discourse/lib/pretty_text.rb

# frozen_string_literal: true

require 'mini_racer'
require 'nokogiri'
require 'erb'

module PrettyText
  DANGEROUS_BIDI_CHARACTERS = [
    "\u202A",
    "\u202B",
    "\u202C",
    "\u202D",
    "\u202E",
    "\u2066",
    "\u2067",
    "\u2068",
    "\u2069",
  ].freeze
  DANGEROUS_BIDI_REGEXP = Regexp.new(DANGEROUS_BIDI_CHARACTERS.join("|")).freeze

  BLOCKED_HOTLINKED_SRC_ATTR = "data-blocked-hotlinked-src"
  BLOCKED_HOTLINKED_SRCSET_ATTR = "data-blocked-hotlinked-srcset"

  @mutex = Mutex.new
  @ctx_init = Mutex.new

  def self.app_root
    Rails.root
  end

  def self.find_file(root, filename)
    return filename if File.file?("#{root}#{filename}")

    es6_name = "#{filename}.js.es6"
    return es6_name if File.file?("#{root}#{es6_name}")

    js_name = "#{filename}.js"
    return js_name if File.file?("#{root}#{js_name}")

    erb_name = "#{filename}.js.es6.erb"
    return erb_name if File.file?("#{root}#{erb_name}")

    erb_name = "#{filename}.js.erb"
    return erb_name if File.file?("#{root}#{erb_name}")
  end

  def self.apply_es6_file(ctx, root_path, part_name)
    filename = find_file(root_path, part_name)
    if filename
      source = File.read("#{root_path}#{filename}")
      source = ERB.new(source).result(binding) if filename =~ /\.erb$/

      transpiler = DiscourseJsProcessor::Transpiler.new
      transpiled = transpiler.perform(source, "#{Rails.root}/app/assets/javascripts/", part_name)
      ctx.eval(transpiled)
    else
      # Look for vendored stuff
      vendor_root = "#{Rails.root}/vendor/assets/javascripts/"
      filename = find_file(vendor_root, part_name)
      if filename
        ctx.eval(File.read("#{vendor_root}#{filename}"))
      end
    end
  end

  def self.ctx_load_directory(ctx, path)
    root_path = "#{Rails.root}/app/assets/javascripts/"
    Dir["#{root_path}#{path}/**/*"].sort.each do |f|
      apply_es6_file(ctx, root_path, f.sub(root_path, '').sub(/\.js(.es6)?$/, ''))
    end
  end

  def self.create_es6_context
    ctx = MiniRacer::Context.new(timeout: 25000, ensure_gc_after_idle: 2000)

    ctx.eval("window = {}; window.devicePixelRatio = 2;") # hack to make code think stuff is retina

    ctx.attach("rails.logger.info", proc { |err| Rails.logger.info(err.to_s) })
    ctx.attach("rails.logger.warn", proc { |err| Rails.logger.warn(err.to_s) })
    ctx.attach("rails.logger.error", proc { |err| Rails.logger.error(err.to_s) })
    ctx.eval <<~JS
      console = {
        prefix: "[PrettyText] ",
        log: function(...args){ rails.logger.info(console.prefix + args.join(" ")); },
        warn: function(...args){ rails.logger.warn(console.prefix + args.join(" ")); },
        error: function(...args){ rails.logger.error(console.prefix + args.join(" ")); }
      }
    JS

    ctx.eval("__PRETTY_TEXT = true")

    PrettyText::Helpers.instance_methods.each do |method|
      ctx.attach("__helpers.#{method}", PrettyText::Helpers.method(method))
    end

    root_path = "#{Rails.root}/app/assets/javascripts/"
    ctx_load(ctx, "#{root_path}/node_modules/loader.js/dist/loader/loader.js")
    ctx_load(ctx, "#{root_path}/handlebars-shim.js")
    ctx_load(ctx, "#{root_path}/node_modules/xss/dist/xss.js")
    ctx.load("#{Rails.root}/lib/pretty_text/vendor-shims.js")
    ctx_load_directory(ctx, "pretty-text/addon")
    ctx_load_directory(ctx, "pretty-text/engines/discourse-markdown")
    ctx_load(ctx, "#{root_path}/node_modules/markdown-it/dist/markdown-it.js")

    apply_es6_file(ctx, root_path, "discourse-common/addon/lib/get-url")
    apply_es6_file(ctx, root_path, "discourse-common/addon/lib/object")
    apply_es6_file(ctx, root_path, "discourse-common/addon/lib/deprecated")
    apply_es6_file(ctx, root_path, "discourse-common/addon/lib/escape")
    apply_es6_file(ctx, root_path, "discourse-common/addon/utils/watched-words")
    apply_es6_file(ctx, root_path, "discourse/app/lib/to-markdown")
    apply_es6_file(ctx, root_path, "discourse/app/lib/utilities")

    ctx.load("#{Rails.root}/lib/pretty_text/shims.js")
    ctx.eval("__setUnicode(#{Emoji.unicode_replacements_json})")

    to_load = []
    DiscoursePluginRegistry.each_globbed_asset do |a|
      to_load << a if File.file?(a) && a =~ /discourse-markdown/
    end
    to_load.uniq.each do |f|
      if f =~ /^.+assets\/javascripts\//
        root = Regexp.last_match[0]
        apply_es6_file(ctx, root, f.sub(root, '').sub(/\.js(\.es6)?$/, ''))
      end
    end

    DiscoursePluginRegistry.vendored_core_pretty_text.each do |vpt|
      ctx.eval(File.read(vpt))
    end

    DiscoursePluginRegistry.vendored_pretty_text.each do |vpt|
      ctx.eval(File.read(vpt))
    end

    ctx
  end

  def self.v8
    return @ctx if @ctx

    # ensure we only init one of these
    @ctx_init.synchronize do
      return @ctx if @ctx
      @ctx = create_es6_context
    end

    @ctx
  end

  def self.reset_translations
    v8.eval("__resetTranslationTree()")
  end

  def self.reset_context
    @ctx_init.synchronize do
      @ctx&.dispose
      @ctx = nil
    end
  end

  # Acceptable options:
  #
  #  disable_emojis    - Disables the emoji markdown engine.
  #  features          - A hash where the key is the markdown feature name and the value is a boolean to enable/disable the markdown feature.
  #                      The hash is merged into the default features set in pretty-text.js which can be used to add new features or disable existing features.
  #  features_override - An array of markdown feature names to override the default markdown feature set. Currently used by plugins to customize what features should be enabled
  #                      when rendering markdown.
  #  markdown_it_rules - An array of markdown rule names which will be applied to the markdown-it engine. Currently used by plugins to customize what markdown-it rules should be
  #                      enabled when rendering markdown.
  #  topic_id          - Topic id for the post being cooked.
  #  user_id           - User id for the post being cooked.
  #  force_quote_link  - Always create the link to the quoted topic for [quote] bbcode. Normally this only happens
  #                      if the topic_id provided is different from the [quote topic:X].
  #  hashtag_context   - Defaults to "topic-composer" if not supplied. Controls the order of #hashtag lookup results
  #                      based on registered hashtag contexts from the `#register_hashtag_search_param` plugin API
  #                      method.
  def self.markdown(text, opts = {})
    # we use the exact same markdown converter as the client
    # TODO: use the same extensions on both client and server (in particular the template for mentions)
    baked = nil
    text = text || ""

    protect do
      context = v8

      custom_emoji = {}
      Emoji.custom.map { |e| custom_emoji[e.name] = e.url }

      # note, any additional options added to __optInput here must be
      # also be added to the buildOptions function in pretty-text.js,
      # otherwise they will be discarded
      buffer = +<<~JS
        __optInput = {};
        __optInput.siteSettings = #{SiteSetting.client_settings_json};
        #{"__optInput.disableEmojis = true" if opts[:disable_emojis]}
        __paths = #{paths_json};
        __optInput.getURL = __getURL;
        #{"__optInput.features = #{opts[:features].to_json};" if opts[:features]}
        #{"__optInput.featuresOverride = #{opts[:features_override].to_json};" if opts[:features_override]}
        #{"__optInput.markdownItRules = #{opts[:markdown_it_rules].to_json};" if opts[:markdown_it_rules]}
        __optInput.getCurrentUser = __getCurrentUser;
        __optInput.lookupAvatar = __lookupAvatar;
        __optInput.lookupPrimaryUserGroup = __lookupPrimaryUserGroup;
        __optInput.formatUsername = __formatUsername;
        __optInput.getTopicInfo = __getTopicInfo;
        __optInput.categoryHashtagLookup = __categoryLookup;
        __optInput.hashtagLookup = __hashtagLookup;
        __optInput.customEmoji = #{custom_emoji.to_json};
        __optInput.customEmojiTranslation = #{Plugin::CustomEmoji.translations.to_json};
        __optInput.emojiUnicodeReplacer = __emojiUnicodeReplacer;
        __optInput.lookupUploadUrls = __lookupUploadUrls;
        __optInput.censoredRegexp = #{WordWatcher.serializable_word_matcher_regexp(:censor).to_json };
        __optInput.watchedWordsReplace = #{WordWatcher.word_matcher_regexps(:replace).to_json};
        __optInput.watchedWordsLink = #{WordWatcher.word_matcher_regexps(:link).to_json};
        __optInput.additionalOptions = #{Site.markdown_additional_options.to_json};
      JS

      if opts[:topic_id]
        buffer << "__optInput.topicId = #{opts[:topic_id].to_i};\n"
      end

      if opts[:force_quote_link]
        buffer << "__optInput.forceQuoteLink = #{opts[:force_quote_link]};\n"
      end

      if opts[:user_id]
        buffer << "__optInput.userId = #{opts[:user_id].to_i};\n"
        buffer << "__optInput.currentUser = #{User.find(opts[:user_id]).to_json}\n"
      end

      opts[:hashtag_context] = opts[:hashtag_context] || "topic-composer"
      hashtag_types_as_js = HashtagAutocompleteService.ordered_types_for_context(
        opts[:hashtag_context]
      ).map { |t| "'#{t}'" }.join(",")
      hashtag_icons_as_js = HashtagAutocompleteService.data_source_icons.map { |i| "'#{i}'" }.join(",")
      buffer << "__optInput.hashtagTypesInPriorityOrder = [#{hashtag_types_as_js}];\n"
      buffer << "__optInput.hashtagIcons = [#{hashtag_icons_as_js}];\n"

      buffer << "__textOptions = __buildOptions(__optInput);\n"
      buffer << ("__pt = new __PrettyText(__textOptions);")

      # Be careful disabling sanitization. We allow for custom emails
      if opts[:sanitize] == false
        buffer << ('__pt.disableSanitizer();')
      end

      opts = context.eval(buffer)

      DiscourseEvent.trigger(:markdown_context, context)
      baked = context.eval("__pt.cook(#{text.inspect})")
    end

    baked
  end

  def self.paths_json
    paths = {
      baseUri: Discourse.base_path,
      CDN: Rails.configuration.action_controller.asset_host,
    }

    if SiteSetting.Upload.enable_s3_uploads
      if SiteSetting.Upload.s3_cdn_url.present?
        paths[:S3CDN] = SiteSetting.Upload.s3_cdn_url
      end
      paths[:S3BaseUrl] = Discourse.store.absolute_base_url
    end

    paths.to_json
  end

  # leaving this here, cause it invokes v8, don't want to implement twice
  def self.avatar_img(avatar_template, size)
    protect do
      v8.eval(<<~JS)
        __paths = #{paths_json};
        __utils.avatarImg({size: #{size.inspect}, avatarTemplate: #{avatar_template.inspect}}, __getURL);
      JS
    end
  end

  def self.unescape_emoji(title)
    return title unless SiteSetting.enable_emoji? && title

    set = SiteSetting.emoji_set.inspect
    custom = Emoji.custom.map { |e| [e.name, e.url] }.to_h.to_json

    protect do
      v8.eval(<<~JS)
        __paths = #{paths_json};
        __performEmojiUnescape(#{title.inspect}, {
          getURL: __getURL,
          emojiSet: #{set},
          emojiCDNUrl: "#{SiteSetting.external_emoji_url.blank? ? "" : SiteSetting.external_emoji_url}",
          customEmoji: #{custom},
          enableEmojiShortcuts: #{SiteSetting.enable_emoji_shortcuts},
          inlineEmoji: #{SiteSetting.enable_inline_emoji_translation}
        });
      JS
    end
  end

  def self.escape_emoji(title)
    return unless title

    replace_emoji_shortcuts = SiteSetting.enable_emoji && SiteSetting.enable_emoji_shortcuts

    protect do
      v8.eval(<<~JS)
        __performEmojiEscape(#{title.inspect}, {
          emojiShortcuts: #{replace_emoji_shortcuts},
          inlineEmoji: #{SiteSetting.enable_inline_emoji_translation}
        });
      JS
    end
  end

  def self.cook(text, opts = {})
    options = opts.dup
    working_text = text.dup

    sanitized = markdown(working_text, options)

    doc = Nokogiri::HTML5.fragment(sanitized)

    add_nofollow = !options[:omit_nofollow] && SiteSetting.add_rel_nofollow_to_user_content
    add_rel_attributes_to_user_content(doc, add_nofollow)
    strip_hidden_unicode_bidirectional_characters(doc)
    sanitize_hotlinked_media(doc)

    if SiteSetting.enable_mentions
      add_mentions(doc, user_id: opts[:user_id])
    end

    scrubber = Loofah::Scrubber.new do |node|
      node.remove if node.name == 'script'
    end
    loofah_fragment = Loofah.fragment(doc.to_html)
    loofah_fragment.scrub!(scrubber).to_html
  end

  def self.strip_hidden_unicode_bidirectional_characters(doc)
    return if !DANGEROUS_BIDI_REGEXP.match?(doc.content)

    doc.css("code,pre").each do |code_tag|
      next if !DANGEROUS_BIDI_REGEXP.match?(code_tag.content)

      DANGEROUS_BIDI_CHARACTERS.each do |bidi|
        next if !code_tag.content.include?(bidi)

        formatted = "&lt;U+#{bidi.ord.to_s(16).upcase}&gt;"
        code_tag.inner_html = code_tag.inner_html.gsub(
          bidi,
          "<span class=\"bidi-warning\" title=\"#{I18n.t("post.hidden_bidi_character")}\">#{formatted}</span>"
        )
      end
    end
  end

  def self.sanitize_hotlinked_media(doc)
    return if !SiteSetting.block_hotlinked_media

    allowed_pattern = allowed_src_pattern

    doc.css("img[src], source[src], source[srcset], track[src]").each do |el|
      if el["src"] && !el["src"].match?(allowed_pattern)
        el[PrettyText::BLOCKED_HOTLINKED_SRC_ATTR] = el.delete("src")
      end

      if el["srcset"]
        srcs = el["srcset"].split(',').map { |e| e.split(' ', 2)[0].presence }
        if srcs.any? { |src| !src.match?(allowed_pattern) }
          el[PrettyText::BLOCKED_HOTLINKED_SRCSET_ATTR] = el.delete("srcset")
        end
      end
    end
  end

  def self.add_rel_attributes_to_user_content(doc, add_nofollow)
    allowlist = []

    domains = SiteSetting.exclude_rel_nofollow_domains
    allowlist = domains.split('|') if domains.present?

    site_uri = nil
    doc.css("a").each do |l|
      href = l["href"].to_s
      l["rel"] = "noopener" if l["target"] == "_blank"

      begin
        uri = URI(UrlHelper.encode_component(href))
        site_uri ||= URI(Discourse.base_url)

        same_domain = !uri.host.present? ||
          uri.host == site_uri.host ||
          uri.host.ends_with?(".#{site_uri.host}") ||
          allowlist.any? { |u| uri.host == u || uri.host.ends_with?(".#{u}") }

        l["rel"] = "noopener nofollow ugc" if add_nofollow && !same_domain
      rescue URI::Error
        # add a nofollow anyway
        l["rel"] = "noopener nofollow ugc"
      end
    end
  end

  class DetectedLink < Struct.new(:url, :is_quote); end

  def self.extract_links(html)
    links = []
    doc = Nokogiri::HTML5.fragment(html)

    # extract onebox links
    doc.css("aside.onebox[data-onebox-src]").each { |onebox| links << DetectedLink.new(onebox["data-onebox-src"], false) }

    # remove href inside quotes & oneboxes & elided part
    doc.css("aside.quote a, aside.onebox a, .elided a").remove

    # remove hotlinked images
    doc.css("a.onebox > img").each { |img| img.parent.remove }

    # extract all links
    doc.css("a").each do |a|
      if a["href"].present? && a["href"][0] != "#"
        links << DetectedLink.new(a["href"], false)
      end
    end

    # extract quotes
    doc.css("aside.quote[data-topic]").each do |aside|
      if aside["data-topic"].present?
        url = +"/t/#{aside["data-topic"]}"
        url << "/#{aside["data-post"]}" if aside["data-post"].present?
        links << DetectedLink.new(url, true)
      end
    end

    # extract Youtube links
    doc.css("div[data-youtube-id]").each do |div|
      if div["data-youtube-id"].present?
        links << DetectedLink.new("https://www.youtube.com/watch?v=#{div['data-youtube-id']}", false)
      end
    end

    links
  end

  def self.excerpt(html, max_length, options = {})
    # TODO: properly fix this HACK in ExcerptParser without introducing XSS
    doc = Nokogiri::HTML5.fragment(html)
    DiscourseEvent.trigger(:reduce_excerpt, doc, options)
    strip_image_wrapping(doc)
    strip_oneboxed_media(doc)
    html = doc.to_html
    ExcerptParser.get_excerpt(html, max_length, options)
  end

  def self.strip_links(string)
    return string if string.blank?

    # If the user is not basic, strip links from their bio
    fragment = Nokogiri::HTML5.fragment(string)
    fragment.css('a').each { |a| a.replace(a.inner_html) }
    fragment.to_html
  end

  def self.make_all_links_absolute(doc)
    site_uri = nil
    doc.css("a").each do |link|
      href = link["href"].to_s
      begin
        uri = URI(href)
        site_uri ||= URI(Discourse.base_url)
        unless uri.host.present? || href.start_with?('mailto')
          link["href"] = "#{site_uri}#{link['href']}"
        end
      rescue URI::Error
        # leave it
      end
    end
  end

  def self.strip_image_wrapping(doc)
    doc.css(".lightbox-wrapper .meta").remove
  end

  def self.strip_oneboxed_media(doc)
    doc.css("audio").remove
    doc.css(".video-onebox,video").remove
  end

  def self.convert_vimeo_iframes(doc)
    doc.css("iframe[src*='player.vimeo.com']").each do |iframe|
      if iframe["data-original-href"].present?
        vimeo_url = UrlHelper.normalized_encode(iframe["data-original-href"])
      else
        vimeo_id = iframe['src'].split('/').last
        vimeo_url = "https://vimeo.com/#{vimeo_id}"
      end
      iframe.replace Nokogiri::HTML5.fragment("<p><a href='#{vimeo_url}'>#{vimeo_url}</a></p>")
    end
  end

  def self.strip_secure_uploads(doc)
    # images inside a lightbox or other link
    doc.css('a[href]').each do |a|
      next if !Upload.secure_uploads_url?(a['href'])

      non_image_media = %w(video audio).include?(a&.parent&.name)
      target = non_image_media ? a.parent : a
      next if target.to_s.include?('stripped-secure-view-media') || target.to_s.include?('stripped-secure-view-upload')

      next if a.css('img[src]').empty? && !non_image_media

      if a.classes.include?('lightbox')
        img = a.css('img[src]').first
        srcset = img&.attributes['srcset']&.value
        if srcset
          # if available, use the first image from the srcset here
          # so we get the optimized image instead of the possibly huge original
          url = srcset.split(',').first
        else
          url = img['src']
        end
        a.add_next_sibling secure_uploads_placeholder(doc, url, width: img['width'], height: img['height'])
        a.remove
      else
        width = non_image_media ? nil : a.at_css('img').attr('width')
        height = non_image_media ? nil : a.at_css('img').attr('height')
        target.add_next_sibling secure_uploads_placeholder(doc, a['href'], width: width, height: height)
        target.remove
      end
    end

    # images by themselves or inside a onebox
    doc.css('img[src]').each do |img|
      url = if img.parent.classes.include?("aspect-image") && img.attributes["srcset"].present?

        # we are using the first image from the srcset here so we get the
        # optimized image instead of the original, because an optimized
        # image may be used for the onebox thumbnail
        srcset = img.attributes["srcset"].value
        srcset.split(",").first
      else
        img['src']
      end

      width = img['width']
      height = img['height']
      onebox_type = nil

      if img.ancestors.css(".onebox-body").any?
        if img.classes.include?("onebox-avatar-inline")
          onebox_type = "avatar-inline"
        else
          onebox_type = "thumbnail"
        end
      end

      # we always want this to be tiny and without any special styles
      if img.classes.include?('site-icon')
        onebox_type = nil
        width = 16
        height = 16
      end

      if Upload.secure_uploads_url?(url)
        img.add_next_sibling secure_uploads_placeholder(doc, url, onebox_type: onebox_type, width: width, height: height)
        img.remove
      end
    end
  end

  def self.secure_uploads_placeholder(doc, url, onebox_type: false, width: nil, height: nil)
    data_width = width ? "data-width=#{width}" : ''
    data_height = height ? "data-height=#{height}" : ''
    data_onebox_type = onebox_type ? "data-onebox-type='#{onebox_type}'" : ''
    <<~HTML
    <div class="secure-upload-notice" data-stripped-secure-upload="#{url}" #{data_onebox_type} #{data_width} #{data_height}>
      #{I18n.t('emails.secure_uploads_placeholder')} <a class='stripped-secure-view-upload' href="#{url}">#{I18n.t("emails.view_redacted_media")}</a>.
    </div>
    HTML
  end

  def self.format_for_email(html, post = nil)
    doc = Nokogiri::HTML5.fragment(html)
    DiscourseEvent.trigger(:reduce_cooked, doc, post)
    strip_secure_uploads(doc) if post&.with_secure_uploads?
    strip_image_wrapping(doc)
    convert_vimeo_iframes(doc)
    make_all_links_absolute(doc)
    doc.to_html
  end

  protected

  class JavaScriptError < StandardError
    attr_accessor :message, :backtrace

    def initialize(message, backtrace)
      @message = message
      @backtrace = backtrace
    end

  end

  def self.protect
    rval = nil
    @mutex.synchronize do
      rval = yield
    end
    rval
  end

  def self.ctx_load(ctx, *files)
    files.each do |file|
      ctx.load(app_root + file)
    end
  end

  private

  USER_TYPE ||= 'user'
  GROUP_TYPE ||= 'group'
  GROUP_MENTIONABLE_TYPE ||= 'group-mentionable'

  def self.add_mentions(doc, user_id: nil)
    elements = doc.css("span.mention")
    names = elements.map { |element| element.text[1..-1] }

    mentions = lookup_mentions(names, user_id: user_id)

    elements.each do |element|
      name = element.text[1..-1]
      name.downcase!

      if type = mentions[name]
        element.name = 'a'

        element.children = PrettyText::Helpers.format_username(
          element.children.text
        )

        case type
        when USER_TYPE
          element['href'] = "#{Discourse.base_path}/u/#{UrlHelper.encode_component(name)}"
        when GROUP_MENTIONABLE_TYPE
          element['class'] = 'mention-group notify'
          element['href'] = "#{Discourse.base_path}/groups/#{UrlHelper.encode_component(name)}"
        when GROUP_TYPE
          element['class'] = 'mention-group'
          element['href'] = "#{Discourse.base_path}/groups/#{UrlHelper.encode_component(name)}"
        end
      end
    end
  end

  def self.lookup_mentions(names, user_id: nil)
    return {} if names.blank?

    sql = <<~SQL
    (
      SELECT
        :user_type AS type,
        username_lower AS name
      FROM users
      WHERE username_lower IN (:names) AND staged = false
    )
    UNION
    (
      SELECT
        :group_type AS type,
        lower(name) AS name
      FROM groups
    )
    UNION
    (
      SELECT
        :group_mentionable_type AS type,
        lower(name) AS name
      FROM groups
      WHERE lower(name) IN (:names) AND (#{Group.mentionable_sql_clause(include_public: false)})
    )
    ORDER BY type
    SQL

    user = User.find_by(id: user_id)
    names.each(&:downcase!)

    results = DB.query(sql,
      names: names,
      user_type: USER_TYPE,
      group_type: GROUP_TYPE,
      group_mentionable_type: GROUP_MENTIONABLE_TYPE,
      levels: Group.alias_levels(user),
      user_id: user_id
    )

    mentions = {}
    results.each { |result| mentions[result.name] = result.type }
    mentions
  end

  def self.allowed_src_pattern
    allowed_src_prefixes = [
      Discourse.base_path,
      Discourse.base_url,
      GlobalSetting.s3_cdn_url,
      GlobalSetting.cdn_url,
      SiteSetting.external_emoji_url.presence,
      *SiteSetting.block_hotlinked_media_exceptions.split("|")
    ]

    patterns = allowed_src_prefixes.compact.map do |url|
      pattern = Regexp.escape(url)

      # If 'https://example.com' is allowed, ensure 'https://example.com.blah.com' is not
      pattern += '(?:/|\z)' if !pattern.ends_with?("\/")

      pattern
    end

    /\A(data:|#{patterns.join("|")})/
  end
end