodeListItem(self):
doc = parseString("")
children = doc.childNodes
docelem = children[0]
self.confirm(children[0] is children.item(0)
and children.item(1) is None
and docelem.childNodes.item(0) is docelem.childNodes[0]
and docelem.childNodes.item(1) is docelem.childNodes[1]
and docelem.childNodes.item(0).childNodes.item(0) is None,
"test NodeList.item()")
doc.unlink()
def testEncodings(self):
doc = parseString('€')
self.assertEqual(doc.toxml(),
'\u20ac')
self.assertEqual(doc.toxml('utf-8'),
b'\xe2\x82\xac')
self.assertEqual(doc.toxml('iso-8859-15'),
b'\xa4')
self.assertEqual(doc.toxml('us-ascii'),
b'€')
self.assertEqual(doc.toxml('utf-16'),
''
'\u20ac'.encode('utf-16'))
# Verify that character decoding errors raise exceptions instead
# of crashing
self.assertRaises(UnicodeDecodeError, parseString,
b'Comment \xe7a va ? Tr\xe8s bien ?')
doc.unlink()
class UserDataHandler:
called = 0
def handle(self, operation, key, data, src, dst):
dst.setUserData(key, data + 1, self)
src.setUserData(key, None, None)
self.called = 1
def testUserData(self):
dom = Document()
n = dom.createElement('e')
self.confirm(n.getUserData("foo") is None)
n.setUserData("foo", None, None)
self.confirm(n.getUserData("foo") is None)
n.setUserData("foo", 12, 12)
n.setUserData("bar", 13, 13)
self.confirm(n.getUserData("foo") == 12)
self.confirm(n.getUserData("bar") == 13)
n.setUserData("foo", None, None)
self.confirm(n.getUserData("foo") is None)
self.confirm(n.getUserData("bar") == 13)
handler = self.UserDataHandler()
n.setUserData("bar", 12, handler)
c = n.cloneNode(1)
self.confirm(handler.called
and n.getUserData("bar") is None
and c.getUserData("bar") == 13)
n.unlink()
c.unlink()
dom.unlink()
def checkRenameNodeSharedConstraints(self, doc, node):
# Make sure illegal NS usage is detected:
self.assertRaises(xml.dom.NamespaceErr, doc.renameNode, node,
"http://xml.python.org/ns", "xmlns:foo")
doc2 = parseString("")
self.assertRaises(xml.dom.WrongDocumentErr, doc2.renameNode, node,
xml.dom.EMPTY_NAMESPACE, "foo")
def testRenameAttribute(self):
doc = parseString("")
elem = doc.documentElement
attrmap = elem.attributes
attr = elem.attributes['a']
# Simple renaming
attr = doc.renameNode(attr, xml.dom.EMPTY_NAMESPACE, "b")
self.confirm(attr.name == "b"
and attr.nodeName == "b"
and attr.localName is None
and attr.namespaceURI == xml.dom.EMPTY_NAMESPACE
and attr.prefix is None
and attr.value == "v"
and elem.getAttributeNode("a") is None
and elem.getAttributeNode("b").isSameNode(attr)
and attrmap["b"].isSameNode(attr)
and attr.ownerDocument.isSameNode(doc)
and attr.ownerElement.isSameNode(elem))
# Rename to have a namespace, no prefix
attr = doc.renameNode(attr, "http://xml.python.org/ns", "c")
self.confirm(attr.name == "c"
and attr.nodeName == "c"
and attr.localName == "c"
and attr.namespaceURI == "http://xml.python.org/ns"
and attr.prefix is None
and attr.value == "v"
and elem.getAttributeNode("a") is None
and elem.getAttributeNode("b") is None
and elem.getAttributeNode("c").isSameNode(attr)
and elem.getAttributeNodeNS(
"http://xml.python.org/ns", "c").isSameNode(attr)
and attrmap["c"].isSameNode(attr)
and attrmap[("http://xml.python.org/ns", "c")].isSameNode(attr))
# Rename to have a namespace, with prefix
attr = doc.renameNode(attr, "http://xml.python.org/ns2", "p:d")
self.confirm(attr.name == "p:d"
and attr.nodeName == "p:d"
and attr.localName == "d"
and attr.namespaceURI == "http://xml.python.org/ns2"
and attr.prefix == "p"
and attr.value == "v"
and elem.getAttributeNode("a") is None
and elem.getAttributeNode("b") is None
and elem.getAttributeNode("c") is None
and elem.getAttributeNodeNS(
"http://xml.python.org/ns", "c") is None
and elem.getAttributeNode("p:d").isSameNode(attr)
and elem.getAttributeNodeNS(
"http://xml.python.org/ns2", "d").isSameNode(attr)
and attrmap["p:d"].isSameNode(attr)
and attrmap[("http://xml.python.org/ns2", "d")].isSameNode(attr))
# Rename back to a simple non-NS node
attr = doc.renameNode(attr, xml.dom.EMPTY_NAMESPACE, "e")
self.confirm(attr.name == "e"
and attr.nodeName == "e"
and attr.localName is None
and attr.namespaceURI == xml.dom.EMPTY_NAMESPACE
and attr.prefix is None
and attr.value == "v"
and elem.getAttributeNode("a") is None
and elem.getAttributeNode("b") is None
and elem.getAttributeNode("c") is None
and elem.getAttributeNode("p:d") is None
and elem.getAttributeNodeNS(
"http://xml.python.org/ns", "c") is None
and elem.getAttributeNode("e").isSameNode(attr)
and attrmap["e"].isSameNode(attr))
self.assertRaises(xml.dom.NamespaceErr, doc.renameNode, attr,
"http://xml.python.org/ns", "xmlns")
self.checkRenameNodeSharedConstraints(doc, attr)
doc.unlink()
def testRenameElement(self):
doc = parseString("")
elem = doc.documentElement
# Simple renaming
elem = doc.renameNode(elem, xml.dom.EMPTY_NAMESPACE, "a")
self.confirm(elem.tagName == "a"
and elem.nodeName == "a"
and elem.localName is None
and elem.namespaceURI == xml.dom.EMPTY_NAMESPACE
and elem.prefix is None
and elem.ownerDocument.isSameNode(doc))
# Rename to have a namespace, no prefix
elem = doc.renameNode(elem, "http://xml.python.org/ns", "b")
self.confirm(elem.tagName == "b"
and elem.nodeName == "b"
and elem.localName == "b"
and elem.namespaceURI == "http://xml.python.org/ns"
and elem.prefix is None
and elem.ownerDocument.isSameNode(doc))
# Rename to have a namespace, with prefix
elem = doc.renameNode(elem, "http://xml.python.org/ns2", "p:c")
self.confirm(elem.tagName == "p:c"
and elem.nodeName == "p:c"
and elem.localName == "c"
and elem.namespaceURI == "http://xml.python.org/ns2"
and elem.prefix == "p"
and elem.ownerDocument.isSameNode(doc))
# Rename back to a simple non-NS node
elem = doc.renameNode(elem, xml.dom.EMPTY_NAMESPACE, "d")
self.confirm(elem.tagName == "d"
and elem.nodeName == "d"
and elem.localName is None
and elem.namespaceURI == xml.dom.EMPTY_NAMESPACE
and elem.prefix is None
and elem.ownerDocument.isSameNode(doc))
self.checkRenameNodeSharedConstraints(doc, elem)
doc.unlink()
def testRenameOther(self):
# We have to create a comment node explicitly since not all DOM
# builders used with minidom add comments to the DOM.
doc = xml.dom.minidom.getDOMImplementation().createDocument(
xml.dom.EMPTY_NAMESPACE, "e", None)
node = doc.createComment("comment")
self.assertRaises(xml.dom.NotSupportedErr, doc.renameNode, node,
xml.dom.EMPTY_NAMESPACE, "foo")
doc.unlink()
def testWholeText(self):
doc = parseString("a")
elem = doc.documentElement
text = elem.childNodes[0]
self.assertEqual(text.nodeType, Node.TEXT_NODE)
self.checkWholeText(text, "a")
elem.appendChild(doc.createTextNode("b"))
self.checkWholeText(text, "ab")
elem.insertBefore(doc.createCDATASection("c"), text)
self.checkWholeText(text, "cab")
# make sure we don't cross other nodes
splitter = doc.createComment("comment")
elem.appendChild(splitter)
text2 = doc.createTextNode("d")
elem.appendChild(text2)
self.checkWholeText(text, "cab")
self.checkWholeText(text2, "d")
x = doc.createElement("x")
elem.replaceChild(x, splitter)
splitter = x
self.checkWholeText(text, "cab")
self.checkWholeText(text2, "d")
x = doc.createProcessingInstruction("y", "z")
elem.replaceChild(x, splitter)
splitter = x
self.checkWholeText(text, "cab")
self.checkWholeText(text2, "d")
elem.removeChild(splitter)
self.checkWholeText(text, "cabd")
self.checkWholeText(text2, "cabd")
def testPatch1094164(self):
doc = parseString("")
elem = doc.documentElement
e = elem.firstChild
self.confirm(e.parentNode is elem, "Before replaceChild()")
# Check that replacing a child with itself leaves the tree unchanged
elem.replaceChild(e, e)
self.confirm(e.parentNode is elem, "After replaceChild()")
def testReplaceWholeText(self):
def setup():
doc = parseString("ad")
elem = doc.documentElement
text1 = elem.firstChild
text2 = elem.lastChild
splitter = text1.nextSibling
elem.insertBefore(doc.createTextNode("b"), splitter)
elem.insertBefore(doc.createCDATASection("c"), text1)
return doc, elem, text1, splitter, text2
doc, elem, text1, splitter, text2 = setup()
text = text1.replaceWholeText("new content")
self.checkWholeText(text, "new content")
self.checkWholeText(text2, "d")
self.confirm(len(elem.childNodes) == 3)
doc, elem, text1, splitter, text2 = setup()
text = text2.replaceWholeText("new content")
self.checkWholeText(text, "new content")
self.checkWholeText(text1, "cab")
self.confirm(len(elem.childNodes) == 5)
doc, elem, text1, splitter, text2 = setup()
text = text1.replaceWholeText("")
self.checkWholeText(text2, "d")
self.confirm(text is None
and len(elem.childNodes) == 2)
def testSchemaType(self):
doc = parseString(
"\n"
" \n"
" \n"
"]>")
elem = doc.documentElement
# We don't want to rely on any specific loader at this point, so
# just make sure we can get to all the names, and that the
# DTD-based namespace is right. The names can vary by loader
# since each supports a different level of DTD information.
t = elem.schemaType
self.confirm(t.name is None
and t.namespace == xml.dom.EMPTY_NAMESPACE)
names = "id notid text enum ref refs ent ents nm nms".split()
for name in names:
a = elem.getAttributeNode(name)
t = a.schemaType
self.confirm(hasattr(t, "name")
and t.namespace == xml.dom.EMPTY_NAMESPACE)
def testSetIdAttribute(self):
doc = parseString("")
e = doc.documentElement
a1 = e.getAttributeNode("a1")
a2 = e.getAttributeNode("a2")
self.confirm(doc.getElementById("v") is None
and not a1.isId
and not a2.isId)
e.setIdAttribute("a1")
self.confirm(e.isSameNode(doc.getElementById("v"))
and a1.isId
and not a2.isId)
e.setIdAttribute("a2")
self.confirm(e.isSameNode(doc.getElementById("v"))
and e.isSameNode(doc.getElementById("w"))
and a1.isId
and a2.isId)
# replace the a1 node; the new node should *not* be an ID
a3 = doc.createAttribute("a1")
a3.value = "v"
e.setAttributeNode(a3)
self.confirm(doc.getElementById("v") is None
and e.isSameNode(doc.getElementById("w"))
and not a1.isId
and a2.isId
and not a3.isId)
# renaming an attribute should not affect its ID-ness:
doc.renameNode(a2, xml.dom.EMPTY_NAMESPACE, "an")
self.confirm(e.isSameNode(doc.getElementById("w"))
and a2.isId)
def testSetIdAttributeNS(self):
NS1 = "http://xml.python.org/ns1"
NS2 = "http://xml.python.org/ns2"
doc = parseString("")
e = doc.documentElement
a1 = e.getAttributeNodeNS(NS1, "a1")
a2 = e.getAttributeNodeNS(NS2, "a2")
self.confirm(doc.getElementById("v") is None
and not a1.isId
and not a2.isId)
e.setIdAttributeNS(NS1, "a1")
self.confirm(e.isSameNode(doc.getElementById("v"))
and a1.isId
and not a2.isId)
e.setIdAttributeNS(NS2, "a2")
self.confirm(e.isSameNode(doc.getElementById("v"))
and e.isSameNode(doc.getElementById("w"))
and a1.isId
and a2.isId)
# replace the a1 node; the new node should *not* be an ID
a3 = doc.createAttributeNS(NS1, "a1")
a3.value = "v"
e.setAttributeNode(a3)
self.confirm(e.isSameNode(doc.getElementById("w")))
self.confirm(not a1.isId)
self.confirm(a2.isId)
self.confirm(not a3.isId)
self.confirm(doc.getElementById("v") is None)
# renaming an attribute should not affect its ID-ness:
doc.renameNode(a2, xml.dom.EMPTY_NAMESPACE, "an")
self.confirm(e.isSameNode(doc.getElementById("w"))
and a2.isId)
def testSetIdAttributeNode(self):
NS1 = "http://xml.python.org/ns1"
NS2 = "http://xml.python.org/ns2"
doc = parseString("")
e = doc.documentElement
a1 = e.getAttributeNodeNS(NS1, "a1")
a2 = e.getAttributeNodeNS(NS2, "a2")
self.confirm(doc.getElementById("v") is None
and not a1.isId
and not a2.isId)
e.setIdAttributeNode(a1)
self.confirm(e.isSameNode(doc.getElementById("v"))
and a1.isId
and not a2.isId)
e.setIdAttributeNode(a2)
self.confirm(e.isSameNode(doc.getElementById("v"))
and e.isSameNode(doc.getElementById("w"))
and a1.isId
and a2.isId)
# replace the a1 node; the new node should *not* be an ID
a3 = doc.createAttributeNS(NS1, "a1")
a3.value = "v"
e.setAttributeNode(a3)
self.confirm(e.isSameNode(doc.getElementById("w")))
self.confirm(not a1.isId)
self.confirm(a2.isId)
self.confirm(not a3.isId)
self.confirm(doc.getElementById("v") is None)
# renaming an attribute should not affect its ID-ness:
doc.renameNode(a2, xml.dom.EMPTY_NAMESPACE, "an")
self.confirm(e.isSameNode(doc.getElementById("w"))
and a2.isId)
def assert_recursive_equal(self, doc, doc2):
stack = [(doc, doc2)]
while stack:
n1, n2 = stack.pop()
self.assertEqual(n1.nodeType, n2.nodeType)
self.assertEqual(len(n1.childNodes), len(n2.childNodes))
self.assertEqual(n1.nodeName, n2.nodeName)
self.assertFalse(n1.isSameNode(n2))
self.assertFalse(n2.isSameNode(n1))
if n1.nodeType == Node.DOCUMENT_TYPE_NODE:
len(n1.entities)
len(n2.entities)
len(n1.notations)
len(n2.notations)
self.assertEqual(len(n1.entities), len(n2.entities))
self.assertEqual(len(n1.notations), len(n2.notations))
for i in range(len(n1.notations)):
# XXX this loop body doesn't seem to be executed?
no1 = n1.notations.item(i)
no2 = n1.notations.item(i)
self.assertEqual(no1.name, no2.name)
self.assertEqual(no1.publicId, no2.publicId)
self.assertEqual(no1.systemId, no2.systemId)
stack.append((no1, no2))
for i in range(len(n1.entities)):
e1 = n1.entities.item(i)
e2 = n2.entities.item(i)
self.assertEqual(e1.notationName, e2.notationName)
self.assertEqual(e1.publicId, e2.publicId)
self.assertEqual(e1.systemId, e2.systemId)
stack.append((e1, e2))
if n1.nodeType != Node.DOCUMENT_NODE:
self.assertTrue(n1.ownerDocument.isSameNode(doc))
self.assertTrue(n2.ownerDocument.isSameNode(doc2))
for i in range(len(n1.childNodes)):
stack.append((n1.childNodes[i], n2.childNodes[i]))
def testPickledDocument(self):
doc = parseString(sample)
for proto in range(2, pickle.HIGHEST_PROTOCOL + 1):
s = pickle.dumps(doc, proto)
doc2 = pickle.loads(s)
self.assert_recursive_equal(doc, doc2)
def testDeepcopiedDocument(self):
doc = parseString(sample)
doc2 = copy.deepcopy(doc)
self.assert_recursive_equal(doc, doc2)
def testSerializeCommentNodeWithDoubleHyphen(self):
doc = create_doc_without_doctype()
doc.appendChild(doc.createComment("foo--bar"))
self.assertRaises(ValueError, doc.toxml)
def testEmptyXMLNSValue(self):
doc = parseString("\n"
"\n")
doc2 = parseString(doc.toxml())
self.confirm(doc2.namespaceURI == xml.dom.EMPTY_NAMESPACE)
def testExceptionOnSpacesInXMLNSValue(self):
with self.assertRaisesRegex(ValueError, 'Unsupported syntax'):
parseString('')
def testDocRemoveChild(self):
doc = parse(tstfile)
title_tag = doc.documentElement.getElementsByTagName("TITLE")[0]
self.assertRaises( xml.dom.NotFoundErr, doc.removeChild, title_tag)
num_children_before = len(doc.childNodes)
doc.removeChild(doc.childNodes[0])
num_children_after = len(doc.childNodes)
self.assertTrue(num_children_after == num_children_before - 1)
def testProcessingInstructionNameError(self):
# wrong variable in .nodeValue property will
# lead to "NameError: name 'data' is not defined"
doc = parse(tstfile)
pi = doc.createProcessingInstruction("y", "z")
pi.nodeValue = "crash"
def test_main():
run_unittest(MinidomTest)
if __name__ == "__main__":
test_main()