# Zulip's main Markdown implementation. See docs/subsystems/markdown.md for # detailed documentation on our Markdown syntax. import datetime import functools import html import logging import os import re import time import urllib import urllib.parse from collections import defaultdict, deque from dataclasses import dataclass from io import StringIO from typing import ( Any, Callable, Dict, Generic, Iterable, List, Optional, Set, Tuple, TypeVar, Union, ) from typing.re import Match, Pattern from xml.etree import ElementTree as etree from xml.etree.ElementTree import Element, SubElement import ahocorasick import dateutil.parser import dateutil.tz import markdown import requests from django.conf import settings from django.db.models import Q from hyperlink import parse from markdown.extensions import codehilite, nl2br, sane_lists, tables from typing_extensions import TypedDict from zerver.lib import mention as mention from zerver.lib.cache import NotFoundInCache, cache_with_key from zerver.lib.camo import get_camo_url from zerver.lib.emoji import ( codepoint_to_name, emoticon_regex, name_to_codepoint, translate_emoticons, ) from zerver.lib.exceptions import MarkdownRenderingException from zerver.lib.markdown import fenced_code from zerver.lib.markdown.fenced_code import FENCE_RE from zerver.lib.mention import extract_user_group, possible_mentions, possible_user_group_mentions from zerver.lib.tex import render_tex from zerver.lib.thumbnail import user_uploads_or_external from zerver.lib.timeout import TimeoutExpired, timeout from zerver.lib.timezone import get_common_timezones from zerver.lib.url_encoding import encode_stream, hash_util_encode from zerver.lib.url_preview import preview as link_preview from zerver.models import ( MAX_MESSAGE_LENGTH, Message, Realm, UserGroup, UserGroupMembership, UserProfile, all_realm_filters, get_active_streams, realm_filters_for_realm, ) ReturnT = TypeVar('ReturnT') def one_time(method: Callable[[], ReturnT]) -> Callable[[], ReturnT]: ''' Use this decorator with extreme caution. The function you wrap should have no dependency on any arguments (no args, no kwargs) nor should it depend on any global state. ''' val = None def cache_wrapper() -> ReturnT: nonlocal val if val is None: val = method() return val return cache_wrapper class FullNameInfo(TypedDict): id: int email: str full_name: str class LinkInfo(TypedDict): parent: Element title: Optional[str] index: Optional[int] remove: Optional[Element] DbData = Dict[str, Any] # Format version of the Markdown rendering; stored along with rendered # messages so that we can efficiently determine what needs to be re-rendered version = 1 _T = TypeVar('_T') ElementStringNone = Union[Element, Optional[str]] EMOJI_REGEX = r'(?P:[\w\-\+]+:)' def verbose_compile(pattern: str) -> Any: return re.compile( f"^(.*?){pattern}(.*?)$", re.DOTALL | re.UNICODE | re.VERBOSE, ) def normal_compile(pattern: str) -> Any: return re.compile( fr"^(.*?){pattern}(.*)$", re.DOTALL | re.UNICODE, ) STREAM_LINK_REGEX = r""" (?[^\*]+) # stream name can contain anything \*\* # ends by double asterisks """ @one_time def get_compiled_stream_link_regex() -> Pattern: return verbose_compile(STREAM_LINK_REGEX) STREAM_TOPIC_LINK_REGEX = r""" (?[^\*>]+) # stream name can contain anything except > > # > acts as separator (?P[^\*]+) # topic name can contain anything \*\* # ends by double asterisks """ @one_time def get_compiled_stream_topic_link_regex() -> Pattern: return verbose_compile(STREAM_TOPIC_LINK_REGEX) LINK_REGEX: Pattern = None def get_web_link_regex() -> str: # We create this one time, but not at startup. So the # first message rendered in any process will have some # extra costs. It's roughly 75ms to run this code, so # caching the value in LINK_REGEX is super important here. global LINK_REGEX if LINK_REGEX is not None: return LINK_REGEX tlds = '|'.join(list_of_tlds()) # A link starts at a word boundary, and ends at space, punctuation, or end-of-input. # # We detect a url either by the `https?://` or by building around the TLD. # In lieu of having a recursive regex (which python doesn't support) to match # arbitrary numbers of nested matching parenthesis, we manually build a regexp that # can match up to six # The inner_paren_contents chunk matches the innermore non-parenthesis-holding text, # and the paren_group matches text with, optionally, a matching set of parens inner_paren_contents = r"[^\s()\"]*" paren_group = r""" [^\s()\"]*? # Containing characters that won't end the URL (?: \( %s \) # and more characters in matched parens [^\s()\"]*? # followed by more characters )* # zero-or-more sets of paired parens """ nested_paren_chunk = paren_group for i in range(6): nested_paren_chunk = nested_paren_chunk % (paren_group,) nested_paren_chunk = nested_paren_chunk % (inner_paren_contents,) file_links = r"| (?:file://(/[^/ ]*)+/?)" if settings.ENABLE_FILE_LINKS else r"" REGEX = fr""" (? # Main group (?:(?: # Domain part https?://[\w.:@-]+? # If it has a protocol, anything goes. |(?: # Or, if not, be more strict to avoid false-positives (?:[\w-]+\.)+ # One or more domain components, separated by dots (?:{tlds}) # TLDs (filled in via format from tlds-alpha-by-domain.txt) ) ) (?:/ # A path, beginning with / {nested_paren_chunk} # zero-to-6 sets of paired parens )?) # Path is optional | (?:[\w.-]+\@[\w.-]+\.[\w]+) # Email is separate, since it can't have a path {file_links} # File path start with file:///, enable by setting ENABLE_FILE_LINKS=True | (?:bitcoin:[13][a-km-zA-HJ-NP-Z1-9]{{25,34}}) # Bitcoin address pattern, see https://mokagio.github.io/tech-journal/2014/11/21/regex-bitcoin.html ) (?= # URL must be followed by (not included in group) [!:;\?\),\.\'\"\>]* # Optional punctuation characters (?:\Z|\s) # followed by whitespace or end of string ) """ LINK_REGEX = verbose_compile(REGEX) return LINK_REGEX def clear_state_for_testing() -> None: # The link regex never changes in production, but our tests # try out both sides of ENABLE_FILE_LINKS, so we need # a way to clear it. global LINK_REGEX LINK_REGEX = None markdown_logger = logging.getLogger() def rewrite_local_links_to_relative(db_data: Optional[DbData], link: str) -> str: """If the link points to a local destination (e.g. #narrow/...), generate a relative link that will open it in the current window. """ if db_data: realm_uri_prefix = db_data['realm_uri'] + "/" if ( link.startswith(realm_uri_prefix) and urllib.parse.urljoin(realm_uri_prefix, link[len(realm_uri_prefix):]) == link ): return link[len(realm_uri_prefix):] return link def url_embed_preview_enabled(message: Optional[Message]=None, realm: Optional[Realm]=None, no_previews: bool=False) -> bool: if not settings.INLINE_URL_EMBED_PREVIEW: return False if no_previews: return False if realm is None: if message is not None: realm = message.get_realm() if realm is None: # realm can be None for odd use cases # like generating documentation or running # test code return True return realm.inline_url_embed_preview def image_preview_enabled(message: Optional[Message]=None, realm: Optional[Realm]=None, no_previews: bool=False) -> bool: if not settings.INLINE_IMAGE_PREVIEW: return False if no_previews: return False if realm is None: if message is not None: realm = message.get_realm() if realm is None: # realm can be None for odd use cases # like generating documentation or running # test code return True return realm.inline_image_preview def list_of_tlds() -> List[str]: # HACK we manually blacklist a few domains blacklist = ['PY\n', "MD\n"] # tlds-alpha-by-domain.txt comes from https://data.iana.org/TLD/tlds-alpha-by-domain.txt tlds_file = os.path.join(os.path.dirname(__file__), 'tlds-alpha-by-domain.txt') tlds = [tld.lower().strip() for tld in open(tlds_file) if tld not in blacklist and not tld[0].startswith('#')] tlds.sort(key=len, reverse=True) return tlds def walk_tree(root: Element, processor: Callable[[Element], Optional[_T]], stop_after_first: bool=False) -> List[_T]: results = [] queue = deque([root]) while queue: currElement = queue.popleft() for child in currElement: if child: queue.append(child) result = processor(child) if result is not None: results.append(result) if stop_after_first: return results return results @dataclass class ElementFamily: grandparent: Optional[Element] parent: Element child: Element in_blockquote: bool T = TypeVar("T") class ResultWithFamily(Generic[T]): family: ElementFamily result: T def __init__(self, family: ElementFamily, result: T): self.family = family self.result = result class ElementPair: parent: Optional["ElementPair"] value: Element def __init__(self, parent: Optional["ElementPair"], value: Element): self.parent = parent self.value = value def walk_tree_with_family(root: Element, processor: Callable[[Element], Optional[_T]], ) -> List[ResultWithFamily[_T]]: results = [] queue = deque([ElementPair(parent=None, value=root)]) while queue: currElementPair = queue.popleft() for child in currElementPair.value: if child: queue.append(ElementPair(parent=currElementPair, value=child)) result = processor(child) if result is not None: if currElementPair.parent is not None: grandparent_element = currElementPair.parent grandparent: Optional[Element] = grandparent_element.value else: grandparent = None family = ElementFamily( grandparent=grandparent, parent=currElementPair.value, child=child, in_blockquote=has_blockquote_ancestor(currElementPair), ) results.append(ResultWithFamily( family=family, result=result, )) return results def has_blockquote_ancestor(element_pair: Optional[ElementPair]) -> bool: if element_pair is None: return False elif element_pair.value.tag == 'blockquote': return True else: return has_blockquote_ancestor(element_pair.parent) @cache_with_key(lambda tweet_id: tweet_id, cache_name="database", with_statsd_key="tweet_data") def fetch_tweet_data(tweet_id: str) -> Optional[Dict[str, Any]]: if settings.TEST_SUITE: from . import testing_mocks res = testing_mocks.twitter(tweet_id) else: creds = { 'consumer_key': settings.TWITTER_CONSUMER_KEY, 'consumer_secret': settings.TWITTER_CONSUMER_SECRET, 'access_token_key': settings.TWITTER_ACCESS_TOKEN_KEY, 'access_token_secret': settings.TWITTER_ACCESS_TOKEN_SECRET, } if not all(creds.values()): return None # We lazily import twitter here because its import process is # surprisingly slow, and doing so has a significant impact on # the startup performance of `manage.py` commands. import twitter try: api = twitter.Api(tweet_mode='extended', **creds) # Sometimes Twitter hangs on responses. Timing out here # will cause the Tweet to go through as-is with no inline # preview, rather than having the message be rejected # entirely. This timeout needs to be less than our overall # formatting timeout. tweet = timeout(3, api.GetStatus, tweet_id) res = tweet.AsDict() except AttributeError: markdown_logger.error('Unable to load twitter api, you may have the wrong ' 'library installed, see https://github.com/zulip/zulip/issues/86') return None except TimeoutExpired: # We'd like to try again later and not cache the bad result, # so we need to re-raise the exception (just as though # we were being rate-limited) raise except twitter.TwitterError as e: t = e.args[0] if len(t) == 1 and ('code' in t[0]): # https://developer.twitter.com/en/docs/basics/response-codes code = t[0]['code'] if code in [34, 144, 421, 422]: # All these "correspond with HTTP 404," and mean # that the message doesn't exist; return None so # that we will cache the error. return None elif code in [63, 179]: # 63 is that the account is suspended, 179 is that # it is now locked; cache the None. return None elif code in [88, 130, 131]: # Code 88 means that we were rate-limited, 130 # means Twitter is having capacity issues, and 131 # is other 400-equivalent; in these cases, raise # the error so we don't cache None and will try # again later. raise # It's not clear what to do in cases of other errors, # but for now it seems reasonable to log at error # level (so that we get notified), but then cache the # failure to proceed with our usual work markdown_logger.exception("Unknown error fetching tweet data", stack_info=True) return None return res HEAD_START_RE = re.compile('^head[ >]') HEAD_END_RE = re.compile('^/head[ >]') META_START_RE = re.compile('^meta[ >]') META_END_RE = re.compile('^/meta[ >]') def fetch_open_graph_image(url: str) -> Optional[Dict[str, Any]]: in_head = False # HTML will auto close meta tags, when we start the next tag add # a closing tag if it has not been closed yet. last_closed = True head = [] # TODO: What if response content is huge? Should we get headers first? try: content = requests.get(url, timeout=1).text except Exception: return None # Extract the head and meta tags # All meta tags are self closing, have no children or are closed # automatically. for part in content.split('<'): if not in_head and HEAD_START_RE.match(part): # Started the head node output it to have a document root in_head = True head.append('') elif in_head and HEAD_END_RE.match(part): # Found the end of the head close any remaining tag then stop # processing in_head = False if not last_closed: last_closed = True head.append('') head.append('') break elif in_head and META_START_RE.match(part): # Found a meta node copy it if not last_closed: head.append('') last_closed = True head.append('<') head.append(part) if '/>' not in part: last_closed = False elif in_head and META_END_RE.match(part): # End of a meta node just copy it to close the tag head.append('<') head.append(part) last_closed = True try: doc = etree.fromstring(''.join(head)) except etree.ParseError: return None og_image = doc.find('meta[@property="og:image"]') og_title = doc.find('meta[@property="og:title"]') og_desc = doc.find('meta[@property="og:description"]') title = None desc = None if og_image is not None: image = og_image.get('content') else: return None if og_title is not None: title = og_title.get('content') if og_desc is not None: desc = og_desc.get('content') return {'image': image, 'title': title, 'desc': desc} def get_tweet_id(url: str) -> Optional[str]: parsed_url = urllib.parse.urlparse(url) if not (parsed_url.netloc == 'twitter.com' or parsed_url.netloc.endswith('.twitter.com')): return None to_match = parsed_url.path # In old-style twitter.com/#!/wdaher/status/1231241234-style URLs, # we need to look at the fragment instead if parsed_url.path == '/' and len(parsed_url.fragment) > 5: to_match = parsed_url.fragment tweet_id_match = re.match(r'^!?/.*?/status(es)?/(?P\d{10,30})(/photo/[0-9])?/?$', to_match) if not tweet_id_match: return None return tweet_id_match.group("tweetid") class InlineHttpsProcessor(markdown.treeprocessors.Treeprocessor): def run(self, root: Element) -> None: # Get all URLs from the blob found_imgs = walk_tree(root, lambda e: e if e.tag == "img" else None) for img in found_imgs: url = img.get("src") assert url is not None if urllib.parse.urlsplit(url).scheme != "http": # Don't rewrite images on our own site (e.g. emoji). continue img.set("src", get_camo_url(url)) class BacktickInlineProcessor(markdown.inlinepatterns.BacktickInlineProcessor): """ Return a `` element containing the matching text. """ def handleMatch(self, m: Match[str], data: str) -> Tuple[Union[None, Element], int, int]: # Let upstream's implementation do its job as it is, we'll # just replace the text to not strip the group because it # makes it impossible to put leading/trailing whitespace in # an inline code block. el, start, end = super().handleMatch(m, data) if m.group(3): # upstream's code here is: m.group(3).strip() rather than m.group(3). el.text = markdown.util.AtomicString(markdown.util.code_escape(m.group(3))) return el, start, end class InlineInterestingLinkProcessor(markdown.treeprocessors.Treeprocessor): TWITTER_MAX_IMAGE_HEIGHT = 400 TWITTER_MAX_TO_PREVIEW = 3 INLINE_PREVIEW_LIMIT_PER_MESSAGE = 5 def __init__(self, md: markdown.Markdown) -> None: markdown.treeprocessors.Treeprocessor.__init__(self, md) def add_a( self, root: Element, url: str, link: str, title: Optional[str]=None, desc: Optional[str]=None, class_attr: str="message_inline_image", data_id: Optional[str]=None, insertion_index: Optional[int]=None, already_thumbnailed: bool=False, ) -> None: desc = desc if desc is not None else "" # Update message.has_image attribute. if 'message_inline_image' in class_attr and self.md.zulip_message: self.md.zulip_message.has_image = True if insertion_index is not None: div = Element("div") root.insert(insertion_index, div) else: div = SubElement(root, "div") div.set("class", class_attr) a = SubElement(div, "a") a.set("href", link) if title is not None: a.set("title", title) if data_id is not None: a.set("data-id", data_id) img = SubElement(a, "img") if settings.THUMBNAIL_IMAGES and (not already_thumbnailed) and user_uploads_or_external(url): # See docs/thumbnailing.md for some high-level documentation. # # We strip leading '/' from relative URLs here to ensure # consistency in what gets passed to /thumbnail url = url.lstrip('/') img.set("src", "/thumbnail?url={}&size=thumbnail".format( urllib.parse.quote(url, safe=''), )) img.set('data-src-fullsize', "/thumbnail?url={}&size=full".format( urllib.parse.quote(url, safe=''), )) else: img.set("src", url) if class_attr == "message_inline_ref": summary_div = SubElement(div, "div") title_div = SubElement(summary_div, "div") title_div.set("class", "message_inline_image_title") title_div.text = title desc_div = SubElement(summary_div, "desc") desc_div.set("class", "message_inline_image_desc") def add_oembed_data(self, root: Element, link: str, extracted_data: Dict[str, Any]) -> bool: oembed_resource_type = extracted_data.get('type', '') title = extracted_data.get('title') if oembed_resource_type == 'photo': image = extracted_data.get('image') if image: self.add_a(root, image, link, title=title) return True elif oembed_resource_type == 'video': html = extracted_data['html'] image = extracted_data['image'] title = extracted_data.get('title') description = extracted_data.get('description') self.add_a(root, image, link, title, description, "embed-video message_inline_image", html, already_thumbnailed=True) return True return False def add_embed(self, root: Element, link: str, extracted_data: Dict[str, Any]) -> None: oembed = extracted_data.get('oembed', False) if oembed and self.add_oembed_data(root, link, extracted_data): return img_link = extracted_data.get('image') if not img_link: # Don't add an embed if an image is not found return container = SubElement(root, "div") container.set("class", "message_embed") parsed_img_link = urllib.parse.urlparse(img_link) # Append domain where relative img_link url is given if not parsed_img_link.netloc: parsed_url = urllib.parse.urlparse(link) domain = '{url.scheme}://{url.netloc}/'.format(url=parsed_url) img_link = urllib.parse.urljoin(domain, img_link) img = SubElement(container, "a") img.set("style", "background-image: url(" + img_link + ")") img.set("href", link) img.set("class", "message_embed_image") data_container = SubElement(container, "div") data_container.set("class", "data-container") title = extracted_data.get('title') if title: title_elm = SubElement(data_container, "div") title_elm.set("class", "message_embed_title") a = SubElement(title_elm, "a") a.set("href", link) a.set("title", title) a.text = title description = extracted_data.get('description') if description: description_elm = SubElement(data_container, "div") description_elm.set("class", "message_embed_description") description_elm.text = description def get_actual_image_url(self, url: str) -> str: # Add specific per-site cases to convert image-preview urls to image urls. # See https://github.com/zulip/zulip/issues/4658 for more information parsed_url = urllib.parse.urlparse(url) if (parsed_url.netloc == 'github.com' or parsed_url.netloc.endswith('.github.com')): # https://github.com/zulip/zulip/blob/master/static/images/logo/zulip-icon-128x128.png -> # https://raw.githubusercontent.com/zulip/zulip/master/static/images/logo/zulip-icon-128x128.png split_path = parsed_url.path.split('/') if len(split_path) > 3 and split_path[3] == "blob": return urllib.parse.urljoin('https://raw.githubusercontent.com', '/'.join(split_path[0:3] + split_path[4:])) return url def is_image(self, url: str) -> bool: if not self.md.image_preview_enabled: return False parsed_url = urllib.parse.urlparse(url) # remove html urls which end with img extensions that can not be shorted if parsed_url.netloc == 'pasteboard.co': return False # List from https://support.google.com/chromeos/bin/answer.py?hl=en&answer=183093 for ext in [".bmp", ".gif", ".jpe", "jpeg", ".jpg", ".png", ".webp"]: if parsed_url.path.lower().endswith(ext): return True return False def corrected_image_source(self, url: str) -> Optional[str]: # This function adjusts any urls from linx.li and # wikipedia.org to point to the actual image url. It's # structurally very similar to dropbox_image, and possibly # should be rewritten to use open graph, but has some value. parsed_url = urllib.parse.urlparse(url) if parsed_url.netloc.lower().endswith('.wikipedia.org'): # Redirecting from "/wiki/File:" to "/wiki/Special:FilePath/File:" # A possible alternative, that avoids the redirect after hitting "Special:" # is using the first characters of md5($filename) to generate the url domain = parsed_url.scheme + "://" + parsed_url.netloc correct_url = domain + parsed_url.path[:6] + 'Special:FilePath' + parsed_url.path[5:] return correct_url if parsed_url.netloc == 'linx.li': return 'https://linx.li/s' + parsed_url.path return None def dropbox_image(self, url: str) -> Optional[Dict[str, Any]]: # TODO: The returned Dict could possibly be a TypedDict in future. parsed_url = urllib.parse.urlparse(url) if (parsed_url.netloc == 'dropbox.com' or parsed_url.netloc.endswith('.dropbox.com')): is_album = parsed_url.path.startswith('/sc/') or parsed_url.path.startswith('/photos/') # Only allow preview Dropbox shared links if not (parsed_url.path.startswith('/s/') or parsed_url.path.startswith('/sh/') or is_album): return None # Try to retrieve open graph protocol info for a preview # This might be redundant right now for shared links for images. # However, we might want to make use of title and description # in the future. If the actual image is too big, we might also # want to use the open graph image. image_info = fetch_open_graph_image(url) is_image = is_album or self.is_image(url) # If it is from an album or not an actual image file, # just use open graph image. if is_album or not is_image: # Failed to follow link to find an image preview so # use placeholder image and guess filename if image_info is None: return None image_info["is_image"] = is_image return image_info # Otherwise, try to retrieve the actual image. # This is because open graph image from Dropbox may have padding # and gifs do not work. # TODO: What if image is huge? Should we get headers first? if image_info is None: image_info = {} image_info['is_image'] = True parsed_url_list = list(parsed_url) parsed_url_list[4] = "dl=1" # Replaces query image_info["image"] = urllib.parse.urlunparse(parsed_url_list) return image_info return None def youtube_id(self, url: str) -> Optional[str]: if not self.md.image_preview_enabled: return None # Youtube video id extraction regular expression from https://pastebin.com/KyKAFv1s # Slightly modified to support URLs of the forms # - youtu.be/ # - youtube.com/playlist?v=&list= # - youtube.com/watch_videos?video_ids=,, # If it matches, match.group(2) is the video id. schema_re = r'(?:https?://)' host_re = r'(?:youtu\.be/|(?:\w+\.)?youtube(?:-nocookie)?\.com/)' param_re = r'(?:(?:(?:v|embed)/)|' + \ r'(?:(?:(?:watch|playlist)(?:_popup|_videos)?(?:\.php)?)?(?:\?|#!?)(?:.+&)?v(?:ideo_ids)?=))' id_re = r'([0-9A-Za-z_-]+)' youtube_re = r'^({schema_re}?{host_re}{param_re}?)?{id_re}(?(1).+)?$' youtube_re = youtube_re.format(schema_re=schema_re, host_re=host_re, id_re=id_re, param_re=param_re) match = re.match(youtube_re, url) # URLs of the form youtube.com/playlist?list= are incorrectly matched if match is None or match.group(2) == 'playlist': return None return match.group(2) def youtube_title(self, extracted_data: Dict[str, Any]) -> Optional[str]: title = extracted_data.get("title") if title is not None: return f"YouTube - {title}" return None def youtube_image(self, url: str) -> Optional[str]: yt_id = self.youtube_id(url) if yt_id is not None: return f"https://i.ytimg.com/vi/{yt_id}/default.jpg" return None def vimeo_id(self, url: str) -> Optional[str]: if not self.md.image_preview_enabled: return None #(http|https)?:\/\/(www\.)?vimeo.com\/(?:channels\/(?:\w+\/)?|groups\/([^\/]*)\/videos\/|)(\d+)(?:|\/\?) # If it matches, match.group('id') is the video id. vimeo_re = r'^((http|https)?:\/\/(www\.)?vimeo.com\/' + \ r'(?:channels\/(?:\w+\/)?|groups\/' + \ r'([^\/]*)\/videos\/|)(\d+)(?:|\/\?))$' match = re.match(vimeo_re, url) if match is None: return None return match.group(5) def vimeo_title(self, extracted_data: Dict[str, Any]) -> Optional[str]: title = extracted_data.get("title") if title is not None: return f"Vimeo - {title}" return None def twitter_text(self, text: str, urls: List[Dict[str, str]], user_mentions: List[Dict[str, Any]], media: List[Dict[str, Any]]) -> Element: """ Use data from the twitter API to turn links, mentions and media into A tags. Also convert unicode emojis to images. This works by using the urls, user_mentions and media data from the twitter API and searching for unicode emojis in the text using `unicode_emoji_regex`. The first step is finding the locations of the URLs, mentions, media and emoji in the text. For each match we build a dictionary with type, the start location, end location, the URL to link to, and the text(codepoint and title in case of emojis) to be used in the link(image in case of emojis). Next we sort the matches by start location. And for each we add the text from the end of the last link to the start of the current link to the output. The text needs to added to the text attribute of the first node (the P tag) or the tail the last link created. Finally we add any remaining text to the last node. """ to_process: List[Dict[str, Any]] = [] # Build dicts for URLs for url_data in urls: short_url = url_data["url"] full_url = url_data["expanded_url"] for match in re.finditer(re.escape(short_url), text, re.IGNORECASE): to_process.append({ 'type': 'url', 'start': match.start(), 'end': match.end(), 'url': short_url, 'text': full_url, }) # Build dicts for mentions for user_mention in user_mentions: screen_name = user_mention['screen_name'] mention_string = '@' + screen_name for match in re.finditer(re.escape(mention_string), text, re.IGNORECASE): to_process.append({ 'type': 'mention', 'start': match.start(), 'end': match.end(), 'url': 'https://twitter.com/' + urllib.parse.quote(screen_name), 'text': mention_string, }) # Build dicts for media for media_item in media: short_url = media_item['url'] expanded_url = media_item['expanded_url'] for match in re.finditer(re.escape(short_url), text, re.IGNORECASE): to_process.append({ 'type': 'media', 'start': match.start(), 'end': match.end(), 'url': short_url, 'text': expanded_url, }) # Build dicts for emojis for match in re.finditer(unicode_emoji_regex, text, re.IGNORECASE): orig_syntax = match.group('syntax') codepoint = unicode_emoji_to_codepoint(orig_syntax) if codepoint in codepoint_to_name: display_string = ':' + codepoint_to_name[codepoint] + ':' to_process.append({ 'type': 'emoji', 'start': match.start(), 'end': match.end(), 'codepoint': codepoint, 'title': display_string, }) to_process.sort(key=lambda x: x['start']) p = current_node = Element('p') def set_text(text: str) -> None: """ Helper to set the text or the tail of the current_node """ if current_node == p: current_node.text = text else: current_node.tail = text db_data = self.md.zulip_db_data current_index = 0 for item in to_process: # The text we want to link starts in already linked text skip it if item['start'] < current_index: continue # Add text from the end of last link to the start of the current # link set_text(text[current_index:item['start']]) current_index = item['end'] if item['type'] != 'emoji': elem = url_to_a(db_data, item['url'], item['text']) assert isinstance(elem, Element) else: elem = make_emoji(item['codepoint'], item['title']) current_node = elem p.append(elem) # Add any unused text set_text(text[current_index:]) return p def twitter_link(self, url: str) -> Optional[Element]: tweet_id = get_tweet_id(url) if tweet_id is None: return None try: res = fetch_tweet_data(tweet_id) if res is None: return None user: Dict[str, Any] = res['user'] tweet = Element("div") tweet.set("class", "twitter-tweet") img_a = SubElement(tweet, 'a') img_a.set("href", url) profile_img = SubElement(img_a, 'img') profile_img.set('class', 'twitter-avatar') # For some reason, for, e.g. tweet 285072525413724161, # python-twitter does not give us a # profile_image_url_https, but instead puts that URL in # profile_image_url. So use _https if available, but fall # back gracefully. image_url = user.get('profile_image_url_https', user['profile_image_url']) profile_img.set('src', image_url) text = html.unescape(res['full_text']) urls = res.get('urls', []) user_mentions = res.get('user_mentions', []) media: List[Dict[str, Any]] = res.get('media', []) p = self.twitter_text(text, urls, user_mentions, media) tweet.append(p) span = SubElement(tweet, 'span') span.text = "- {} (@{})".format(user['name'], user['screen_name']) # Add image previews for media_item in media: # Only photos have a preview image if media_item['type'] != 'photo': continue # Find the image size that is smaller than # TWITTER_MAX_IMAGE_HEIGHT px tall or the smallest size_name_tuples = list(media_item['sizes'].items()) size_name_tuples.sort(reverse=True, key=lambda x: x[1]['h']) for size_name, size in size_name_tuples: if size['h'] < self.TWITTER_MAX_IMAGE_HEIGHT: break media_url = '{}:{}'.format(media_item['media_url_https'], size_name) img_div = SubElement(tweet, 'div') img_div.set('class', 'twitter-image') img_a = SubElement(img_div, 'a') img_a.set('href', media_item['url']) img = SubElement(img_a, 'img') img.set('src', media_url) return tweet except Exception: # We put this in its own try-except because it requires external # connectivity. If Twitter flakes out, we don't want to not-render # the entire message; we just want to not show the Twitter preview. markdown_logger.warning("Error building Twitter link", exc_info=True) return None def get_url_data(self, e: Element) -> Optional[Tuple[str, Optional[str]]]: if e.tag == "a": url = e.get("href") assert url is not None return (url, e.text) return None def get_inlining_information( self, root: Element, found_url: ResultWithFamily[Tuple[str, Optional[str]]], ) -> LinkInfo: grandparent = found_url.family.grandparent parent = found_url.family.parent ahref_element = found_url.family.child (url, text) = found_url.result # url != text usually implies a named link, which we opt not to remove url_eq_text = text is None or url == text title = None if url_eq_text else text info: LinkInfo = { 'parent': root, 'title': title, 'index': None, 'remove': None, } if parent.tag == 'li': info['parent'] = parent if not parent.text and not ahref_element.tail and url_eq_text: info['remove'] = ahref_element elif parent.tag == 'p': assert grandparent is not None parent_index = None for index, uncle in enumerate(grandparent): if uncle is parent: parent_index = index break # Append to end of list of grandparent's children as normal info['parent'] = grandparent if (len(parent) == 1 and (not parent.text or parent.text == "\n") and not ahref_element.tail and url_eq_text): info['remove'] = parent if parent_index is not None: info['index'] = self.find_proper_insertion_index(grandparent, parent, parent_index) return info def handle_image_inlining( self, root: Element, found_url: ResultWithFamily[Tuple[str, Optional[str]]], ) -> None: info = self.get_inlining_information(root, found_url) (url, text) = found_url.result actual_url = self.get_actual_image_url(url) self.add_a(info['parent'], actual_url, url, title=info['title'], insertion_index=info['index']) if info['remove'] is not None: info['parent'].remove(info['remove']) def handle_tweet_inlining( self, root: Element, found_url: ResultWithFamily[Tuple[str, Optional[str]]], twitter_data: Element, ) -> None: info = self.get_inlining_information(root, found_url) if info['index'] is not None: div = Element("div") root.insert(info['index'], div) else: div = SubElement(root, "div") div.set("class", "inline-preview-twitter") div.insert(0, twitter_data) def handle_youtube_url_inlining( self, root: Element, found_url: ResultWithFamily[Tuple[str, Optional[str]]], yt_image: str, ) -> None: info = self.get_inlining_information(root, found_url) (url, text) = found_url.result yt_id = self.youtube_id(url) self.add_a(info['parent'], yt_image, url, None, None, "youtube-video message_inline_image", yt_id, insertion_index=info['index'], already_thumbnailed=True) def find_proper_insertion_index(self, grandparent: Element, parent: Element, parent_index_in_grandparent: int) -> int: # If there are several inline images from same paragraph, ensure that # they are in correct (and not opposite) order by inserting after last # inline image from paragraph 'parent' parent_links = [ele.attrib['href'] for ele in parent.iter(tag="a")] insertion_index = parent_index_in_grandparent while True: insertion_index += 1 if insertion_index >= len(grandparent): return insertion_index uncle = grandparent[insertion_index] inline_image_classes = ['message_inline_image', 'message_inline_ref', 'inline-preview-twitter'] if ( uncle.tag != 'div' or 'class' not in uncle.keys() or uncle.attrib['class'] not in inline_image_classes ): return insertion_index uncle_link = list(uncle.iter(tag="a"))[0].attrib['href'] if uncle_link not in parent_links: return insertion_index def is_absolute_url(self, url: str) -> bool: return bool(urllib.parse.urlparse(url).netloc) def run(self, root: Element) -> None: # Get all URLs from the blob found_urls = walk_tree_with_family(root, self.get_url_data) unique_urls = {found_url.result[0] for found_url in found_urls} # Collect unique URLs which are not quoted as we don't do # inline previews for links inside blockquotes. unique_previewable_urls = {found_url.result[0] for found_url in found_urls if not found_url.family.in_blockquote} # Set has_link and similar flags whenever a message is processed by Markdown if self.md.zulip_message: self.md.zulip_message.has_link = len(found_urls) > 0 self.md.zulip_message.has_image = False # This is updated in self.add_a self.md.zulip_message.potential_attachment_path_ids = [] for url in unique_urls: # Due to rewrite_local_links_to_relative, we need to # handle both relative URLs beginning with # `/user_uploads` and beginning with `user_uploads`. # This urllib construction converts the latter into # the former. parsed_url = urllib.parse.urlsplit(urllib.parse.urljoin("/", url)) host = parsed_url.netloc if host != '' and host != self.md.zulip_realm.host: continue if not parsed_url.path.startswith("/user_uploads/"): continue path_id = parsed_url.path[len("/user_uploads/"):] self.md.zulip_message.potential_attachment_path_ids.append(path_id) if len(found_urls) == 0: return if len(unique_previewable_urls) > self.INLINE_PREVIEW_LIMIT_PER_MESSAGE: return processed_urls: Set[str] = set() rendered_tweet_count = 0 for found_url in found_urls: (url, text) = found_url.result if url in unique_previewable_urls and url not in processed_urls: processed_urls.add(url) else: continue if not self.is_absolute_url(url): if self.is_image(url): self.handle_image_inlining(root, found_url) # We don't have a strong use case for doing url preview for relative links. continue dropbox_image = self.dropbox_image(url) if dropbox_image is not None: class_attr = "message_inline_ref" is_image = dropbox_image["is_image"] if is_image: class_attr = "message_inline_image" # Not making use of title and description of images self.add_a(root, dropbox_image['image'], url, title=dropbox_image.get('title'), desc=dropbox_image.get('desc', ""), class_attr=class_attr, already_thumbnailed=True) continue if self.is_image(url): image_source = self.corrected_image_source(url) if image_source is not None: found_url = ResultWithFamily( family=found_url.family, result=(image_source, image_source), ) self.handle_image_inlining(root, found_url) continue if get_tweet_id(url) is not None: if rendered_tweet_count >= self.TWITTER_MAX_TO_PREVIEW: # Only render at most one tweet per message continue twitter_data = self.twitter_link(url) if twitter_data is None: # This link is not actually a tweet known to twitter continue rendered_tweet_count += 1 self.handle_tweet_inlining(root, found_url, twitter_data) continue youtube = self.youtube_image(url) if youtube is not None: self.handle_youtube_url_inlining(root, found_url, youtube) # NOTE: We don't `continue` here, to allow replacing the URL with # the title, if INLINE_URL_EMBED_PREVIEW feature is enabled. # The entire preview would ideally be shown only if the feature # is enabled, but URL previews are a beta feature and YouTube # previews are pretty stable. db_data = self.md.zulip_db_data if db_data and db_data['sent_by_bot']: continue if not self.md.url_embed_preview_enabled: continue try: extracted_data = link_preview.link_embed_data_from_cache(url) except NotFoundInCache: self.md.zulip_message.links_for_preview.add(url) continue if extracted_data: if youtube is not None: title = self.youtube_title(extracted_data) if title is not None: found_url.family.child.text = title continue self.add_embed(root, url, extracted_data) if self.vimeo_id(url): title = self.vimeo_title(extracted_data) if title: found_url.family.child.text = title class Timestamp(markdown.inlinepatterns.Pattern): def handleMatch(self, match: Match[str]) -> Optional[Element]: time_input_string = match.group('time') timestamp = None try: timestamp = dateutil.parser.parse(time_input_string, tzinfos=get_common_timezones()) except ValueError: try: timestamp = datetime.datetime.fromtimestamp(float(time_input_string)) except ValueError: pass if not timestamp: error_element = Element('span') error_element.set('class', 'timestamp-error') error_element.text = markdown.util.AtomicString( f"Invalid time format: {time_input_string}") return error_element # Use HTML5