cam4-anonymous/progress_211.py

150 lines
4.6 KiB
Python
Raw Normal View History

import sys
from collections import deque
from shutil import get_terminal_size
from time import time
from streamlink_cli.compat import is_win32
PROGRESS_FORMATS = (
"[download] => {prefix} ({written} @ {speed}/s) ",
)
# widths generated from
# http://www.unicode.org/Public/4.0-Update/EastAsianWidth-4.0.0.txt
widths = [
(13, 1), (15, 0), (126, 1), (159, 0), (687, 1), (710, 0), # noqa: E241
(711, 1), (727, 0), (733, 1), (879, 0), (1154, 1), (1161, 0), # noqa: E241
(4347, 1), (4447, 2), (7467, 1), (7521, 0), (8369, 1), (8426, 0), # noqa: E241
(9000, 1), (9002, 2), (11021, 1), (12350, 2), (12351, 1), (12438, 2), # noqa: E241
(12442, 0), (19893, 2), (19967, 1), (55203, 2), (63743, 1), (64106, 2), # noqa: E241
(65039, 1), (65059, 0), (65131, 2), (65279, 1), (65376, 2), (65500, 1), # noqa: E241
(65510, 2), (120831, 1), (262141, 2), (1114109, 1) # noqa: E241
]
def get_width(o):
"""Returns the screen column width for unicode ordinal."""
for num, wid in widths:
if o <= num:
return wid
return 1
def terminal_width(value):
"""Returns the width of the string it would be when displayed."""
if isinstance(value, bytes):
value = value.decode("utf8", "ignore")
return sum(map(get_width, map(ord, value)))
def get_cut_prefix(value, max_len):
"""Drops Characters by unicode not by bytes."""
should_convert = isinstance(value, bytes)
if should_convert:
value = value.decode("utf8", "ignore")
for i in range(len(value)):
if terminal_width(value[i:]) <= max_len:
break
return value[i:].encode("utf8", "ignore") if should_convert else value[i:]
def print_inplace(msg):
"""Clears out the previous line and prints a new one."""
term_width = get_terminal_size().columns
spacing = term_width - terminal_width(msg)
# On windows we need one less space or we overflow the line for some reason.
if is_win32:
spacing -= 1
sys.stderr.write("\r{0}".format(msg))
sys.stderr.write(" " * max(0, spacing))
sys.stderr.flush()
def format_filesize(size):
"""Formats the file size into a human readable format."""
for suffix in ("bytes", "KB", "MB", "GB", "TB"):
if size < 1024.0:
if suffix in ("GB", "TB"):
return "{0:3.2f} {1}".format(size, suffix)
else:
return "{0:3.1f} {1}".format(size, suffix)
size /= 1024.0
def format_time(elapsed):
"""Formats elapsed seconds into a human readable format."""
hours = int(elapsed / (60 * 60))
minutes = int((elapsed % (60 * 60)) / 60)
seconds = int(elapsed % 60)
rval = ""
if hours:
rval += "{0}h".format(hours)
if elapsed > 60:
rval += "{0}m".format(minutes)
rval += "{0}s".format(seconds)
return rval
def create_status_line(**params):
"""Creates a status line with appropriate size."""
max_size = get_terminal_size().columns - 1
for fmt in PROGRESS_FORMATS:
status = fmt.format(**params)
if len(status) <= max_size:
break
return status
def progress(iterator, prefix):
"""Progress an iterator and updates a pretty status line to the terminal.
The status line contains:
- Amount of data read from the iterator
- Time elapsed
- Average speed, based on the last few seconds.
"""
if terminal_width(prefix) > 52:
prefix = (get_cut_prefix(prefix, 50))
speed_updated = start = time()
speed_written = written = 0
speed_history = deque(maxlen=5)
for data in iterator:
yield data
now = time()
elapsed = now - start
written += len(data)
speed_elapsed = now - speed_updated
if speed_elapsed >= 0.5:
speed_history.appendleft((
written - speed_written,
speed_updated,
))
speed_updated = now
speed_written = written
speed_history_written = sum(h[0] for h in speed_history)
speed_history_elapsed = now - speed_history[-1][1]
speed = speed_history_written / speed_history_elapsed
status = create_status_line(
prefix=prefix,
written=format_filesize(written),
elapsed=format_time(elapsed),
speed=format_filesize(speed)
)
print_inplace(status)
sys.stderr.write("\n")
sys.stderr.flush()