You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
bazarr/libs/pysubs2/substation.py

263 lines
9.9 KiB

from __future__ import print_function, division, unicode_literals
import re
from numbers import Number
from .formatbase import FormatBase
from .ssaevent import SSAEvent
from .ssastyle import SSAStyle
from .common import text_type, Color
from .time import make_time, ms_to_times, timestamp_to_ms, TIMESTAMP
SSA_ALIGNMENT = (1, 2, 3, 9, 10, 11, 5, 6, 7)
def ass_to_ssa_alignment(i):
return SSA_ALIGNMENT[i-1]
def ssa_to_ass_alignment(i):
return SSA_ALIGNMENT.index(i) + 1
SECTION_HEADING = re.compile(r"^.{,3}\[[^\]]+\]") # allow for UTF-8 BOM, which is 3 bytes
STYLE_FORMAT_LINE = {
"ass": "Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic,"
" Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment,"
" MarginL, MarginR, MarginV, Encoding",
"ssa": "Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, TertiaryColour, BackColour, Bold, Italic,"
" BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, AlphaLevel, Encoding"
}
STYLE_FIELDS = {
"ass": ["fontname", "fontsize", "primarycolor", "secondarycolor", "outlinecolor", "backcolor", "bold", "italic",
"underline", "strikeout", "scalex", "scaley", "spacing", "angle", "borderstyle", "outline", "shadow",
"alignment", "marginl", "marginr", "marginv", "encoding"],
"ssa": ["fontname", "fontsize", "primarycolor", "secondarycolor", "tertiarycolor", "backcolor", "bold", "italic",
"borderstyle", "outline", "shadow", "alignment", "marginl", "marginr", "marginv", "alphalevel", "encoding"]
}
EVENT_FORMAT_LINE = {
"ass": "Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text",
"ssa": "Format: Marked, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text"
}
EVENT_FIELDS = {
"ass": ["layer", "start", "end", "style", "name", "marginl", "marginr", "marginv", "effect", "text"],
"ssa": ["marked", "start", "end", "style", "name", "marginl", "marginr", "marginv", "effect", "text"]
}
#: Largest timestamp allowed in SubStation, ie. 9:59:59.99.
MAX_REPRESENTABLE_TIME = make_time(h=10) - 10
def ms_to_timestamp(ms):
"""Convert ms to 'H:MM:SS.cc'"""
# XXX throw on overflow/underflow?
if ms < 0: ms = 0
if ms > MAX_REPRESENTABLE_TIME: ms = MAX_REPRESENTABLE_TIME
h, m, s, ms = ms_to_times(ms)
return "%01d:%02d:%02d.%02d" % (h, m, s, ms//10)
def color_to_ass_rgba(c):
return "&H%08X" % ((c.a << 24) | (c.b << 16) | (c.g << 8) | c.r)
def color_to_ssa_rgb(c):
return "%d" % ((c.b << 16) | (c.g << 8) | c.r)
def ass_rgba_to_color(s):
x = int(s[2:], base=16)
r = x & 0xff
g = (x >> 8) & 0xff
b = (x >> 16) & 0xff
a = (x >> 24) & 0xff
return Color(r, g, b, a)
def ssa_rgb_to_color(s):
x = int(s)
r = x & 0xff
g = (x >> 8) & 0xff
b = (x >> 16) & 0xff
return Color(r, g, b)
def is_valid_field_content(s):
"""
Returns True if string s can be stored in a SubStation field.
Fields are written in CSV-like manner, thus commas and/or newlines
are not acceptable in the string.
"""
return "\n" not in s and "," not in s
def parse_tags(text, style=SSAStyle.DEFAULT_STYLE, styles={}):
"""
Split text into fragments with computed SSAStyles.
Returns list of tuples (fragment, style), where fragment is a part of text
between two brace-delimited override sequences, and style is the computed
styling of the fragment, ie. the original style modified by all override
sequences before the fragment.
Newline and non-breakable space overrides are left as-is.
Supported override tags:
- i, b, u, s
- r (with or without style name)
"""
fragments = SSAEvent.OVERRIDE_SEQUENCE.split(text)
if len(fragments) == 1:
return [(text, style)]
def apply_overrides(all_overrides):
s = style.copy()
for tag in re.findall(r"\\[ibus][10]|\\r[a-zA-Z_0-9 ]*", all_overrides):
if tag == r"\r":
s = style.copy() # reset to original line style
elif tag.startswith(r"\r"):
name = tag[2:]
if name in styles:
s = styles[name].copy() # reset to named style
else:
if "i" in tag: s.italic = "1" in tag
elif "b" in tag: s.bold = "1" in tag
elif "u" in tag: s.underline = "1" in tag
elif "s" in tag: s.strikeout = "1" in tag
return s
overrides = SSAEvent.OVERRIDE_SEQUENCE.findall(text)
overrides_prefix_sum = ["".join(overrides[:i]) for i in range(len(overrides) + 1)]
computed_styles = map(apply_overrides, overrides_prefix_sum)
return list(zip(fragments, computed_styles))
NOTICE = "Script generated by pysubs2\nhttps://pypi.python.org/pypi/pysubs2"
class SubstationFormat(FormatBase):
@classmethod
def guess_format(cls, text):
if "V4+ Styles" in text:
return "ass"
elif "V4 Styles" in text:
return "ssa"
@classmethod
def from_file(cls, subs, fp, format_, **kwargs):
def string_to_field(f, v):
if f in {"start", "end"}:
return timestamp_to_ms(TIMESTAMP.match(v).groups())
elif "color" in f:
if format_ == "ass":
return ass_rgba_to_color(v)
else:
try:
return ssa_rgb_to_color(v)
except ValueError:
try:
return ass_rgba_to_color(v)
except:
return Color(255, 255, 255, 0)
elif f in {"bold", "underline", "italic", "strikeout"}:
return v == "-1"
elif f in {"borderstyle", "encoding", "marginl", "marginr", "marginv", "layer", "alphalevel"}:
return int(v)
elif f in {"fontsize", "scalex", "scaley", "spacing", "angle", "outline", "shadow"}:
return float(v)
elif f == "marked":
return v.endswith("1")
elif f == "alignment":
i = int(v)
if format_ == "ass":
return i
else:
return ssa_to_ass_alignment(i)
else:
return v
subs.info.clear()
subs.aegisub_project.clear()
subs.styles.clear()
inside_info_section = False
inside_aegisub_section = False
for line in fp:
line = line.strip()
if SECTION_HEADING.match(line):
inside_info_section = "Info" in line
inside_aegisub_section = "Aegisub" in line
elif inside_info_section or inside_aegisub_section:
if line.startswith(";"): continue # skip comments
try:
k, v = line.split(": ", 1)
if inside_info_section:
subs.info[k] = v
elif inside_aegisub_section:
subs.aegisub_project[k] = v
except ValueError:
pass
elif line.startswith("Style:"):
_, rest = line.split(": ", 1)
buf = rest.strip().split(",")
name, raw_fields = buf[0], buf[1:] # splat workaround for Python 2.7
field_dict = {f: string_to_field(f, v) for f, v in zip(STYLE_FIELDS[format_], raw_fields)}
sty = SSAStyle(**field_dict)
subs.styles[name] = sty
elif line.startswith("Dialogue:") or line.startswith("Comment:"):
ev_type, rest = line.split(": ", 1)
raw_fields = rest.strip().split(",", len(EVENT_FIELDS[format_])-1)
field_dict = {f: string_to_field(f, v) for f, v in zip(EVENT_FIELDS[format_], raw_fields)}
field_dict["type"] = ev_type
ev = SSAEvent(**field_dict)
subs.events.append(ev)
@classmethod
def to_file(cls, subs, fp, format_, header_notice=NOTICE, **kwargs):
print("[Script Info]", file=fp)
for line in header_notice.splitlines(False):
print(";", line, file=fp)
subs.info["ScriptType"] = "v4.00+" if format_ == "ass" else "v4.00"
for k, v in subs.info.items():
print(k, v, sep=": ", file=fp)
if subs.aegisub_project:
print("\n[Aegisub Project Garbage]", file=fp)
for k, v in subs.aegisub_project.items():
print(k, v, sep=": ", file=fp)
def field_to_string(f, v):
if f in {"start", "end"}:
return ms_to_timestamp(v)
elif f == "marked":
return "Marked=%d" % v
elif f == "alignment" and format_ == "ssa":
return text_type(ass_to_ssa_alignment(v))
elif isinstance(v, bool):
return "-1" if v else "0"
elif isinstance(v, (text_type, Number)):
return text_type(v)
elif isinstance(v, Color):
if format_ == "ass":
return color_to_ass_rgba(v)
else:
return color_to_ssa_rgb(v)
else:
raise TypeError("Unexpected type when writing a SubStation field")
print("\n[V4+ Styles]" if format_ == "ass" else "\n[V4 Styles]", file=fp)
print(STYLE_FORMAT_LINE[format_], file=fp)
for name, sty in subs.styles.items():
fields = [field_to_string(f, getattr(sty, f)) for f in STYLE_FIELDS[format_]]
print("Style: %s" % name, *fields, sep=",", file=fp)
print("\n[Events]", file=fp)
print(EVENT_FORMAT_LINE[format_], file=fp)
for ev in subs.events:
fields = [field_to_string(f, getattr(ev, f)) for f in EVENT_FIELDS[format_]]
print(ev.type, end=": ", file=fp)
print(*fields, sep=",", file=fp)