Fix XML document rendering in Firefox

[ghostwords]

Fixes #1949.

This only seems to happen in Firefox. Does Chrome handle the checks we add in this PR internally?

Sample XML document: https://sourceforge.net/sitemap.xml
Sample XHTML document: http://tieba.baidu.com/mo/q---2799060AC2198378C599200012D229ED%3AFG%3D1--1-3-0--2--wapp_1440757687244_809/m?kw=%E9%9A%90%E6%9D%80&lp=1030

This seems like a good thing to have automated tests for, but I haven't yet figured out how to detect the difference between a pretty-rendered XML document in Firefox and a non-pretty-thanks-to-an-injected-script-tag version of the same with Selenium. The problem seems to be that Firefox handles its pretty rendering HTML specially; it's not part of the document (doesn't seem to be even inspectable by dev tools).

Here is my (non-working) test attempt:

diff --git a/tests/selenium/breakage_test.py b/tests/selenium/breakage_test.py
index 33a3743..db86fd8 100644
--- a/tests/selenium/breakage_test.py
+++ b/tests/selenium/breakage_test.py
@@ -1,14 +1,16 @@
 #!/usr/bin/env python
 # -*- coding: UTF-8 -*-
 
+import os
 import unittest
+
 import pbtest
+
 from selenium.webdriver.support.ui import WebDriverWait
 from selenium.webdriver.support import expected_conditions as EC
-import re
 
 
-class Test(pbtest.PBSeleniumTest):
+class BreakageTest(pbtest.PBSeleniumTest):
     """Make sure the extension doesn't break common sites and use cases.
     e.g. we should be able to load a website, search on Google.
     TODO: Add tests to simulate most common web use cases:
@@ -27,5 +29,14 @@ class Test(pbtest.PBSeleniumTest):
         qry_el.submit()
         WebDriverWait(self.driver, 10).until(EC.title_contains("EFF"))
 
+    # https://github.com/EFForg/privacybadger/issues/1949
+    def test_xml_rendering(self):
+        xml_fixture_path = os.path.join(
+            pbtest.get_git_root(), 'tests', 'fixtures', 'xml_document.xml')
+        self.load_url('file://' + xml_fixture_path)
+        page_contents = self.js("return document.outerHTML;")
+        self.assertEqual(page_contents, '<root>Test</root>',
+            "XML document's contents should have been left alone.")
+
 if __name__ == "__main__":
     unittest.main()

$ cat tests/fixtures/xml_document.xml 
<root>Test</root>

[ghostwords]

This PR also adds a check to avoid injecting the canvas fingerprinting detection script on first-party frames. This check already exists when receiving canvas reports, so why not avoid modifying page objects unless we need to do it. This may help with some canvas-related site breakages (if the canvas is first-party, the breakage will go away).

Oops, I goofed, 7f7ff1d (rebased out) doesn't work since we need to compare script URL to frame URL and we don't have script URL at the time of injection yet.

[bcyphers]

This looks good. I tried for a while, but could not find any way for selenium to differentiate between XML that is rendered as a document tree and the broken pseudo-HTML that we get in the error case. Chrome seems to rewrite the document itself, but firefox leaves the source exactly the same.

Anyway -- it seems like more trouble than its worth to write a test for this right now.