Implement lazy chunk downloading

using `http_dash_segments_generator` as seen in youtube live-from-start
this way it only goes through the chunks etc one a format has actually been selected
so we don't have it downloading eg 20 chunks of every stream only for most to be discarded

closes #29

2 files changed, 40 insertions, 20 deletions

diff --git a/yt_dlp_plugins/extractor/radiko.py b/yt_dlp_plugins/extractor/radiko.py
index 6718201..2ec363a 100644
--- a/yt_dlp_plugins/extractor/radiko.py
+++ b/yt_dlp_plugins/extractor/radiko.py
@@ -310,13 +310,15 @@ class _RadikoBaseIE(InfoExtractor):
 
 			if delivered_live and timefree and do_as_live_chunks:
 
-				chunks_playlist = hacks._generate_as_live_playlist(
-					self, playlist_url, start_at, end_at, domain, auth_headers
-				)
+				def fragments_generator(_):
+					return hacks._generate_as_live_fragments(
+						self, playlist_url, start_at, end_at, domain, auth_headers
+					)
 
 				m3u8_formats = [{
 					"format_id": join_nonempty(domain, "chunked"),
-					"hls_media_playlist_data": chunks_playlist,
+					"fragments": fragments_generator,
+					"protocol": "http_dash_segments_generator",
 					"preference": preference,
 					"ext": "m4a",
 					"vcodec": "none",
@@ -324,6 +326,7 @@ class _RadikoBaseIE(InfoExtractor):
 					# fallback to live for ffmpeg etc
 					"url": playlist_url,
 					"http_headers": auth_headers,
+					"is_live": "yesn't",
 				}]
 				format_note.append("Chunked")
 			else:
diff --git a/yt_dlp_plugins/extractor/radiko_hacks.py b/yt_dlp_plugins/extractor/radiko_hacks.py
index 6486034..418aa70 100644
--- a/yt_dlp_plugins/extractor/radiko_hacks.py
+++ b/yt_dlp_plugins/extractor/radiko_hacks.py
@@ -10,8 +10,8 @@ from yt_dlp.utils import (
 
 # "hacks" as in great jank/schizo shit that works anyway
 
-def _generate_as_live_playlist(self, playlist_url, start_at, end_at, domain, headers={}):
-	playlist = ""
+def _generate_as_live_fragments(self, playlist_base_url, start_at, end_at, domain, headers={}):
+	playlist = []
 	chunk_length = 300  # max the api allows
 
 	duration = int(end_at.timestamp() - start_at.timestamp())
@@ -21,18 +21,19 @@ def _generate_as_live_playlist(self, playlist_url, start_at, end_at, domain, hea
 		chunk_length = min(chunk_length, duration - cursor)
 
 		chunk_start = start_at + datetime.timedelta(seconds=cursor)
-		chunk_url = update_url_query(playlist_url, {
+		chunk_url = update_url_query(playlist_base_url, {
 			"seek": chunk_start.timestring(),
 			"l": chunk_length,
 		})
 
-		chunk_playlist, real_chunk_length = _get_chunk_playlist(self, chunk_url, domain, chunk_num, headers)
+		chunk_fragments, real_chunk_length = _get_chunk_playlist(self, chunk_url, domain, chunk_num, headers)
 
-		playlist += chunk_playlist
-		cursor += real_chunk_length
+		cursor += round(real_chunk_length)
 		chunk_num += 1
 
-	return playlist
+		for frag in chunk_fragments:
+			yield frag
+
 
 def _get_chunk_playlist(self, chunk_url, src_id, chunk_num, headers={}):
 	EXTINF_duration = re.compile(r"^#EXTINF:([\d.]+),", flags=re.MULTILINE)
@@ -41,20 +42,23 @@ def _get_chunk_playlist(self, chunk_url, src_id, chunk_num, headers={}):
 	chunk_id = join_nonempty(src_id, chunk_num)
 	base_format = self._extract_m3u8_formats(
 		chunk_url, chunk_id, fatal=False, headers=headers,
-		note=f"Preparing {src_id} chunk {chunk_num}"
+#		note=f"Preparing {src_id} chunk {chunk_num}"
+		note=False,
+		errnote=f"Failed to get {src_id} chunk {chunk_num} base format",
 	)
 	m3u8_url = traverse_obj(base_format, (..., "url",), get_all=False)
-	playlist = self._download_webpage(m3u8_url, chunk_id, note=f"Getting {src_id} chunk {chunk_num} fragments")
+	playlist = self._download_webpage(m3u8_url, chunk_id, note=False, errnote=f"Failed to get {src_id} chunk {chunk_num} playlist")
+	#note=f"Getting {src_id} chunk {chunk_num} fragments")
+
+	return _parse_hls(playlist)
 
-	real_duration = 0
-	for i in EXTINF_duration.findall(playlist):
-		real_duration += float(i)
-	real_duration = round(real_duration)
+def _parse_hls(m3u8_doc):
+	fragments = []
 
 	# playlists can sometimes be longer than they should
-	# wowza stream does some strange things
+	# e.g. wowza stream does some strange things
 	# it goes along fine with every fragment 5s long as normal
-	# and then during the ad break it does one with a different length (2s here)
+	# and then during the ad break it does one with a different length (eg 2s)
 	# i assume so they have a clean split to do ad insertion in? idk
 
 	# but anyway now the chunks aren't always a clean 5mins long
@@ -62,4 +66,17 @@ def _get_chunk_playlist(self, chunk_url, src_id, chunk_num, headers={}):
 
 	# so to work around this, we track the real duration from the #EXTINF tags
 
-	return playlist, real_duration
+	playlist_duration = 0
+	fragment_duration = None
+	for line in m3u8_doc.splitlines():
+		if line.startswith("#EXTINF:"):
+			fragment_duration = float(line[len('#EXTINF:'):].split(',')[0])  # from common._parse_m3u8_vod_duration
+			continue
+		elif line.startswith("#"):
+			continue
+
+		fragments.append({"url": line, "duration": fragment_duration})
+		playlist_duration += fragment_duration or 0
+		fragment_duration = None
+
+	return fragments, playlist_duration