# -*- coding: utf-8 -*- # This file is part of beets. # Copyright 2016, Fabrice Laporte. # # Permission is hereby granted, free of charge, to any person obtaining # a copy of this software and associated documentation files (the # "Software"), to deal in the Software without restriction, including # without limitation the rights to use, copy, modify, merge, publish, # distribute, sublicense, and/or sell copies of the Software, and to # permit persons to whom the Software is furnished to do so, subject to # the following conditions: # # The above copyright notice and this permission notice shall be # included in all copies or substantial portions of the Software. """Tests for the 'lyrics' plugin.""" from __future__ import absolute_import, division, print_function import os import re import six import sys import unittest from mock import patch from test import _common from beets import logging from beets.library import Item from beets.util import bytestring_path import confuse from beetsplug import lyrics from mock import MagicMock log = logging.getLogger('beets.test_lyrics') raw_backend = lyrics.Backend({}, log) google = lyrics.Google(MagicMock(), log) class LyricsPluginTest(unittest.TestCase): def setUp(self): """Set up configuration.""" lyrics.LyricsPlugin() def test_search_artist(self): item = Item(artist='Alice ft. Bob', title='song') self.assertIn(('Alice ft. Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice feat Bob', title='song') self.assertIn(('Alice feat Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice feat. Bob', title='song') self.assertIn(('Alice feat. Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice feats Bob', title='song') self.assertIn(('Alice feats Bob', ['song']), lyrics.search_pairs(item)) self.assertNotIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice featuring Bob', title='song') self.assertIn(('Alice featuring Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice & Bob', title='song') self.assertIn(('Alice & Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice and Bob', title='song') self.assertIn(('Alice and Bob', ['song']), lyrics.search_pairs(item)) self.assertIn(('Alice', ['song']), lyrics.search_pairs(item)) item = Item(artist='Alice and Bob', title='song') self.assertEqual(('Alice and Bob', ['song']), list(lyrics.search_pairs(item))[0]) def test_search_pairs_multi_titles(self): item = Item(title='1 / 2', artist='A') self.assertIn(('A', ['1 / 2']), lyrics.search_pairs(item)) self.assertIn(('A', ['1', '2']), lyrics.search_pairs(item)) item = Item(title='1/2', artist='A') self.assertIn(('A', ['1/2']), lyrics.search_pairs(item)) self.assertIn(('A', ['1', '2']), lyrics.search_pairs(item)) def test_search_pairs_titles(self): item = Item(title='Song (live)', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song (live)']), lyrics.search_pairs(item)) item = Item(title='Song (live) (new)', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song (live) (new)']), lyrics.search_pairs(item)) item = Item(title='Song (live (new))', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song (live (new))']), lyrics.search_pairs(item)) item = Item(title='Song ft. B', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song ft. B']), lyrics.search_pairs(item)) item = Item(title='Song featuring B', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song featuring B']), lyrics.search_pairs(item)) item = Item(title='Song and B', artist='A') self.assertNotIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song and B']), lyrics.search_pairs(item)) item = Item(title='Song: B', artist='A') self.assertIn(('A', ['Song']), lyrics.search_pairs(item)) self.assertIn(('A', ['Song: B']), lyrics.search_pairs(item)) def test_remove_credits(self): self.assertEqual( lyrics.remove_credits("""It's close to midnight Lyrics brought by example.com"""), "It's close to midnight" ) self.assertEqual( lyrics.remove_credits("""Lyrics brought by example.com"""), "" ) # don't remove 2nd verse for the only reason it contains 'lyrics' word text = """Look at all the shit that i done bought her See lyrics ain't nothin if the beat aint crackin""" self.assertEqual(lyrics.remove_credits(text), text) def test_is_lyrics(self): texts = ['LyricsMania.com - Copyright (c) 2013 - All Rights Reserved'] texts += ["""All material found on this site is property\n of mywickedsongtext brand"""] for t in texts: self.assertFalse(google.is_lyrics(t)) def test_slugify(self): text = u"http://site.com/\xe7afe-au_lait(boisson)" self.assertEqual(google.slugify(text), 'http://site.com/cafe_au_lait') def test_scrape_strip_cruft(self): text = u"""  one
two !

four""" self.assertEqual(lyrics._scrape_strip_cruft(text, True), "one\ntwo !\n\nfour") def test_scrape_strip_scripts(self): text = u"""foobaz""" self.assertEqual(lyrics._scrape_strip_cruft(text, True), "foobaz") def test_scrape_strip_tag_in_comment(self): text = u"""fooqux""" self.assertEqual(lyrics._scrape_strip_cruft(text, True), "fooqux") def test_scrape_merge_paragraphs(self): text = u"one

two

three" self.assertEqual(lyrics._scrape_merge_paragraphs(text), "one\ntwo\nthree") def test_missing_lyrics(self): self.assertFalse(google.is_lyrics(LYRICS_TEXTS['missing_texts'])) def url_to_filename(url): url = re.sub(r'https?://|www.', '', url) fn = "".join(x for x in url if (x.isalnum() or x == '/')) fn = fn.split('/') fn = os.path.join(LYRICS_ROOT_DIR, bytestring_path(fn[0]), bytestring_path(fn[-1] + '.txt')) return fn class MockFetchUrl(object): def __init__(self, pathval='fetched_path'): self.pathval = pathval self.fetched = None def __call__(self, url, filename=None): self.fetched = url fn = url_to_filename(url) with open(fn, 'r') as f: content = f.read() return content def is_lyrics_content_ok(title, text): """Compare lyrics text to expected lyrics for given title.""" if not text: return keywords = set(LYRICS_TEXTS[google.slugify(title)].split()) words = set(x.strip(".?, ") for x in text.lower().split()) return keywords <= words LYRICS_ROOT_DIR = os.path.join(_common.RSRC, b'lyrics') yaml_path = os.path.join(_common.RSRC, b'lyricstext.yaml') LYRICS_TEXTS = confuse.load_yaml(yaml_path) class LyricsGoogleBaseTest(unittest.TestCase): def setUp(self): """Set up configuration.""" try: __import__('bs4') except ImportError: self.skipTest('Beautiful Soup 4 not available') if sys.version_info[:3] < (2, 7, 3): self.skipTest("Python's built-in HTML parser is not good enough") class LyricsPluginSourcesTest(LyricsGoogleBaseTest): """Check that beets google custom search engine sources are correctly scraped. """ DEFAULT_SONG = dict(artist=u'The Beatles', title=u'Lady Madonna') DEFAULT_SOURCES = [ dict(DEFAULT_SONG, backend=lyrics.LyricsWiki), dict(artist=u'Santana', title=u'Black magic woman', backend=lyrics.MusiXmatch), dict(DEFAULT_SONG, backend=lyrics.Genius), ] GOOGLE_SOURCES = [ dict(DEFAULT_SONG, url=u'http://www.absolutelyrics.com', path=u'/lyrics/view/the_beatles/lady_madonna'), dict(DEFAULT_SONG, url=u'http://www.azlyrics.com', path=u'/lyrics/beatles/ladymadonna.html'), dict(DEFAULT_SONG, url=u'http://www.chartlyrics.com', path=u'/_LsLsZ7P4EK-F-LD4dJgDQ/Lady+Madonna.aspx'), dict(DEFAULT_SONG, url=u'http://www.elyricsworld.com', path=u'/lady_madonna_lyrics_beatles.html'), dict(url=u'http://www.lacoccinelle.net', artist=u'Jacques Brel', title=u"Amsterdam", path=u'/paroles-officielles/275679.html'), dict(DEFAULT_SONG, url=u'http://letras.mus.br/', path=u'the-beatles/275/'), dict(DEFAULT_SONG, url='http://www.lyricsmania.com/', path='lady_madonna_lyrics_the_beatles.html'), dict(DEFAULT_SONG, url=u'http://lyrics.wikia.com/', path=u'The_Beatles:Lady_Madonna'), dict(DEFAULT_SONG, url=u'http://www.lyricsmode.com', path=u'/lyrics/b/beatles/lady_madonna.html'), dict(url=u'http://www.lyricsontop.com', artist=u'Amy Winehouse', title=u"Jazz'n'blues", path=u'/amy-winehouse-songs/jazz-n-blues-lyrics.html'), dict(DEFAULT_SONG, url='http://www.metrolyrics.com/', path='lady-madonna-lyrics-beatles.html'), dict(url='http://www.musica.com/', path='letras.asp?letra=2738', artist=u'Santana', title=u'Black magic woman'), dict(url=u'http://www.paroles.net/', artist=u'Lilly Wood & the prick', title=u"Hey it's ok", path=u'lilly-wood-the-prick/paroles-hey-it-s-ok'), dict(DEFAULT_SONG, url='http://www.songlyrics.com', path=u'/the-beatles/lady-madonna-lyrics'), dict(DEFAULT_SONG, url=u'http://www.sweetslyrics.com', path=u'/761696.The%20Beatles%20-%20Lady%20Madonna.html') ] def setUp(self): LyricsGoogleBaseTest.setUp(self) self.plugin = lyrics.LyricsPlugin() @unittest.skipUnless(os.environ.get( 'BEETS_TEST_LYRICS_SOURCES', '0') == '1', 'lyrics sources testing not enabled') def test_backend_sources_ok(self): """Test default backends with songs known to exist in respective databases. """ errors = [] for s in self.DEFAULT_SOURCES: res = s['backend'](self.plugin.config, self.plugin._log).fetch( s['artist'], s['title']) if not is_lyrics_content_ok(s['title'], res): errors.append(s['backend'].__name__) self.assertFalse(errors) @unittest.skipUnless(os.environ.get( 'BEETS_TEST_LYRICS_SOURCES', '0') == '1', 'lyrics sources testing not enabled') def test_google_sources_ok(self): """Test if lyrics present on websites registered in beets google custom search engine are correctly scraped. """ for s in self.GOOGLE_SOURCES: url = s['url'] + s['path'] res = lyrics.scrape_lyrics_from_html( raw_backend.fetch_url(url)) self.assertTrue(google.is_lyrics(res), url) self.assertTrue(is_lyrics_content_ok(s['title'], res), url) class LyricsGooglePluginMachineryTest(LyricsGoogleBaseTest): """Test scraping heuristics on a fake html page. """ source = dict(url=u'http://www.example.com', artist=u'John Doe', title=u'Beets song', path=u'/lyrics/beetssong') def setUp(self): """Set up configuration""" LyricsGoogleBaseTest.setUp(self) self.plugin = lyrics.LyricsPlugin() @patch.object(lyrics.Backend, 'fetch_url', MockFetchUrl()) def test_mocked_source_ok(self): """Test that lyrics of the mocked page are correctly scraped""" url = self.source['url'] + self.source['path'] res = lyrics.scrape_lyrics_from_html(raw_backend.fetch_url(url)) self.assertTrue(google.is_lyrics(res), url) self.assertTrue(is_lyrics_content_ok(self.source['title'], res), url) @patch.object(lyrics.Backend, 'fetch_url', MockFetchUrl()) def test_is_page_candidate_exact_match(self): """Test matching html page title with song infos -- when song infos are present in the title. """ from bs4 import SoupStrainer, BeautifulSoup s = self.source url = six.text_type(s['url'] + s['path']) html = raw_backend.fetch_url(url) soup = BeautifulSoup(html, "html.parser", parse_only=SoupStrainer('title')) self.assertEqual( google.is_page_candidate(url, soup.title.string, s['title'], s['artist']), True, url) def test_is_page_candidate_fuzzy_match(self): """Test matching html page title with song infos -- when song infos are not present in the title. """ s = self.source url = s['url'] + s['path'] url_title = u'example.com | Beats song by John doe' # very small diffs (typo) are ok eg 'beats' vs 'beets' with same artist self.assertEqual(google.is_page_candidate(url, url_title, s['title'], s['artist']), True, url) # reject different title url_title = u'example.com | seets bong lyrics by John doe' self.assertEqual(google.is_page_candidate(url, url_title, s['title'], s['artist']), False, url) def test_is_page_candidate_special_chars(self): """Ensure that `is_page_candidate` doesn't crash when the artist and such contain special regular expression characters. """ # https://github.com/beetbox/beets/issues/1673 s = self.source url = s['url'] + s['path'] url_title = u'foo' google.is_page_candidate(url, url_title, s['title'], u'Sunn O)))') class SlugTests(unittest.TestCase): def test_slug(self): # plain ascii passthrough text = u"test" self.assertEqual(lyrics.slug(text), 'test') # german unicode and capitals text = u"Mørdag" self.assertEqual(lyrics.slug(text), 'mordag') # more accents and quotes text = u"l'été c'est fait pour jouer" self.assertEqual(lyrics.slug(text), 'l-ete-c-est-fait-pour-jouer') # accents, parens and spaces text = u"\xe7afe au lait (boisson)" self.assertEqual(lyrics.slug(text), 'cafe-au-lait-boisson') text = u"Multiple spaces -- and symbols! -- merged" self.assertEqual(lyrics.slug(text), 'multiple-spaces-and-symbols-merged') def suite(): return unittest.TestLoader().loadTestsFromName(__name__) if __name__ == '__main__': unittest.main(defaultTest='suite')