import re

from .common import InfoExtractor


class EchoMskIE(InfoExtractor):
    _VALID_URL = r'https?://(?:www\.)?echo\.msk\.ru/sounds/(?P<id>\d+)'
    _TEST = {
        'url': 'http://www.echo.msk.ru/sounds/1464134.html',
        'md5': '2e44b3b78daff5b458e4dbc37f191f7c',
        'info_dict': {
            'id': '1464134',
            'ext': 'mp3',
            'title': 'Особое мнение - 29 декабря 2014, 19:08',
        },
    }

    def _real_extract(self, url):
        video_id = self._match_id(url)

        webpage = self._download_webpage(url, video_id)

        audio_url = self._search_regex(
            r'<a rel="mp3" href="([^"]+)">', webpage, 'audio URL')

        title = self._html_search_regex(
            r'<a href="/programs/[^"]+" target="_blank">([^<]+)</a>',
            webpage, 'title')

        air_date = self._html_search_regex(
            r'(?s)<div class="date">(.+?)</div>',
            webpage, 'date', fatal=False, default=None)

        if air_date:
            air_date = re.sub(r'(\s)\1+', r'\1', air_date)
            if air_date:
                title = '%s - %s' % (title, air_date)

        return {
            'id': video_id,
            'url': audio_url,
            'title': title,
        }