X-Git-Url: http://git.bitcoin.ninja/index.cgi?a=blobdiff_plain;f=youtube_dl%2Fdownloader%2Ff4m.py;h=0e7a1c20075499e58b977da4154ce287b144f958;hb=16e7711e22648027739096560914a976b8eea786;hp=c460c167a2db78be7a1cdb8e81a3efe89e677ed1;hpb=bc694039e47cc871c98abacdf1c0a2e5a257a8a4;p=youtube-dl diff --git a/youtube_dl/downloader/f4m.py b/youtube_dl/downloader/f4m.py index c460c167a..0e7a1c200 100644 --- a/youtube_dl/downloader/f4m.py +++ b/youtube_dl/downloader/f4m.py @@ -177,13 +177,12 @@ def build_fragments_list(boot_info): """ Return a list of (segment, fragment) for each fragment in the video """ res = [] segment_run_table = boot_info['segments'][0] - # I've only found videos with one segment - segment_run_entry = segment_run_table['segment_run'][0] - n_frags = segment_run_entry[1] fragment_run_entry_table = boot_info['fragments'][0]['fragments'] first_frag_number = fragment_run_entry_table[0]['first'] - for (i, frag_number) in zip(range(1, n_frags + 1), itertools.count(first_frag_number)): - res.append((1, frag_number)) + fragments_counter = itertools.count(first_frag_number) + for segment, fragments_count in segment_run_table['segment_run']: + for _ in range(fragments_count): + res.append((segment, next(fragments_counter))) return res @@ -231,6 +230,23 @@ class F4mFD(FileDownloader): A downloader for f4m manifests or AdobeHDS. """ + def _get_unencrypted_media(self, doc): + media = doc.findall(_add_ns('media')) + if not media: + self.report_error('No media found') + for e in (doc.findall(_add_ns('drmAdditionalHeader')) + + doc.findall(_add_ns('drmAdditionalHeaderSet'))): + # If id attribute is missing it's valid for all media nodes + # without drmAdditionalHeaderId or drmAdditionalHeaderSetId attribute + if 'id' not in e.attrib: + self.report_error('Missing ID in f4m DRM') + media = list(filter(lambda e: 'drmAdditionalHeaderId' not in e.attrib and + 'drmAdditionalHeaderSetId' not in e.attrib, + media)) + if not media: + self.report_error('Unsupported DRM') + return media + def real_download(self, filename, info_dict): man_url = info_dict['url'] requested_bitrate = info_dict.get('tbr') @@ -249,7 +265,8 @@ class F4mFD(FileDownloader): ) doc = etree.fromstring(manifest) - formats = [(int(f.attrib.get('bitrate', -1)), f) for f in doc.findall(_add_ns('media'))] + formats = [(int(f.attrib.get('bitrate', -1)), f) + for f in self._get_unencrypted_media(doc)] if requested_bitrate is None: # get the best format formats = sorted(formats, key=lambda f: f[0])