diff --git a/mautrix_telegram/formatter/from_matrix/__init__.py b/mautrix_telegram/formatter/from_matrix/__init__.py
index 206165cf..e9861b44 100644
--- a/mautrix_telegram/formatter/from_matrix/__init__.py
+++ b/mautrix_telegram/formatter/from_matrix/__init__.py
@@ -26,12 +26,7 @@ from ...types import TelegramID, MatrixRoomID
 from ...db import Message as DBMessage
 from ..util import (add_surrogates, remove_surrogates, trim_reply_fallback_html,
                     trim_reply_fallback_text)
-from .parser_common import ParsedMessage
-
-try:
-    from mautrix_telegram.formatter.from_matrix.parser_lxml import parse_html
-except ImportError:
-    from mautrix_telegram.formatter.from_matrix.parser_htmlparser import parse_html
+from .parser import ParsedMessage, parse_html
 
 if TYPE_CHECKING:
     from ...context import Context
diff --git a/mautrix_telegram/formatter/from_matrix/html_reader.py b/mautrix_telegram/formatter/from_matrix/html_reader.py
new file mode 100644
index 00000000..f1e4b178
--- /dev/null
+++ b/mautrix_telegram/formatter/from_matrix/html_reader.py
@@ -0,0 +1,4 @@
+try:
+    from .html_reader_lxml import HTMLNode, read_html
+except ImportError:
+    from .html_reader_htmlparser import HTMLNode, read_html
diff --git a/mautrix_telegram/formatter/from_matrix/html_reader.pyi b/mautrix_telegram/formatter/from_matrix/html_reader.pyi
new file mode 100644
index 00000000..d292ff3c
--- /dev/null
+++ b/mautrix_telegram/formatter/from_matrix/html_reader.pyi
@@ -0,0 +1,11 @@
+from typing import Dict, List
+
+
+class HTMLNode(List['HTMLNode']):
+    tag: str
+    text: str
+    tail: str
+    attrib: Dict[str, str]
+
+
+def read_html(data: str) -> HTMLNode: ...
diff --git a/mautrix_telegram/formatter/from_matrix/html_reader_htmlparser.py b/mautrix_telegram/formatter/from_matrix/html_reader_htmlparser.py
new file mode 100644
index 00000000..9ac10cf2
--- /dev/null
+++ b/mautrix_telegram/formatter/from_matrix/html_reader_htmlparser.py
@@ -0,0 +1,58 @@
+# -*- coding: future_fstrings -*-
+# mautrix-telegram - A Matrix-Telegram puppeting bridge
+# Copyright (C) 2018 Tulir Asokan
+#
+# This program is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU Affero General Public License for more details.
+#
+# You should have received a copy of the GNU Affero General Public License
+# along with this program.  If not, see <https://www.gnu.org/licenses/>.
+from typing import Dict, List, Tuple
+
+from html.parser import HTMLParser
+
+
+class HTMLNode(list):
+    def __init__(self, tag: str, attrs: List[Tuple[str, str]]):
+        super().__init__()
+        self.tag = tag  # type: str
+        self.text = ""  # type: str
+        self.tail = ""  # type: str
+        self.attrib = dict(attrs)  # type: Dict[str, str]
+
+
+class NodeifyingParser(HTMLParser):
+    def __init__(self):
+        super().__init__()
+        self.stack = [HTMLNode("html", [])]  # type: List[HTMLNode]
+
+    def handle_starttag(self, tag, attrs):
+        node = HTMLNode(tag, attrs)
+        self.stack[-1].append(node)
+        self.stack.append(node)
+
+    def handle_endtag(self, tag):
+        if tag == self.stack[-1].tag:
+            self.stack.pop()
+
+    def handle_data(self, data):
+        if len(self.stack[-1]) > 0:
+            self.stack[-1][-1].tail += data
+        else:
+            self.stack[-1].text += data
+
+    def error(self, message):
+        pass
+
+
+def read_html(data: str) -> HTMLNode:
+    parser = NodeifyingParser()
+    parser.feed(data)
+    return parser.stack[0]
diff --git a/mautrix_telegram/formatter/from_matrix/parser_common.py b/mautrix_telegram/formatter/from_matrix/html_reader_lxml.py
similarity index 51%
rename from mautrix_telegram/formatter/from_matrix/parser_common.py
rename to mautrix_telegram/formatter/from_matrix/html_reader_lxml.py
index 9b04b026..36d6d56e 100644
--- a/mautrix_telegram/formatter/from_matrix/parser_common.py
+++ b/mautrix_telegram/formatter/from_matrix/html_reader_lxml.py
@@ -14,23 +14,10 @@
 #
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <https://www.gnu.org/licenses/>.
-import re
-from typing import List, Tuple, Pattern
-from telethon.tl.types import TypeMessageEntity
+from lxml import html
+
+HTMLNode = html.HtmlElement
 
 
-class MatrixParserCommon:
-    mention_regex = re.compile("https://matrix.to/#/(@.+:.+)")  # type: Pattern
-    room_regex = re.compile("https://matrix.to/#/(#.+:.+)")  # type: Pattern
-    block_tags = ("p", "pre", "blockquote",
-                  "ol", "ul", "li",
-                  "h1", "h2", "h3", "h4", "h5", "h6",
-                  "div", "hr", "table")  # type: Tuple[str, ...]
-    list_bullets = ("●", "○", "■", "‣")  # type: Tuple[str, ...]
-
-    @classmethod
-    def list_bullet(cls, depth: int) -> str:
-        return cls.list_bullets[(depth - 1) % len(cls.list_bullets)] + " "
-
-
-ParsedMessage = Tuple[str, List[TypeMessageEntity]]
+def read_html(data: str) -> HTMLNode:
+    return html.fromstring(data)
diff --git a/mautrix_telegram/formatter/from_matrix/parser_lxml.py b/mautrix_telegram/formatter/from_matrix/parser.py
similarity index 83%
rename from mautrix_telegram/formatter/from_matrix/parser_lxml.py
rename to mautrix_telegram/formatter/from_matrix/parser.py
index 70057245..e53c816b 100644
--- a/mautrix_telegram/formatter/from_matrix/parser_lxml.py
+++ b/mautrix_telegram/formatter/from_matrix/parser.py
@@ -14,21 +14,26 @@
 #
 # You should have received a copy of the GNU Affero General Public License
 # along with this program.  If not, see <https://www.gnu.org/licenses/>.
-from typing import List, Tuple
-from lxml import html
+from typing import List, Tuple, Pattern
+import re
 
 from telethon.tl.types import (MessageEntityMention as Mention, MessageEntityBotCommand as Command,
                                MessageEntityMentionName as MentionName, MessageEntityEmail as Email,
                                MessageEntityUrl as URL, MessageEntityTextUrl as TextURL,
                                MessageEntityBold as Bold, MessageEntityItalic as Italic,
-                               MessageEntityCode as Code, MessageEntityPre as Pre)
+                               MessageEntityCode as Code, MessageEntityPre as Pre,
+                               TypeMessageEntity)
 
 from ... import user as u, puppet as pu, portal as po
 from ...types import MatrixUserID
 from ..util import html_to_unicode
-from .parser_common import MatrixParserCommon, ParsedMessage
 from .telegram_message import TelegramMessage, Entity, offset_length_multiply
 
+from .html_reader import HTMLNode, read_html
+
+
+ParsedMessage = Tuple[str, List[TypeMessageEntity]]
+
 
 def parse_html(input_html: str) -> ParsedMessage:
     return MatrixParser.parse(input_html)
@@ -52,9 +57,21 @@ class RecursionContext:
         return RecursionContext(strip_linebreaks=False, ul_depth=self.ul_depth)
 
 
-class MatrixParser(MatrixParserCommon):
+class MatrixParser:
+    mention_regex = re.compile("https://matrix.to/#/(@.+:.+)")  # type: Pattern
+    room_regex = re.compile("https://matrix.to/#/(#.+:.+)")  # type: Pattern
+    block_tags = ("p", "pre", "blockquote",
+                  "ol", "ul", "li",
+                  "h1", "h2", "h3", "h4", "h5", "h6",
+                  "div", "hr", "table")  # type: Tuple[str, ...]
+    list_bullets = ("●", "○", "■", "‣")  # type: Tuple[str, ...]
+
     @classmethod
-    def list_to_tmessage(cls, node: html.HtmlElement, ctx: RecursionContext) -> TelegramMessage:
+    def list_bullet(cls, depth: int) -> str:
+        return cls.list_bullets[(depth - 1) % len(cls.list_bullets)] + " "
+
+    @classmethod
+    def list_to_tmessage(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         ordered = node.tag == "ol"
         tagged_children = cls.node_to_tagged_tmessages(node, ctx)
         counter = 1
@@ -86,23 +103,21 @@ class MatrixParser(MatrixParserCommon):
         return TelegramMessage.join(children, "\n")
 
     @classmethod
-    def blockquote_to_tmessage(cls, node: html.HtmlElement, ctx: RecursionContext
-                               ) -> TelegramMessage:
+    def blockquote_to_tmessage(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         msg = cls.tag_aware_parse_node(node, ctx)
         children = msg.trim().split("\n")
         children = [child.prepend("> ") for child in children]
         return TelegramMessage.join(children, "\n")
 
     @classmethod
-    def header_to_tmessage(cls, node: html.HtmlElement, ctx: RecursionContext) -> TelegramMessage:
+    def header_to_tmessage(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         children = cls.node_to_tmessages(node, ctx)
         length = int(node.tag[1])
         prefix = "#" * length + " "
         return TelegramMessage.join(children, "").prepend(prefix).format(Bold)
 
     @classmethod
-    def basic_format_to_tmessage(cls, node: html.HtmlElement, ctx: RecursionContext
-                                 ) -> TelegramMessage:
+    def basic_format_to_tmessage(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         msg = cls.tag_aware_parse_node(node, ctx)
         if node.tag in ("b", "strong"):
             msg.format(Bold)
@@ -121,7 +136,7 @@ class MatrixParser(MatrixParserCommon):
         return msg
 
     @classmethod
-    def link_to_tstring(cls, node: html.HtmlElement, ctx: RecursionContext) -> TelegramMessage:
+    def link_to_tstring(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         msg = cls.tag_aware_parse_node(node, ctx)
         href = node.attrib.get("href", "")
         if not href:
@@ -156,7 +171,7 @@ class MatrixParser(MatrixParserCommon):
                 else msg.format(TextURL, url=href))
 
     @classmethod
-    def node_to_tmessage(cls, node: html.HtmlElement, ctx: RecursionContext) -> TelegramMessage:
+    def node_to_tmessage(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         if node.tag == "blockquote":
             return cls.blockquote_to_tmessage(node, ctx)
         elif node.tag == "ol":
@@ -193,7 +208,7 @@ class MatrixParser(MatrixParserCommon):
         return TelegramMessage(text)
 
     @classmethod
-    def node_to_tagged_tmessages(cls, node: html.HtmlElement, ctx: RecursionContext
+    def node_to_tagged_tmessages(cls, node: HTMLNode, ctx: RecursionContext
                                  ) -> List[Tuple[TelegramMessage, str]]:
         output = []
 
@@ -206,12 +221,12 @@ class MatrixParser(MatrixParserCommon):
         return output
 
     @classmethod
-    def node_to_tmessages(cls, node: html.HtmlElement, ctx: RecursionContext
+    def node_to_tmessages(cls, node: HTMLNode, ctx: RecursionContext
                           ) -> List[TelegramMessage]:
         return [msg for (msg, tag) in cls.node_to_tagged_tmessages(node, ctx)]
 
     @classmethod
-    def tag_aware_parse_node(cls, node: html.HtmlElement, ctx: RecursionContext
+    def tag_aware_parse_node(cls, node: HTMLNode, ctx: RecursionContext
                              ) -> TelegramMessage:
         msgs = cls.node_to_tagged_tmessages(node, ctx)
         output = TelegramMessage()
@@ -226,11 +241,10 @@ class MatrixParser(MatrixParserCommon):
         return output.trim()
 
     @classmethod
-    def parse_node(cls, node: html.HtmlElement, ctx: RecursionContext) -> TelegramMessage:
+    def parse_node(cls, node: HTMLNode, ctx: RecursionContext) -> TelegramMessage:
         return TelegramMessage.join(cls.node_to_tmessages(node, ctx))
 
     @classmethod
     def parse(cls, data: str) -> ParsedMessage:
-        document = html.fromstring(f"<html>{data}</html>")
-        msg = cls.parse_node(document, RecursionContext())
+        msg = cls.node_to_tmessage(read_html(f"<body>{data}</body>"), RecursionContext())
         return msg.text, msg.entities
diff --git a/mautrix_telegram/formatter/from_matrix/parser_htmlparser.py b/mautrix_telegram/formatter/from_matrix/parser_htmlparser.py
deleted file mode 100644
index 8988228b..00000000
--- a/mautrix_telegram/formatter/from_matrix/parser_htmlparser.py
+++ /dev/null
@@ -1,241 +0,0 @@
-# -*- coding: future_fstrings -*-
-# mautrix-telegram - A Matrix-Telegram puppeting bridge
-# Copyright (C) 2018 Tulir Asokan
-#
-# This program is free software: you can redistribute it and/or modify
-# it under the terms of the GNU Affero General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU Affero General Public License for more details.
-#
-# You should have received a copy of the GNU Affero General Public License
-# along with this program.  If not, see <https://www.gnu.org/licenses/>.
-from typing import (Optional, List, Tuple, Type, Dict, Any, TYPE_CHECKING, Match)
-from html import unescape
-from html.parser import HTMLParser
-from collections import deque
-import math
-
-from telethon.tl.types import (MessageEntityMention, MessageEntityMentionName, MessageEntityEmail,
-                               MessageEntityUrl, MessageEntityTextUrl, MessageEntityBold,
-                               MessageEntityItalic, MessageEntityCode, MessageEntityPre,
-                               MessageEntityBotCommand, TypeMessageEntity)
-
-from ... import user as u, puppet as pu, portal as po
-from ...types import MatrixUserID
-from ..util import html_to_unicode
-from .parser_common import MatrixParserCommon, ParsedMessage
-
-if TYPE_CHECKING:
-    from typing import Deque
-
-
-def parse_html(html: str) -> ParsedMessage:
-    parser = MatrixParser()
-    parser.feed(html)
-    return parser.text, parser.entities
-
-
-class MatrixParser(HTMLParser, MatrixParserCommon):
-    def __init__(self):
-        super(MatrixParser, self).__init__()
-        self.text = ""  # type: str
-        self.entities = []  # type: List[TypeMessageEntity]
-        self._building_entities = {}  # type: Dict[str, TypeMessageEntity]
-        self._list_counter = 0  # type: int
-        self._open_tags = deque()  # type: Deque[str]
-        self._open_tags_meta = deque()  # type: Deque[Any]
-        self._line_is_new = True  # type: bool
-        self._list_entry_is_new = False  # type: bool
-
-    def _parse_url(self, url: str, args: Dict[str, Any]
-                   ) -> Tuple[Optional[Type[TypeMessageEntity]], Optional[str]]:
-        mention = self.mention_regex.match(url)  # type: Match
-        if mention:
-            mxid = MatrixUserID(mention.group(1))
-            user = (pu.Puppet.get_by_mxid(mxid)
-                    or u.User.get_by_mxid(mxid, create=False))
-            if not user:
-                return None, None
-            if user.username:
-                return MessageEntityMention, f"@{user.username}"
-            elif user.tgid:
-                args["user_id"] = user.tgid
-                return MessageEntityMentionName, user.displayname or None
-            else:
-                return None, None
-
-        room = self.room_regex.match(url)  # type: Match
-        if room:
-            username = po.Portal.get_username_from_mx_alias(room.group(1))
-            portal = po.Portal.find_by_username(username)
-            if portal and portal.username:
-                return MessageEntityMention, f"@{portal.username}"
-
-        if url.startswith("mailto:"):
-            return MessageEntityEmail, url[len("mailto:"):]
-        elif self.get_starttag_text() == url:
-            return MessageEntityUrl, url
-        else:
-            args["url"] = url
-            return MessageEntityTextUrl, None
-
-    def handle_starttag(self, tag: str, attrs_list: List[Tuple[str, str]]):
-        self._open_tags.appendleft(tag)
-        self._open_tags_meta.appendleft(0)
-
-        attrs = dict(attrs_list)
-        entity_type = None  # type: Optional[Type[TypeMessageEntity]]
-        args = {}  # type: Dict[str, Any]
-        if tag in ("strong", "b"):
-            entity_type = MessageEntityBold
-        elif tag in ("em", "i"):
-            entity_type = MessageEntityItalic
-        elif tag == "code":
-            try:
-                pre = self._building_entities["pre"]
-                try:
-                    # Pre tag and language found, add language to MessageEntityPre
-                    pre.language = attrs["class"][len("language-"):]
-                except KeyError:
-                    # Pre tag found, but language not found, keep pre as-is
-                    pass
-            except KeyError:
-                # No pre tag found, this is inline code
-                entity_type = MessageEntityCode
-        elif tag == "pre":
-            entity_type = MessageEntityPre
-            args["language"] = ""
-        elif tag == "command":
-            entity_type = MessageEntityBotCommand
-        elif tag == "li":
-            self._list_entry_is_new = True
-        elif tag == "a":
-            try:
-                url = attrs["href"]
-            except KeyError:
-                return
-            entity_type, url = self._parse_url(url, args)
-            self._open_tags_meta.popleft()
-            self._open_tags_meta.appendleft(url)
-
-        if (tag in self.block_tags and ("blockquote" not in self._open_tags)) or tag == "br":
-            self._newline()
-
-        if entity_type and tag not in self._building_entities:
-            offset = len(self.text)
-            self._building_entities[tag] = entity_type(offset=offset, length=0, **args)
-
-    @property
-    def _list_indent(self) -> int:
-        indent = 0
-        first_skipped = False
-        for index, tag in enumerate(self._open_tags):
-            if not first_skipped and tag in ("ol", "ul"):
-                # The first list level isn't indented, so skip it.
-                first_skipped = True
-                continue
-            if tag == "ol":
-                n = self._open_tags_meta[index]
-                extra_length_for_long_index = (int(math.log(n, 10)) - 1) * 3
-                indent += 4 + extra_length_for_long_index
-            elif tag == "ul":
-                indent += 3
-        return indent
-
-    def _newline(self, allow_multi: bool = False):
-        if self._line_is_new and not allow_multi:
-            return
-        self.text += "\n"
-        self._line_is_new = True
-        for entity in self._building_entities.values():
-            entity.length += 1
-
-    def _handle_special_previous_tags(self, text: str) -> str:
-        if "pre" not in self._open_tags and "code" not in self._open_tags:
-            text = text.replace("\n", "")
-        else:
-            text = text.strip()
-
-        previous_tag = self._open_tags[0] if len(self._open_tags) > 0 else ""
-        if previous_tag == "a":
-            url = self._open_tags_meta[0]
-            if url:
-                text = url
-        elif previous_tag == "command":
-            text = f"/{text}"
-        return text
-
-    def _html_to_unicode(self, text: str) -> str:
-        strikethrough, underline = "del" in self._open_tags, "u" in self._open_tags
-        if strikethrough and underline:
-            text = html_to_unicode(text, "\u0336\u0332")
-        elif strikethrough:
-            text = html_to_unicode(text, "\u0336")
-        elif underline:
-            text = html_to_unicode(text, "\u0332")
-        return text
-
-    def _handle_tags_for_data(self, text: str) -> Tuple[str, int]:
-        extra_offset = 0
-        list_entry_handled_once = False
-        # In order to maintain order of things like blockquotes in lists or lists in blockquotes,
-        # we can't just have ifs/elses and we need to actually loop through the open tags in order.
-        for index, tag in enumerate(self._open_tags):
-            if tag == "blockquote" and self._line_is_new:
-                text = f"> {text}"
-                extra_offset += 2
-            elif tag == "li" and not list_entry_handled_once:
-                list_type_index = index + 1
-                list_type = self._open_tags[list_type_index]
-                indent = self._list_indent * " " if self._line_is_new else ""
-                if list_type == "ol":
-                    n = self._open_tags_meta[list_type_index]
-                    if self._list_entry_is_new:
-                        n += 1
-                        self._open_tags_meta[list_type_index] = n
-                        prefix = f"{n}. "
-                    else:
-                        prefix = int(math.log(n, 10)) * 3 * " " + 4 * " "
-                else:
-                    prefix = (self.list_bullet(self._open_tags.count('ul'))
-                              if self._list_entry_is_new else 3 * " ")
-                if not self._list_entry_is_new and not self._line_is_new:
-                    prefix = ""
-                extra_offset += len(indent) + len(prefix)
-                text = indent + prefix + text
-                self._list_entry_is_new = False
-                list_entry_handled_once = True
-        return text, extra_offset
-
-    def _extend_entities_in_construction(self, text: str, extra_offset: int):
-        for tag, entity in self._building_entities.items():
-            entity.length += len(text) - extra_offset
-            entity.offset += extra_offset
-
-    def handle_data(self, text: str):
-        text = unescape(text)
-        text = self._handle_special_previous_tags(text)
-        text = self._html_to_unicode(text)
-        text, extra_offset = self._handle_tags_for_data(text)
-        self._extend_entities_in_construction(text, extra_offset)
-        self._line_is_new = False
-        self.text += text
-
-    def handle_endtag(self, tag: str):
-        try:
-            self._open_tags.popleft()
-            self._open_tags_meta.popleft()
-        except IndexError:
-            pass
-
-        entity = self._building_entities.pop(tag, None)
-        if entity:
-            self.entities.append(entity)
-
-        if tag in self.block_tags and tag != "br" and "blockquote" not in self._open_tags:
-            self._newline(allow_multi=tag == "br")
diff --git a/mautrix_telegram/formatter/from_matrix/telegram_message.py b/mautrix_telegram/formatter/from_matrix/telegram_message.py
index c849cc00..f78af066 100644
--- a/mautrix_telegram/formatter/from_matrix/telegram_message.py
+++ b/mautrix_telegram/formatter/from_matrix/telegram_message.py
@@ -153,5 +153,6 @@ class TelegramMessage:
                 msg = TelegramMessage(text=msg)
             main.entities += Entity.adjust(msg.entities, offset_diff(len(main.text)))
             main.text += msg.text + separator
-        main.text = main.text[:-len(separator)]
+        if len(separator) > 0:
+            main.text = main.text[:-len(separator)]
         return main