usse/funda-scraper/venv/lib/python3.10/site-packages/rich/ansi.py

238 lines
6.7 KiB
Python
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

import re
import sys
from contextlib import suppress
from typing import Iterable, NamedTuple, Optional
from .color import Color
from .style import Style
from .text import Text
re_ansi = re.compile(
r"""
(?:\x1b\](.*?)\x1b\\)|
(?:\x1b([(@-Z\\-_]|\[[0-?]*[ -/]*[@-~]))
""",
re.VERBOSE,
)
class _AnsiToken(NamedTuple):
"""Result of ansi tokenized string."""
plain: str = ""
sgr: Optional[str] = ""
osc: Optional[str] = ""
def _ansi_tokenize(ansi_text: str) -> Iterable[_AnsiToken]:
"""Tokenize a string in to plain text and ANSI codes.
Args:
ansi_text (str): A String containing ANSI codes.
Yields:
AnsiToken: A named tuple of (plain, sgr, osc)
"""
position = 0
sgr: Optional[str]
osc: Optional[str]
for match in re_ansi.finditer(ansi_text):
start, end = match.span(0)
osc, sgr = match.groups()
if start > position:
yield _AnsiToken(ansi_text[position:start])
if sgr:
if sgr.endswith("m"):
yield _AnsiToken("", sgr[1:-1], osc)
else:
yield _AnsiToken("", sgr, osc)
position = end
if position < len(ansi_text):
yield _AnsiToken(ansi_text[position:])
SGR_STYLE_MAP = {
1: "bold",
2: "dim",
3: "italic",
4: "underline",
5: "blink",
6: "blink2",
7: "reverse",
8: "conceal",
9: "strike",
21: "underline2",
22: "not dim not bold",
23: "not italic",
24: "not underline",
25: "not blink",
26: "not blink2",
27: "not reverse",
28: "not conceal",
29: "not strike",
30: "color(0)",
31: "color(1)",
32: "color(2)",
33: "color(3)",
34: "color(4)",
35: "color(5)",
36: "color(6)",
37: "color(7)",
39: "default",
40: "on color(0)",
41: "on color(1)",
42: "on color(2)",
43: "on color(3)",
44: "on color(4)",
45: "on color(5)",
46: "on color(6)",
47: "on color(7)",
49: "on default",
51: "frame",
52: "encircle",
53: "overline",
54: "not frame not encircle",
55: "not overline",
90: "color(8)",
91: "color(9)",
92: "color(10)",
93: "color(11)",
94: "color(12)",
95: "color(13)",
96: "color(14)",
97: "color(15)",
100: "on color(8)",
101: "on color(9)",
102: "on color(10)",
103: "on color(11)",
104: "on color(12)",
105: "on color(13)",
106: "on color(14)",
107: "on color(15)",
}
class AnsiDecoder:
"""Translate ANSI code in to styled Text."""
def __init__(self) -> None:
self.style = Style.null()
def decode(self, terminal_text: str) -> Iterable[Text]:
"""Decode ANSI codes in an interable of lines.
Args:
lines (Iterable[str]): An iterable of lines of terminal output.
Yields:
Text: Marked up Text.
"""
for line in terminal_text.splitlines():
yield self.decode_line(line)
def decode_line(self, line: str) -> Text:
"""Decode a line containing ansi codes.
Args:
line (str): A line of terminal output.
Returns:
Text: A Text instance marked up according to ansi codes.
"""
from_ansi = Color.from_ansi
from_rgb = Color.from_rgb
_Style = Style
text = Text()
append = text.append
line = line.rsplit("\r", 1)[-1]
for plain_text, sgr, osc in _ansi_tokenize(line):
if plain_text:
append(plain_text, self.style or None)
elif osc is not None:
if osc.startswith("8;"):
_params, semicolon, link = osc[2:].partition(";")
if semicolon:
self.style = self.style.update_link(link or None)
elif sgr is not None:
# Translate in to semi-colon separated codes
# Ignore invalid codes, because we want to be lenient
codes = [
min(255, int(_code) if _code else 0)
for _code in sgr.split(";")
if _code.isdigit() or _code == ""
]
iter_codes = iter(codes)
for code in iter_codes:
if code == 0:
# reset
self.style = _Style.null()
elif code in SGR_STYLE_MAP:
# styles
self.style += _Style.parse(SGR_STYLE_MAP[code])
elif code == 38:
#  Foreground
with suppress(StopIteration):
color_type = next(iter_codes)
if color_type == 5:
self.style += _Style.from_color(
from_ansi(next(iter_codes))
)
elif color_type == 2:
self.style += _Style.from_color(
from_rgb(
next(iter_codes),
next(iter_codes),
next(iter_codes),
)
)
elif code == 48:
# Background
with suppress(StopIteration):
color_type = next(iter_codes)
if color_type == 5:
self.style += _Style.from_color(
None, from_ansi(next(iter_codes))
)
elif color_type == 2:
self.style += _Style.from_color(
None,
from_rgb(
next(iter_codes),
next(iter_codes),
next(iter_codes),
),
)
return text
if sys.platform != "win32" and __name__ == "__main__": # pragma: no cover
import io
import os
import pty
import sys
decoder = AnsiDecoder()
stdout = io.BytesIO()
def read(fd: int) -> bytes:
data = os.read(fd, 1024)
stdout.write(data)
return data
pty.spawn(sys.argv[1:], read)
from .console import Console
console = Console(record=True)
stdout_result = stdout.getvalue().decode("utf-8")
print(stdout_result)
for line in decoder.decode(stdout_result):
console.print(line)
console.save_html("stdout.html")