2015-09-07 17:22:01 +00:00
|
|
|
# -*- coding: utf-8 -*-
|
2014-07-12 13:46:55 +00:00
|
|
|
from searx.testing import SearxTestCase
|
|
|
|
from searx import utils
|
|
|
|
|
|
|
|
|
|
|
|
class TestUtils(SearxTestCase):
|
|
|
|
|
|
|
|
def test_gen_useragent(self):
|
|
|
|
self.assertIsInstance(utils.gen_useragent(), str)
|
|
|
|
self.assertIsNotNone(utils.gen_useragent())
|
|
|
|
self.assertTrue(utils.gen_useragent().startswith('Mozilla'))
|
|
|
|
|
2015-01-27 19:03:33 +00:00
|
|
|
def test_searx_useragent(self):
|
|
|
|
self.assertIsInstance(utils.searx_useragent(), str)
|
|
|
|
self.assertIsNotNone(utils.searx_useragent())
|
|
|
|
self.assertTrue(utils.searx_useragent().startswith('searx'))
|
|
|
|
|
|
|
|
def test_html_to_text(self):
|
|
|
|
html = """
|
|
|
|
<a href="/testlink" class="link_access_account">
|
|
|
|
<span class="toto">
|
|
|
|
<span>
|
|
|
|
<img src="test.jpg" />
|
|
|
|
</span>
|
|
|
|
</span>
|
|
|
|
<span class="titi">
|
|
|
|
Test text
|
|
|
|
</span>
|
|
|
|
</a>
|
|
|
|
"""
|
2020-08-06 15:42:46 +00:00
|
|
|
self.assertIsInstance(utils.html_to_text(html), str)
|
2015-01-27 19:03:33 +00:00
|
|
|
self.assertIsNotNone(utils.html_to_text(html))
|
|
|
|
self.assertEqual(utils.html_to_text(html), "Test text")
|
|
|
|
|
2020-09-11 08:23:56 +00:00
|
|
|
def test_html_to_text_invalid(self):
|
|
|
|
html = '<p><b>Lorem ipsum</i>dolor sit amet</p>'
|
|
|
|
self.assertEqual(utils.html_to_text(html), "Lorem ipsum")
|
|
|
|
|
2018-03-01 04:30:48 +00:00
|
|
|
def test_match_language(self):
|
|
|
|
self.assertEqual(utils.match_language('es', ['es']), 'es')
|
|
|
|
self.assertEqual(utils.match_language('es', [], fallback='fallback'), 'fallback')
|
|
|
|
self.assertEqual(utils.match_language('ja', ['jp'], {'ja': 'jp'}), 'jp')
|
|
|
|
|
|
|
|
aliases = {'en-GB': 'en-UK', 'he': 'iw'}
|
|
|
|
|
|
|
|
# guess country
|
|
|
|
self.assertEqual(utils.match_language('de-DE', ['de']), 'de')
|
|
|
|
self.assertEqual(utils.match_language('de', ['de-DE']), 'de-DE')
|
|
|
|
self.assertEqual(utils.match_language('es-CO', ['es-AR', 'es-ES', 'es-MX']), 'es-ES')
|
|
|
|
self.assertEqual(utils.match_language('es-CO', ['es-MX']), 'es-MX')
|
|
|
|
self.assertEqual(utils.match_language('en-UK', ['en-AU', 'en-GB', 'en-US']), 'en-GB')
|
|
|
|
self.assertEqual(utils.match_language('en-GB', ['en-AU', 'en-UK', 'en-US'], aliases), 'en-UK')
|
|
|
|
|
|
|
|
# language aliases
|
|
|
|
self.assertEqual(utils.match_language('iw', ['he']), 'he')
|
|
|
|
self.assertEqual(utils.match_language('he', ['iw'], aliases), 'iw')
|
|
|
|
self.assertEqual(utils.match_language('iw-IL', ['he']), 'he')
|
|
|
|
self.assertEqual(utils.match_language('he-IL', ['iw'], aliases), 'iw')
|
|
|
|
self.assertEqual(utils.match_language('iw', ['he-IL']), 'he-IL')
|
|
|
|
self.assertEqual(utils.match_language('he', ['iw-IL'], aliases), 'iw-IL')
|
|
|
|
self.assertEqual(utils.match_language('iw-IL', ['he-IL']), 'he-IL')
|
|
|
|
self.assertEqual(utils.match_language('he-IL', ['iw-IL'], aliases), 'iw-IL')
|
|
|
|
|
2019-08-02 11:37:13 +00:00
|
|
|
def test_ecma_unscape(self):
|
|
|
|
self.assertEqual(utils.ecma_unescape('text%20with%20space'), 'text with space')
|
|
|
|
self.assertEqual(utils.ecma_unescape('text using %xx: %F3'),
|
|
|
|
u'text using %xx: ó')
|
|
|
|
self.assertEqual(utils.ecma_unescape('text using %u: %u5409, %u4E16%u754c'),
|
|
|
|
u'text using %u: 吉, 世界')
|
|
|
|
|
2014-07-12 13:46:55 +00:00
|
|
|
|
|
|
|
class TestHTMLTextExtractor(SearxTestCase):
|
|
|
|
|
|
|
|
def setUp(self):
|
|
|
|
self.html_text_extractor = utils.HTMLTextExtractor()
|
|
|
|
|
|
|
|
def test__init__(self):
|
|
|
|
self.assertEqual(self.html_text_extractor.result, [])
|
|
|
|
|
|
|
|
def test_handle_charref(self):
|
|
|
|
self.html_text_extractor.handle_charref('xF')
|
2020-08-06 15:42:46 +00:00
|
|
|
self.assertIn('\x0f', self.html_text_extractor.result)
|
2014-07-12 13:46:55 +00:00
|
|
|
self.html_text_extractor.handle_charref('XF')
|
2020-08-06 15:42:46 +00:00
|
|
|
self.assertIn('\x0f', self.html_text_extractor.result)
|
2014-07-12 13:46:55 +00:00
|
|
|
|
|
|
|
self.html_text_extractor.handle_charref('97')
|
2020-08-06 15:42:46 +00:00
|
|
|
self.assertIn('a', self.html_text_extractor.result)
|
2014-07-12 13:46:55 +00:00
|
|
|
|
|
|
|
def test_handle_entityref(self):
|
|
|
|
entity = 'test'
|
|
|
|
self.html_text_extractor.handle_entityref(entity)
|
|
|
|
self.assertIn(entity, self.html_text_extractor.result)
|
|
|
|
|
2020-09-11 08:23:56 +00:00
|
|
|
def test_invalid_html(self):
|
|
|
|
text = '<p><b>Lorem ipsum</i>dolor sit amet</p>'
|
|
|
|
with self.assertRaises(utils.HTMLTextExtractorException):
|
|
|
|
self.html_text_extractor.feed(text)
|