mirror of
https://github.com/ihabunek/twitch-dl
synced 2024-08-30 18:32:25 +00:00
261 lines
7.9 KiB
Python
261 lines
7.9 KiB
Python
import m3u8
|
|
import re
|
|
import requests
|
|
import shutil
|
|
import subprocess
|
|
import tempfile
|
|
|
|
from os import path
|
|
from pathlib import Path
|
|
from urllib.parse import urlparse
|
|
|
|
from twitchdl import twitch, utils
|
|
from twitchdl.download import download_file, download_files
|
|
from twitchdl.exceptions import ConsoleError
|
|
from twitchdl.output import print_out
|
|
|
|
|
|
def _parse_playlists(playlists_m3u8):
|
|
playlists = m3u8.loads(playlists_m3u8)
|
|
|
|
for p in playlists.playlists:
|
|
name = p.media[0].name if p.media else ""
|
|
resolution = "x".join(str(r) for r in p.stream_info.resolution)
|
|
yield name, resolution, p.uri
|
|
|
|
|
|
def _get_playlist_by_name(playlists, quality):
|
|
if quality == "source":
|
|
_, _, uri = playlists[0]
|
|
return uri
|
|
|
|
for name, _, uri in playlists:
|
|
if name == quality:
|
|
return uri
|
|
|
|
available = ", ".join([name for (name, _, _) in playlists])
|
|
msg = "Quality '{}' not found. Available qualities are: {}".format(quality, available)
|
|
raise ConsoleError(msg)
|
|
|
|
|
|
def _select_playlist_interactive(playlists):
|
|
print_out("\nAvailable qualities:")
|
|
for n, (name, resolution, uri) in enumerate(playlists):
|
|
print_out("{}) {} [{}]".format(n + 1, name, resolution))
|
|
|
|
no = utils.read_int("Choose quality", min=1, max=len(playlists) + 1, default=1)
|
|
_, _, uri = playlists[no - 1]
|
|
return uri
|
|
|
|
|
|
def _join_vods(playlist_path, target, overwrite):
|
|
command = [
|
|
"ffmpeg",
|
|
"-i", playlist_path,
|
|
"-c", "copy",
|
|
target,
|
|
"-stats",
|
|
"-loglevel", "warning",
|
|
]
|
|
|
|
if overwrite:
|
|
command.append("-y")
|
|
|
|
print_out("<dim>{}</dim>".format(" ".join(command)))
|
|
result = subprocess.run(command)
|
|
if result.returncode != 0:
|
|
raise ConsoleError("Joining files failed")
|
|
|
|
|
|
def _video_target_filename(video, format):
|
|
match = re.search(r"^(\d{4})-(\d{2})-(\d{2})T", video['publishedAt'])
|
|
date = "".join(match.groups())
|
|
|
|
name = "_".join([
|
|
date,
|
|
video['id'][1:],
|
|
video['creator']['login'],
|
|
utils.slugify(video['title']),
|
|
])
|
|
|
|
return name + "." + format
|
|
|
|
|
|
def _clip_target_filename(clip):
|
|
url = clip["videoQualities"][0]["sourceURL"]
|
|
_, ext = path.splitext(url)
|
|
ext = ext.lstrip(".")
|
|
|
|
match = re.search(r"^(\d{4})-(\d{2})-(\d{2})T", clip["createdAt"])
|
|
date = "".join(match.groups())
|
|
|
|
name = "_".join([
|
|
date,
|
|
clip["id"],
|
|
clip["broadcaster"]["login"],
|
|
utils.slugify(clip["title"]),
|
|
])
|
|
|
|
return "{}.{}".format(name, ext)
|
|
|
|
|
|
def _get_vod_paths(playlist, start, end):
|
|
"""Extract unique VOD paths for download from playlist."""
|
|
files = []
|
|
vod_start = 0
|
|
for segment in playlist.segments:
|
|
vod_end = vod_start + segment.duration
|
|
|
|
# `vod_end > start` is used here becuase it's better to download a bit
|
|
# more than a bit less, similar for the end condition
|
|
start_condition = not start or vod_end > start
|
|
end_condition = not end or vod_start < end
|
|
|
|
if start_condition and end_condition and segment.uri not in files:
|
|
files.append(segment.uri)
|
|
|
|
vod_start = vod_end
|
|
|
|
return files
|
|
|
|
|
|
def _crete_temp_dir(base_uri):
|
|
"""Create a temp dir to store downloads if it doesn't exist."""
|
|
path = urlparse(base_uri).path.lstrip("/")
|
|
temp_dir = Path(tempfile.gettempdir(), "twitch-dl", path)
|
|
temp_dir.mkdir(parents=True, exist_ok=True)
|
|
return temp_dir
|
|
|
|
|
|
def download(args):
|
|
video_id = utils.parse_video_identifier(args.video)
|
|
if video_id:
|
|
return _download_video(video_id, args)
|
|
|
|
clip_slug = utils.parse_clip_identifier(args.video)
|
|
if clip_slug:
|
|
return _download_clip(clip_slug, args)
|
|
|
|
raise ConsoleError("Invalid input: {}".format(args.video))
|
|
|
|
|
|
def _get_clip_url(clip, args):
|
|
qualities = clip["videoQualities"]
|
|
|
|
# Quality given as an argument
|
|
if args.quality:
|
|
if args.quality == "source":
|
|
return qualities[0]["sourceURL"]
|
|
|
|
selected_quality = args.quality.rstrip("p") # allow 720p as well as 720
|
|
for q in qualities:
|
|
if q["quality"] == selected_quality:
|
|
return q["sourceURL"]
|
|
|
|
available = ", ".join([str(q["quality"]) for q in qualities])
|
|
msg = "Quality '{}' not found. Available qualities are: {}".format(args.quality, available)
|
|
raise ConsoleError(msg)
|
|
|
|
# Ask user to select quality
|
|
print_out("\nAvailable qualities:")
|
|
for n, q in enumerate(qualities):
|
|
print_out("{}) {} [{} fps]".format(n + 1, q["quality"], q["frameRate"]))
|
|
print_out()
|
|
|
|
no = utils.read_int("Choose quality", min=1, max=len(qualities), default=1)
|
|
selected_quality = qualities[no - 1]
|
|
return selected_quality["sourceURL"]
|
|
|
|
|
|
def _download_clip(slug, args):
|
|
print_out("<dim>Looking up clip...</dim>")
|
|
clip = twitch.get_clip(slug)
|
|
|
|
if not clip:
|
|
raise ConsoleError("Clip '{}' not found".format(slug))
|
|
|
|
print_out("Found: <green>{}</green> by <yellow>{}</yellow>, playing <blue>{}</blue> ({})".format(
|
|
clip["title"],
|
|
clip["broadcaster"]["displayName"],
|
|
clip["game"]["name"],
|
|
utils.format_duration(clip["durationSeconds"])
|
|
))
|
|
|
|
url = _get_clip_url(clip, args)
|
|
print_out("<dim>Selected URL: {}</dim>".format(url))
|
|
|
|
target = _clip_target_filename(clip)
|
|
|
|
print_out("Downloading clip...")
|
|
download_file(url, target)
|
|
|
|
print_out("Downloaded: {}".format(target))
|
|
|
|
|
|
def _download_video(video_id, args):
|
|
if args.start and args.end and args.end <= args.start:
|
|
raise ConsoleError("End time must be greater than start time")
|
|
|
|
print_out("<dim>Looking up video...</dim>")
|
|
video = twitch.get_video(video_id)
|
|
|
|
print_out("Found: <blue>{}</blue> by <yellow>{}</yellow>".format(
|
|
video['title'], video['creator']['displayName']))
|
|
|
|
print_out("<dim>Fetching access token...</dim>")
|
|
access_token = twitch.get_access_token(video_id)
|
|
|
|
print_out("<dim>Fetching playlists...</dim>")
|
|
playlists_m3u8 = twitch.get_playlists(video_id, access_token)
|
|
playlists = list(_parse_playlists(playlists_m3u8))
|
|
playlist_uri = (_get_playlist_by_name(playlists, args.quality) if args.quality
|
|
else _select_playlist_interactive(playlists))
|
|
|
|
print_out("<dim>Fetching playlist...</dim>")
|
|
response = requests.get(playlist_uri)
|
|
response.raise_for_status()
|
|
playlist = m3u8.loads(response.text)
|
|
|
|
base_uri = re.sub("/[^/]+$", "/", playlist_uri)
|
|
target_dir = _crete_temp_dir(base_uri)
|
|
vod_paths = _get_vod_paths(playlist, args.start, args.end)
|
|
|
|
# Save playlists for debugging purposes
|
|
with open(path.join(target_dir, "playlists.m3u8"), "w") as f:
|
|
f.write(playlists_m3u8)
|
|
with open(path.join(target_dir, "playlist.m3u8"), "w") as f:
|
|
f.write(response.text)
|
|
|
|
print_out("\nDownloading {} VODs using {} workers to {}".format(
|
|
len(vod_paths), args.max_workers, target_dir))
|
|
path_map = download_files(base_uri, target_dir, vod_paths, args.max_workers)
|
|
|
|
# Make a modified playlist which references downloaded VODs
|
|
# Keep only the downloaded segments and skip the rest
|
|
org_segments = playlist.segments.copy()
|
|
playlist.segments.clear()
|
|
for segment in org_segments:
|
|
if segment.uri in path_map:
|
|
segment.uri = path_map[segment.uri]
|
|
playlist.segments.append(segment)
|
|
|
|
playlist_path = path.join(target_dir, "playlist_downloaded.m3u8")
|
|
playlist.dump(playlist_path)
|
|
|
|
if args.no_join:
|
|
print_out("\n\n<dim>Skipping joining files...</dim>")
|
|
print_out("VODs downloaded to:\n<blue>{}</blue>".format(target_dir))
|
|
return
|
|
|
|
print_out("\n\nJoining files...")
|
|
target = _video_target_filename(video, args.format)
|
|
_join_vods(playlist_path, target, args.overwrite)
|
|
|
|
if args.keep:
|
|
print_out("\n<dim>Temporary files not deleted: {}</dim>".format(target_dir))
|
|
else:
|
|
print_out("\n<dim>Deleting temporary files...</dim>")
|
|
shutil.rmtree(target_dir)
|
|
|
|
print_out("\nDownloaded: <green>{}</green>".format(target))
|