Add HTML style parse mode
This commit is contained in:
parent
70be5558c9
commit
0e3d08ae75
@ -47,7 +47,7 @@ from pyrogram.api.types import (
|
|||||||
InputPeerUser, InputPeerChat, InputPeerChannel
|
InputPeerUser, InputPeerChat, InputPeerChannel
|
||||||
)
|
)
|
||||||
from pyrogram.crypto import CTR
|
from pyrogram.crypto import CTR
|
||||||
from pyrogram.extensions import Markdown
|
from pyrogram.extensions import Markdown, HTML
|
||||||
from pyrogram.session import Auth, Session
|
from pyrogram.session import Auth, Session
|
||||||
|
|
||||||
log = logging.getLogger(__name__)
|
log = logging.getLogger(__name__)
|
||||||
@ -89,6 +89,7 @@ class Client:
|
|||||||
self.peers_by_username = {}
|
self.peers_by_username = {}
|
||||||
|
|
||||||
self.markdown = Markdown(self.peers_by_id)
|
self.markdown = Markdown(self.peers_by_id)
|
||||||
|
self.html = HTML(self.peers_by_id)
|
||||||
|
|
||||||
self.config = None
|
self.config = None
|
||||||
self.proxy = None
|
self.proxy = None
|
||||||
|
@ -16,4 +16,5 @@
|
|||||||
# You should have received a copy of the GNU Lesser General Public License
|
# You should have received a copy of the GNU Lesser General Public License
|
||||||
# along with Pyrogram. If not, see <http://www.gnu.org/licenses/>.
|
# along with Pyrogram. If not, see <http://www.gnu.org/licenses/>.
|
||||||
|
|
||||||
|
from .html import HTML
|
||||||
from .markdown import Markdown
|
from .markdown import Markdown
|
||||||
|
103
pyrogram/extensions/html.py
Normal file
103
pyrogram/extensions/html.py
Normal file
@ -0,0 +1,103 @@
|
|||||||
|
import re
|
||||||
|
from struct import unpack
|
||||||
|
|
||||||
|
from pyrogram.api.types import (
|
||||||
|
MessageEntityBold as Bold,
|
||||||
|
MessageEntityItalic as Italic,
|
||||||
|
MessageEntityCode as Code,
|
||||||
|
MessageEntityTextUrl as Url,
|
||||||
|
MessageEntityPre as Pre,
|
||||||
|
InputMessageEntityMentionName as Mention
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
|
class HTML:
|
||||||
|
SMP_RE = re.compile(r"[\U00010000-\U0010FFFF]")
|
||||||
|
|
||||||
|
BOLD_RE = r"(?P<b><b>(?P<b_body>.*?)</b>)"
|
||||||
|
STRONG_RE = r"(?P<strong><strong>(?P<strong_body>.*?)</strong>)"
|
||||||
|
ITALIC_RE = r"(?P<i><i>(?P<i_body>.*?)</i>)"
|
||||||
|
EMPATHIZE_RE = r"(?P<em><em>(?P<em_body>.*?)</em>)"
|
||||||
|
CODE_RE = r"(?P<code><code>(?P<code_body>.*?)</code>)"
|
||||||
|
PRE_RE = r"(?P<pre><pre>(?P<pre_body>.*?)</pre>)"
|
||||||
|
MENTION_RE = r"(?P<mention><a href=\"tg://user\?id=(?P<user_id>\d+?)\">(?P<mention_text>.*?)</a>)"
|
||||||
|
URL_RE = r"(?P<url><a href=\"(?P<url_path>.*?)\">(?P<url_text>.*?)</a>)"
|
||||||
|
|
||||||
|
HTML_RE = re.compile("|".join([BOLD_RE, STRONG_RE, ITALIC_RE, EMPATHIZE_RE, CODE_RE, PRE_RE, MENTION_RE, URL_RE]))
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def add_surrogates(cls, text):
|
||||||
|
return cls.SMP_RE.sub(
|
||||||
|
lambda match: # Split SMP in two surrogates
|
||||||
|
"".join(chr(i) for i in unpack("<HH", match.group().encode("utf-16le"))),
|
||||||
|
text
|
||||||
|
)
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def remove_surrogates(text):
|
||||||
|
return text.encode("utf-16", "surrogatepass").decode("utf-16")
|
||||||
|
|
||||||
|
def __init__(self, peers_by_id):
|
||||||
|
self.peers_by_id = peers_by_id
|
||||||
|
|
||||||
|
def parse(self, text):
|
||||||
|
entities = []
|
||||||
|
text = self.add_surrogates(text)
|
||||||
|
offset = 0
|
||||||
|
|
||||||
|
# TODO: Beautify ifs
|
||||||
|
for match in self.HTML_RE.finditer(text):
|
||||||
|
start = match.start() - offset
|
||||||
|
|
||||||
|
if match.group("b"):
|
||||||
|
pattern = match.group("b")
|
||||||
|
body = match.group("b_body")
|
||||||
|
entity = Bold(start, len(body))
|
||||||
|
offset += 7
|
||||||
|
elif match.group("strong"):
|
||||||
|
pattern = match.group("strong")
|
||||||
|
body = match.group("strong_body")
|
||||||
|
entity = Bold(start, len(body))
|
||||||
|
offset += 17
|
||||||
|
elif match.group("i"):
|
||||||
|
pattern = match.group("i")
|
||||||
|
body = match.group("i_body")
|
||||||
|
entity = Italic(start, len(body))
|
||||||
|
offset += 7
|
||||||
|
elif match.group("em"):
|
||||||
|
pattern = match.group("em")
|
||||||
|
body = match.group("em_body")
|
||||||
|
entity = Italic(start, len(body))
|
||||||
|
offset += 9
|
||||||
|
elif match.group("code"):
|
||||||
|
pattern = match.group("code")
|
||||||
|
body = match.group("code_body")
|
||||||
|
entity = Code(start, len(body))
|
||||||
|
offset += 13
|
||||||
|
elif match.group("pre"):
|
||||||
|
pattern = match.group("pre")
|
||||||
|
body = match.group("pre_body")
|
||||||
|
entity = Pre(start, len(body), "")
|
||||||
|
offset += 11
|
||||||
|
elif match.group("mention"):
|
||||||
|
pattern = match.group("mention")
|
||||||
|
body = match.group("mention_text")
|
||||||
|
user_id = match.group("user_id")
|
||||||
|
entity = Mention(start, len(body), self.peers_by_id[int(user_id)])
|
||||||
|
offset += len(user_id) + 28
|
||||||
|
elif match.group("url"):
|
||||||
|
pattern = match.group("url")
|
||||||
|
body = match.group("url_text")
|
||||||
|
path = match.group("url_path")
|
||||||
|
entity = Url(start, len(body), path)
|
||||||
|
offset += len(path) + 15
|
||||||
|
else:
|
||||||
|
continue
|
||||||
|
|
||||||
|
entities.append(entity)
|
||||||
|
text = text.replace(pattern, body)
|
||||||
|
|
||||||
|
return dict(
|
||||||
|
message=self.remove_surrogates(text),
|
||||||
|
entities=entities
|
||||||
|
)
|
Loading…
Reference in New Issue
Block a user