mirror of https://github.com/morpheus65535/bazarr
211 lines
7.4 KiB
Python
211 lines
7.4 KiB
Python
"""
|
|
pygments.lexers.mime
|
|
~~~~~~~~~~~~~~~~~~~~
|
|
|
|
Lexer for Multipurpose Internet Mail Extensions (MIME) data.
|
|
|
|
:copyright: Copyright 2006-2021 by the Pygments team, see AUTHORS.
|
|
:license: BSD, see LICENSE for details.
|
|
"""
|
|
|
|
import re
|
|
|
|
from pygments.lexer import RegexLexer, include
|
|
from pygments.lexers import get_lexer_for_mimetype
|
|
from pygments.token import Text, Name, String, Operator, Comment, Other
|
|
from pygments.util import get_int_opt, ClassNotFound
|
|
|
|
__all__ = ["MIMELexer"]
|
|
|
|
|
|
class MIMELexer(RegexLexer):
|
|
"""
|
|
Lexer for Multipurpose Internet Mail Extensions (MIME) data. This lexer is
|
|
designed to process nested multipart data.
|
|
|
|
It assumes that the given data contains both header and body (and is
|
|
split at an empty line). If no valid header is found, then the entire data
|
|
will be treated as body.
|
|
|
|
Additional options accepted:
|
|
|
|
`MIME-max-level`
|
|
Max recursion level for nested MIME structure. Any negative number
|
|
would treated as unlimited. (default: -1)
|
|
|
|
`Content-Type`
|
|
Treat the data as a specific content type. Useful when header is
|
|
missing, or this lexer would try to parse from header. (default:
|
|
`text/plain`)
|
|
|
|
`Multipart-Boundary`
|
|
Set the default multipart boundary delimiter. This option is only used
|
|
when `Content-Type` is `multipart` and header is missing. This lexer
|
|
would try to parse from header by default. (default: None)
|
|
|
|
`Content-Transfer-Encoding`
|
|
Treat the data as a specific encoding. Or this lexer would try to parse
|
|
from header by default. (default: None)
|
|
|
|
.. versionadded:: 2.5
|
|
"""
|
|
|
|
name = "MIME"
|
|
aliases = ["mime"]
|
|
mimetypes = ["multipart/mixed",
|
|
"multipart/related",
|
|
"multipart/alternative"]
|
|
|
|
def __init__(self, **options):
|
|
super().__init__(**options)
|
|
self.boundary = options.get("Multipart-Boundary")
|
|
self.content_transfer_encoding = options.get("Content_Transfer_Encoding")
|
|
self.content_type = options.get("Content_Type", "text/plain")
|
|
self.max_nested_level = get_int_opt(options, "MIME-max-level", -1)
|
|
|
|
def get_header_tokens(self, match):
|
|
field = match.group(1)
|
|
|
|
if field.lower() in self.attention_headers:
|
|
yield match.start(1), Name.Tag, field + ":"
|
|
yield match.start(2), Text.Whitespace, match.group(2)
|
|
|
|
pos = match.end(2)
|
|
body = match.group(3)
|
|
for i, t, v in self.get_tokens_unprocessed(body, ("root", field.lower())):
|
|
yield pos + i, t, v
|
|
|
|
else:
|
|
yield match.start(), Comment, match.group()
|
|
|
|
def get_body_tokens(self, match):
|
|
pos_body_start = match.start()
|
|
entire_body = match.group()
|
|
|
|
# skip first newline
|
|
if entire_body[0] == '\n':
|
|
yield pos_body_start, Text.Whitespace, '\n'
|
|
pos_body_start = pos_body_start + 1
|
|
entire_body = entire_body[1:]
|
|
|
|
# if it is not a mulitpart
|
|
if not self.content_type.startswith("multipart") or not self.boundary:
|
|
for i, t, v in self.get_bodypart_tokens(entire_body):
|
|
yield pos_body_start + i, t, v
|
|
return
|
|
|
|
# find boundary
|
|
bdry_pattern = r"^--%s(--)?\n" % re.escape(self.boundary)
|
|
bdry_matcher = re.compile(bdry_pattern, re.MULTILINE)
|
|
|
|
# some data has prefix text before first boundary
|
|
m = bdry_matcher.search(entire_body)
|
|
if m:
|
|
pos_part_start = pos_body_start + m.end()
|
|
pos_iter_start = lpos_end = m.end()
|
|
yield pos_body_start, Text, entire_body[:m.start()]
|
|
yield pos_body_start + lpos_end, String.Delimiter, m.group()
|
|
else:
|
|
pos_part_start = pos_body_start
|
|
pos_iter_start = 0
|
|
|
|
# process tokens of each body part
|
|
for m in bdry_matcher.finditer(entire_body, pos_iter_start):
|
|
# bodypart
|
|
lpos_start = pos_part_start - pos_body_start
|
|
lpos_end = m.start()
|
|
part = entire_body[lpos_start:lpos_end]
|
|
for i, t, v in self.get_bodypart_tokens(part):
|
|
yield pos_part_start + i, t, v
|
|
|
|
# boundary
|
|
yield pos_body_start + lpos_end, String.Delimiter, m.group()
|
|
pos_part_start = pos_body_start + m.end()
|
|
|
|
# some data has suffix text after last boundary
|
|
lpos_start = pos_part_start - pos_body_start
|
|
if lpos_start != len(entire_body):
|
|
yield pos_part_start, Text, entire_body[lpos_start:]
|
|
|
|
def get_bodypart_tokens(self, text):
|
|
# return if:
|
|
# * no content
|
|
# * no content type specific
|
|
# * content encoding is not readable
|
|
# * max recurrsion exceed
|
|
if not text.strip() or not self.content_type:
|
|
return [(0, Other, text)]
|
|
|
|
cte = self.content_transfer_encoding
|
|
if cte and cte not in {"8bit", "7bit", "quoted-printable"}:
|
|
return [(0, Other, text)]
|
|
|
|
if self.max_nested_level == 0:
|
|
return [(0, Other, text)]
|
|
|
|
# get lexer
|
|
try:
|
|
lexer = get_lexer_for_mimetype(self.content_type)
|
|
except ClassNotFound:
|
|
return [(0, Other, text)]
|
|
|
|
if isinstance(lexer, type(self)):
|
|
lexer.max_nested_level = self.max_nested_level - 1
|
|
|
|
return lexer.get_tokens_unprocessed(text)
|
|
|
|
def store_content_type(self, match):
|
|
self.content_type = match.group(1)
|
|
|
|
prefix_len = match.start(1) - match.start(0)
|
|
yield match.start(0), Text.Whitespace, match.group(0)[:prefix_len]
|
|
yield match.start(1), Name.Label, match.group(2)
|
|
yield match.end(2), String.Delimiter, '/'
|
|
yield match.start(3), Name.Label, match.group(3)
|
|
|
|
def get_content_type_subtokens(self, match):
|
|
yield match.start(1), Text, match.group(1)
|
|
yield match.start(2), Text.Whitespace, match.group(2)
|
|
yield match.start(3), Name.Attribute, match.group(3)
|
|
yield match.start(4), Operator, match.group(4)
|
|
yield match.start(5), String, match.group(5)
|
|
|
|
if match.group(3).lower() == "boundary":
|
|
boundary = match.group(5).strip()
|
|
if boundary[0] == '"' and boundary[-1] == '"':
|
|
boundary = boundary[1:-1]
|
|
self.boundary = boundary
|
|
|
|
def store_content_transfer_encoding(self, match):
|
|
self.content_transfer_encoding = match.group(0).lower()
|
|
yield match.start(0), Name.Constant, match.group(0)
|
|
|
|
attention_headers = {"content-type", "content-transfer-encoding"}
|
|
|
|
tokens = {
|
|
"root": [
|
|
(r"^([\w-]+):( *)([\s\S]*?\n)(?![ \t])", get_header_tokens),
|
|
(r"^$[\s\S]+", get_body_tokens),
|
|
],
|
|
"header": [
|
|
# folding
|
|
(r"\n[ \t]", Text.Whitespace),
|
|
(r"\n(?![ \t])", Text.Whitespace, "#pop"),
|
|
],
|
|
"content-type": [
|
|
include("header"),
|
|
(
|
|
r"^\s*((multipart|application|audio|font|image|model|text|video"
|
|
r"|message)/([\w-]+))",
|
|
store_content_type,
|
|
),
|
|
(r'(;)((?:[ \t]|\n[ \t])*)([\w:-]+)(=)([\s\S]*?)(?=;|\n(?![ \t]))',
|
|
get_content_type_subtokens),
|
|
(r';[ \t]*\n(?![ \t])', Text, '#pop'),
|
|
],
|
|
"content-transfer-encoding": [
|
|
include("header"),
|
|
(r"([\w-]+)", store_content_transfer_encoding),
|
|
],
|
|
}
|