FIX: Prefer HTML in incoming emails, heavily refactor email receiver

This commit heavily refactors Email::Receiver to both better handle
different emails and improve testability.

A primary focus of the refactor is reducing the usage of class
variables, in favor of actually passing parameters - making it possible
for multiple tests to use the same Receiver instance.

The EmailLog reported when a topic is created is reflected to put the
user's email in the to_address field, instead of the system address.

The discourse_email_parser function is renamed to
discourse_email_trimmer, and additional stopping conditions are added to
make up for EmailReplyParser's inability to deal with html at the start
of a line.

The force_encoding calls are refactored out to a 'fix_charset' method.

parse_body is renamed to select_body, and the scrub_html method is
dropped in favor of the new HtmlCleaner class.

A new parse_body method is added, which performs the job of the removed
lines of code in the 'process' method.

EmailUnparsableError is redefined again, to be encoding errors (when the
declared encoding is not what was delivered).

lib/email/receiver.rb

@@ -1,3 +1,4 @@
+require 'email/html_cleaner'
 #
 # Handles an incoming message
 #
@@ -26,20 +27,12 @@ module Email
     def process
       raise EmptyEmailError if @raw.blank?
 
-      @message = Mail.new(@raw)
+      message = Mail.new(@raw)
 
-      # First remove the known discourse stuff.
-      parse_body
-      raise EmptyEmailError if @body.blank?
-
-      # Then run the github EmailReplyParser on it in case we didn't catch it
-      @body = EmailReplyParser.read(@body).visible_text.force_encoding('UTF-8')
-
-      discourse_email_parser
-      raise EmailUnparsableError if @body.blank?
+      body = parse_body message
 
       dest_info = {type: :invalid, obj: nil}
-      @message.to.each do |to_address|
+      message.to.each do |to_address|
         if dest_info[:type] == :invalid
           dest_info = check_address to_address
         end
@@ -47,6 +40,10 @@ module Email
 
       raise BadDestinationAddress if dest_info[:type] == :invalid
 
+      # TODO get to a state where we can remove this
+      @message = message
+      @body = body
+
       if dest_info[:type] == :category
         raise BadDestinationAddress unless SiteSetting.email_in
         category = dest_info[:obj]
@@ -74,6 +71,8 @@ module Email
 
         create_reply
       end
+    rescue Encoding::UndefinedConversionError, Encoding::InvalidByteSequenceError => e
+      raise EmailUnparsableError.new(e)
     end
 
     def check_address(address)
@@ -94,56 +93,63 @@ module Email
       {type: :invalid, obj: nil}
     end
 
+    def parse_body(message)
+      body = select_body message
+      raise EmptyEmailError if body.strip.blank?
 
-    def parse_body
+      body = discourse_email_trimmer body
+      raise EmptyEmailError if body.strip.blank?
+
+      body = EmailReplyParser.parse_reply body
+      raise EmptyEmailError if body.strip.blank?
+
+      body
+    end
+
+    def select_body(message)
       html = nil
-
-      # If the message is multipart, find the best type for our purposes
-      if @message.multipart?
-        if p = @message.text_part
-          @body = p.charset ? p.body.decoded.force_encoding(p.charset).encode("UTF-8").to_s : p.body.to_s
-          return @body
-        elsif p = @message.html_part
-          html = p.charset ? p.body.decoded.force_encoding(p.charset).encode("UTF-8").to_s : p.body.to_s
+      # If the message is multipart, return that part (favor html)
+      if message.multipart?
+        html = fix_charset message.html_part
+        text = fix_charset message.text_part
+        # TODO picking text if available may be better
+        if text && !html
+          return text
         end
+      elsif message.content_type =~ /text\/html/
+        html = fix_charset message
       end
 
-      if @message.content_type =~ /text\/html/
-        if defined? @message.charset
-          html = @message.body.decoded.force_encoding(@message.charset).encode("UTF-8").to_s
-        else
-          html = @message.body.to_s
-        end
+      if html
+        body = HtmlCleaner.new(html).output_html
+      else
+        body = fix_charset message
       end
 
-      if html.present?
-        @body = scrub_html(html)
-        return @body
-      end
-
-      @body = @message.charset ? @message.body.decoded.force_encoding(@message.charset).encode("UTF-8").to_s.strip : @message.body.to_s
-
       # Certain trigger phrases that means we didn't parse correctly
-      @body = nil if @body =~ /Content\-Type\:/ ||
-                     @body =~ /multipart\/alternative/ ||
-                     @body =~ /text\/plain/
+      if body =~ /Content\-Type\:/ || body =~ /multipart\/alternative/ || body =~ /text\/plain/
+        raise EmptyEmailError
+      end
 
-      @body
+      body
     end
 
-    def scrub_html(html)
-      # If we have an HTML message, strip the markup
-      doc = Nokogiri::HTML(html)
+    # Force encoding to UTF-8 on a Mail::Message or Mail::Part
+    def fix_charset(object)
+      return nil if object.nil?
 
-      # Blackberry is annoying in that it only provides HTML. We can easily extract it though
-      content = doc.at("#BB10_response_div")
-      return content.text if content.present?
-
-      doc.xpath("//text()").text
+      if object.charset
+        object.body.decoded.force_encoding(object.charset).encode("UTF-8").to_s
+      else
+        object.body.to_s
+      end
     end
 
-    def discourse_email_parser
-      lines = @body.scrub.lines.to_a
+    REPLYING_HEADER_LABELS = ['From', 'Sent', 'To', 'Subject', 'Reply To']
+    REPLYING_HEADER_REGEX = Regexp.union(REPLYING_HEADER_LABELS.map { |lbl| "#{lbl}:" })
+
+    def discourse_email_trimmer(body)
+      lines = body.scrub.lines.to_a
       range_end = 0
 
       lines.each_with_index do |l, idx|
@@ -154,11 +160,15 @@ module Email
                  # Let's try it and see how well it works.
                  (l =~ /\d{4}/ && l =~ /\d:\d\d/ && l =~ /\:$/)
 
+        # Headers on subsequent lines
+        break if (0..2).all? { |off| lines[idx+off] =~ REPLYING_HEADER_REGEX }
+        # Headers on the same line
+        break if REPLYING_HEADER_LABELS.count { |lbl| l.include? lbl } >= 3
+
         range_end = idx
       end
 
-      @body = lines[0..range_end].join
-      @body.strip!
+      lines[0..range_end].join.strip
     end
 
     def wrap_body_in_quote(user_email)