1
0
Fork 0
mirror of https://github.com/morpheus65535/bazarr synced 2024-12-27 10:07:22 +00:00
bazarr/libs/srt_tools/utils.py
Michiel van Baak Jansen 4a0932b5d3
Update ffsubsync and srt module
* Update ffsubsync to 0.4.11
* Update srt to 3.4.1
2021-04-13 00:02:29 -04:00

218 lines
6.8 KiB
Python

#!/usr/bin/env python
import argparse
import codecs
import srt
import logging
import sys
import itertools
import collections
import os
PROG_NAME = os.path.basename(sys.argv[0]).replace("-", " ", 1)
STDIN_BYTESTREAM = getattr(sys.stdin, "buffer", sys.stdin)
STDOUT_BYTESTREAM = getattr(sys.stdout, "buffer", sys.stdout)
DASH_STREAM_MAP = {"input": STDIN_BYTESTREAM, "output": STDOUT_BYTESTREAM}
log = logging.getLogger(__name__)
def noop(stream):
"""
Used when we didn't explicitly specify a stream to avoid using
codecs.get{reader,writer}
"""
return stream
def dash_to_stream(arg, arg_type):
if arg == "-":
return DASH_STREAM_MAP[arg_type]
return arg
def basic_parser(
description=None,
multi_input=False,
no_output=False,
examples=None,
hide_no_strict=False,
):
example_lines = []
if examples is not None:
example_lines.append("examples:")
for desc, code in examples.items():
example_lines.append(" {}".format(desc))
example_lines.append(" $ {}\n".format(code))
parser = argparse.ArgumentParser(
prog=PROG_NAME,
description=description,
epilog="\n".join(example_lines),
formatter_class=argparse.RawDescriptionHelpFormatter,
)
# Cannot use argparse.FileType as we need to know the encoding from the
# args
if multi_input:
parser.add_argument(
"--input",
"-i",
metavar="FILE",
action="append",
type=lambda arg: dash_to_stream(arg, "input"),
help="the files to process",
required=True,
)
else:
parser.add_argument(
"--input",
"-i",
metavar="FILE",
default=STDIN_BYTESTREAM,
type=lambda arg: dash_to_stream(arg, "input"),
help="the file to process (default: stdin)",
)
if not no_output:
parser.add_argument(
"--output",
"-o",
metavar="FILE",
default=STDOUT_BYTESTREAM,
type=lambda arg: dash_to_stream(arg, "output"),
help="the file to write to (default: stdout)",
)
if not multi_input:
parser.add_argument(
"--inplace",
"-p",
action="store_true",
help="modify file in place",
)
shelp = "allow blank lines in output, your media player may explode"
if hide_no_strict:
shelp = argparse.SUPPRESS
parser.add_argument("--no-strict", action="store_false", dest="strict", help=shelp)
parser.add_argument(
"--debug",
action="store_const",
dest="log_level",
const=logging.DEBUG,
default=logging.INFO,
help="enable debug logging",
)
parser.add_argument(
"--ignore-parsing-errors",
"-c",
action="store_true",
help="try to keep going, even if there are parsing errors",
)
parser.add_argument(
"--encoding", "-e", help="the encoding to read/write files in (default: utf8)"
)
return parser
def set_basic_args(args):
# TODO: dedupe some of this
if getattr(args, "inplace", None):
if args.input == DASH_STREAM_MAP["input"]:
raise ValueError("Cannot use --inplace on stdin")
if args.output != DASH_STREAM_MAP["output"]:
raise ValueError("Cannot use -o and -p together")
args.output = args.input
for stream_name in ("input", "output"):
log.debug('Processing stream "%s"', stream_name)
try:
stream = getattr(args, stream_name)
except AttributeError:
# For example, in the case of no_output
continue
# We don't use system default encoding, because usually one runs this
# on files they got from elsewhere. As such, be opinionated that these
# files are probably UTF-8. Looking for the BOM on reading allows us to
# be more liberal with what we accept, without adding BOMs on write.
read_encoding = args.encoding or "utf-8-sig"
write_encoding = args.encoding or "utf-8"
r_enc = codecs.getreader(read_encoding)
w_enc = codecs.getwriter(write_encoding)
log.debug("Got %r as stream", stream)
# We don't use encoding= option to open because we want to have the
# same universal newlines behaviour as STD{IN,OUT}_BYTESTREAM
if stream in DASH_STREAM_MAP.values():
log.debug("%s in DASH_STREAM_MAP", stream_name)
if stream is args.input:
args.input = srt.parse(
r_enc(args.input).read(), ignore_errors=args.ignore_parsing_errors
)
elif stream is args.output:
# Since args.output is not in text mode (since we didn't
# earlier know the encoding), we have no universal newline
# support and need to do it ourselves
args.output = w_enc(args.output)
else:
log.debug("%s not in DASH_STREAM_MAP", stream_name)
if stream is args.input:
if isinstance(args.input, collections.MutableSequence):
for i, input_fn in enumerate(args.input):
if input_fn in DASH_STREAM_MAP.values():
if stream is args.input:
args.input[i] = srt.parse(
r_enc(input_fn).read(),
ignore_errors=args.ignore_parsing_errors,
)
else:
f = r_enc(open(input_fn, "rb"))
with f:
args.input[i] = srt.parse(
f.read(), ignore_errors=args.ignore_parsing_errors
)
else:
f = r_enc(open(stream, "rb"))
with f:
args.input = srt.parse(
f.read(), ignore_errors=args.ignore_parsing_errors
)
else:
args.output = w_enc(open(args.output, "wb"))
def compose_suggest_on_fail(subs, strict=True):
try:
return srt.compose(subs, strict=strict, eol=os.linesep, in_place=True)
except srt.SRTParseError as thrown_exc:
# Since `subs` is actually a generator
log.critical(
"Parsing failed, maybe you need to pass a different encoding "
"with --encoding?"
)
raise
def sliding_window(seq, width=2):
seq_iter = iter(seq)
sliced = tuple(itertools.islice(seq_iter, width))
if len(sliced) == width:
yield sliced
for elem in seq_iter:
sliced = sliced[1:] + (elem,)
yield sliced