path: root/app/models/incoming_message.rb

# == Schema Information
# Schema version: 114
#
# Table name: incoming_messages
#
#  id                             :integer         not null, primary key
#  info_request_id                :integer         not null
#  created_at                     :datetime        not null
#  updated_at                     :datetime        not null
#  raw_email_id                   :integer         not null
#  cached_attachment_text_clipped :text
#  cached_main_body_text_folded   :text
#  cached_main_body_text_unfolded :text
#  subject                        :text
#  mail_from_domain               :text
#  valid_to_reply_to              :boolean
#  last_parsed                    :datetime
#  mail_from                      :text
#  sent_at                        :datetime
#

# encoding: UTF-8

# models/incoming_message.rb:
# An (email) message from really anybody to be logged with a request. e.g. A
# response from the public body.
#
# Copyright (c) 2007 UK Citizens Online Democracy. All rights reserved.
# Email: francis@mysociety.org; WWW: http://www.mysociety.org/
#
# $Id: incoming_message.rb,v 1.228 2009-10-21 11:24:14 francis Exp $

# TODO
# Move some of the (e.g. quoting) functions here into rblib, as they feel
# general not specific to IncomingMessage.

require 'alaveteli_file_types'
require 'htmlentities'
require 'rexml/document'
require 'zip/zip'
require 'mapi/msg'
require 'mapi/convert'

# Monkeypatch! Adding some extra members to store extra info in.
module TMail
    class Mail
        attr_accessor :url_part_number
        attr_accessor :rfc822_attachment # when a whole email message is attached as text
        attr_accessor :within_rfc822_attachment # for parts within a message attached as text (for getting subject mainly)
    end
end

class IncomingMessage < ActiveRecord::Base
    belongs_to :info_request
    validates_presence_of :info_request

    validates_presence_of :raw_email

    has_many :outgoing_message_followups, :foreign_key => 'incoming_message_followup_id', :class_name => 'OutgoingMessage'
    has_many :foi_attachments, :order => 'id'
    has_many :info_request_events # never really has many, but could in theory

    belongs_to :raw_email

    # See binary_mask_stuff function below. It just test for inclusion
    # in this hash, not the value of the right hand side.
    DoNotBinaryMask = {
        'image/tiff' => 1,
        'image/gif' => 1,
        'image/jpeg' => 1,
        'image/png' => 1,
        'image/bmp' => 1,
        'application/zip' => 1,
    }

    # Return the structured TMail::Mail object
    # Documentation at http://i.loveruby.net/en/projects/tmail/doc/
    def mail(force = nil)
        if (!force.nil? || @mail.nil?) && !self.raw_email.nil?
            # Hack round bug in TMail's MIME decoding. Example request which provokes it:
            # http://www.whatdotheyknow.com/request/reviews_of_unduly_lenient_senten#incoming-4830
            # Report of TMail bug:
            # http://rubyforge.org/tracker/index.php?func=detail&aid=21810&group_id=4512&atid=17370
            copy_of_raw_data = self.raw_email.data.gsub(/; boundary=\s+"/ims,'; boundary="')

            @mail = TMail::Mail.parse(copy_of_raw_data)
            @mail.base64_decode
        end
        @mail
    end

    # Returns the name of the person the incoming message is from, or nil if
    # there isn't one or if there is only an email address. XXX can probably
    # remove from_name_if_present (which is a monkey patch) by just calling
    # .from_addrs[0].name here instead?

    # Return false if for some reason this is a message that we shouldn't let them reply to
    def _calculate_valid_to_reply_to
        # check validity of email
        if self.mail.from_addrs.nil? || self.mail.from_addrs.size == 0
            return false
        end
        email = self.mail.from_addrs[0].spec
        if !MySociety::Validate.is_valid_email(email)
            return false
        end

        # reject postmaster - authorities seem to nearly always not respond to
        # email to postmaster, and it tends to only happen after delivery failure.
        # likewise Mailer-Daemon, Auto_Reply...
        prefix = email
        prefix =~ /^(.*)@/
        prefix = $1
        if !prefix.nil? && prefix.downcase.match(/^(postmaster|mailer-daemon|auto_reply|donotreply|no.reply)$/)
            return false
        end
        if !self.mail['return-path'].nil? && self.mail['return-path'].addr == "<>"
            return false
        end
        if !self.mail['auto-submitted'].nil?
            return false
        end
        return true
    end

    def parse_raw_email!(force = nil)
        # The following fields may be absent; we treat them as cached
        # values in case we want to regenerate them (due to mail
        # parsing bugs, etc).
        if self.raw_email.nil?
            raise "Incoming message id=#{id} has no raw_email"
        end
        if (!force.nil? || self.last_parsed.nil?)
            ActiveRecord::Base.transaction do
                self.extract_attachments!
                self.sent_at = self.mail.date || self.created_at
                self.subject = self.mail.subject
                # XXX can probably remove from_name_if_present (which is a
                # monkey patch) by just calling .from_addrs[0].name here
                # instead?
                self.mail_from = self.mail.from_name_if_present
                begin
                    self.mail_from_domain = PublicBody.extract_domain_from_email(self.mail.from_addrs[0].spec)
                rescue NoMethodError
                    self.mail_from_domain = ""
                end
                self.valid_to_reply_to = self._calculate_valid_to_reply_to
                self.last_parsed = Time.now
                self.save!
            end
        end
    end

    def valid_to_reply_to?
        return self.valid_to_reply_to
    end

    # The cached fields mentioned in the previous comment
    # XXX there must be a nicer way to do this without all that
    # repetition.  I tried overriding method_missing but got some
    # unpredictable results.
    def valid_to_reply_to
        parse_raw_email!
        super
    end
    def sent_at
        parse_raw_email!
        super
    end
    def subject
        parse_raw_email!
        super
    end
    def mail_from
        parse_raw_email!
        super
    end
    def safe_mail_from
        if !self.mail_from.nil?
            mail_from = self.mail_from.dup
            self.info_request.apply_censor_rules_to_text!(mail_from)
            return mail_from
        end
    end
    def mail_from_domain
        parse_raw_email!
        super
    end

    # Number the attachments in depth first tree order, for use in URLs.
    # XXX This fills in part.rfc822_attachment and part.url_part_number within
    # all the parts of the email (see TMail monkeypatch above for how these
    # attributes are added). ensure_parts_counted must be called before using
    # the attributes.
    def ensure_parts_counted
        @count_parts_count = 0
        _count_parts_recursive(self.mail)
        # we carry on using these numeric ids for attachments uudecoded from within text parts
        @count_first_uudecode_count = @count_parts_count
    end
    def _count_parts_recursive(part)
        if part.multipart?
            part.parts.each do |p|
                _count_parts_recursive(p)
            end
        else
            part_filename = TMail::Mail.get_part_file_name(part)
            begin
                if part.content_type == 'message/rfc822'
                    # An email attached as text
                    # e.g. http://www.whatdotheyknow.com/request/64/response/102
                    part.rfc822_attachment = TMail::Mail.parse(part.body)
                elsif part.content_type == 'application/vnd.ms-outlook' || part_filename && AlaveteliFileTypes.filename_to_mimetype(part_filename) == 'application/vnd.ms-outlook'
                    # An email attached as an Outlook file
                    # e.g. http://www.whatdotheyknow.com/request/chinese_names_for_british_politi
                    msg = Mapi::Msg.open(StringIO.new(part.body))
                    part.rfc822_attachment = TMail::Mail.parse(msg.to_mime.to_s)
                elsif part.content_type == 'application/ms-tnef'
                    # A set of attachments in a TNEF file
                    part.rfc822_attachment = TNEF.as_tmail(part.body)
                end
            rescue
                # If attached mail doesn't parse, treat it as text part
                part.rfc822_attachment = nil
            else
                unless part.rfc822_attachment.nil?
                    _count_parts_recursive(part.rfc822_attachment)
                end
            end
            if part.rfc822_attachment.nil?
                @count_parts_count += 1
                part.url_part_number = @count_parts_count
            end
        end
    end
    # And look up by URL part number to get an attachment
    # XXX relies on extract_attachments calling ensure_parts_counted
    def self.get_attachment_by_url_part_number(attachments, found_url_part_number)
        attachments.each do |a|
            if a.url_part_number == found_url_part_number
                return a
            end
        end
        return nil
    end

    # Converts email addresses we know about into textual descriptions of them
    def mask_special_emails!(text)
        # XXX can later display some of these special emails as actual emails,
        # if they are public anyway.  For now just be precautionary and only
        # put in descriptions of them in square brackets.
        if self.info_request.public_body.is_followupable?
            text.gsub!(self.info_request.public_body.request_email, _("[{{public_body}} request email]", :public_body => self.info_request.public_body.short_or_long_name))
        end
        text.gsub!(self.info_request.incoming_email, _('[FOI #{{request}} email]', :request => self.info_request.id.to_s) )
        text.gsub!(MySociety::Config.get("CONTACT_EMAIL", 'contact@localhost'), _("[{{site_name}} contact email]", :site_name => MySociety::Config.get('SITE_NAME', 'Alaveteli')) )
    end

    # Replaces all email addresses in (possibly binary data) with equal length alternative ones.
    # Also replaces censor items
    def binary_mask_stuff!(text, content_type)
        # See if content type is one that we mask - things like zip files and
        # images may get broken if we try to. We err on the side of masking too
        # much, as many unknown types will really be text.
        if DoNotBinaryMask.include?(content_type)
            return
        end

        # Special cases for some content types
        if content_type == 'application/pdf'
            uncompressed_text = nil
            uncompressed_text = AlaveteliExternalCommand.run("pdftk", "-", "output", "-", "uncompress", :stdin_string => text)
            # if we managed to uncompress the PDF...
            if !uncompressed_text.nil? && !uncompressed_text.empty?
                # then censor stuff (making a copy so can compare again in a bit)
                censored_uncompressed_text = uncompressed_text.dup
                self._binary_mask_stuff_internal!(censored_uncompressed_text)
                # if the censor rule removed something...
                if censored_uncompressed_text != uncompressed_text
                    # then use the altered file (recompressed)
                    recompressed_text = nil
                    if MySociety::Config.get('USE_GHOSTSCRIPT_COMPRESSION') == true
                        command = ["gs", "-sDEVICE=pdfwrite", "-dCompatibilityLevel=1.4", "-dPDFSETTINGS=/screen", "-dNOPAUSE", "-dQUIET", "-dBATCH", "-sOutputFile=-", "-"]
                    else
                        command = ["pdftk", "-", "output", "-", "compress"]
                    end
                    recompressed_text = AlaveteliExternalCommand.run(*(command + [{:stdin_string=>censored_uncompressed_text}]))
                    if recompressed_text.nil? || recompressed_text.empty?
                        # buggy versions of pdftk sometimes fail on
                        # compression, I don't see it's a disaster in
                        # these cases to save an uncompressed version?
                        recompressed_text = censored_uncompressed_text
                        logger.warn "Unable to compress PDF; problem with your pdftk version?"
                    end
                    if !recompressed_text.nil? && !recompressed_text.empty?
                        text[0..-1] = recompressed_text # [0..-1] makes it change the 'text' string in place
                    end
                end
            end
            return
        end

        self._binary_mask_stuff_internal!(text)
    end

    # Used by binary_mask_stuff - replace text in place
    def _binary_mask_stuff_internal!(text)
        # Keep original size, so can check haven't resized it
        orig_size = text.size

        # Replace ASCII email addresses...
        text.gsub!(MySociety::Validate.email_find_regexp) do |email|
            email.gsub(/[^@.]/, 'x')
        end

        # And replace UCS-2 ones (for Microsoft Office documents)...
        # Find emails, by finding them in parts of text that have ASCII
        # equivalents to the UCS-2
        ascii_chars = text.gsub(/\0/, "")
        emails = ascii_chars.scan(MySociety::Validate.email_find_regexp)
        # Convert back to UCS-2, making a mask at the same time
        emails.map! {|email| [
                Iconv.conv('ucs-2le', 'ascii', email[0]),
                Iconv.conv('ucs-2le', 'ascii', email[0].gsub(/[^@.]/, 'x'))
        ] }
        # Now search and replace the UCS-2 email with the UCS-2 mask
        for email, mask in emails
            text.gsub!(email, mask)
        end

        # Replace censor items
        self.info_request.apply_censor_rules_to_binary!(text)

        raise "internal error in binary_mask_stuff" if text.size != orig_size
        return text
    end

    # Removes censored stuff from from HTML conversion of downloaded binaries
    def html_mask_stuff!(html)
        self.mask_special_emails!(html)
        self.remove_privacy_sensitive_things!(html)
    end

    # Lotus notes quoting yeuch!
    def remove_lotus_quoting(text, replacement = "FOLDED_QUOTED_SECTION")
        text = text.dup
        name = Regexp.escape(self.info_request.user_name)

        # To end of message sections
        # http://www.whatdotheyknow.com/request/university_investment_in_the_arm
        text.gsub!(/^#{name}[^\n]+\nSent by:[^\n]+\n.*/ims, "\n\n" + replacement)

        # Some other sort of forwarding quoting
        # http://www.whatdotheyknow.com/request/224/response/326
        text.gsub!(/^#{name}[^\n]+\n[0-9\/:\s]+\s+To\s+FOI requests at.*/ims, "\n\n" + replacement)

        # http://www.whatdotheyknow.com/request/how_do_the_pct_deal_with_retirin_33#incoming-930
        # http://www.whatdotheyknow.com/request/229/response/809
        text.gsub!(/^From: [^\n]+\nSent: [^\n]+\nTo:\s+['"?]#{name}['"]?\nSubject:.*/ims, "\n\n" + replacement)

        return text

    end

    # Remove emails, mobile phones and other details FOI officers ask us to remove.
    def remove_privacy_sensitive_things!(text)
        # Remove any email addresses - we don't want bounce messages to leak out
        # either the requestor's email address or the request's response email
        # address out onto the internet
        text.gsub!(MySociety::Validate.email_find_regexp, "[email address]")

        # Mobile phone numbers
        # http://www.whatdotheyknow.com/request/failed_test_purchases_off_licenc#incoming-1013
        # http://www.whatdotheyknow.com/request/selective_licensing_statistics_i#incoming-550
        # http://www.whatdotheyknow.com/request/common_purpose_training_graduate#incoming-774
        text.gsub!(/(Mobile|Mob)([\s\/]*(Fax|Tel))*\s*:?[\s\d]*\d/, "[mobile number]")

        # Remove WhatDoTheyKnow signup links
        domain = MySociety::Config.get('DOMAIN')
        text.gsub!(/http:\/\/#{domain}\/c\/[^\s]+/, "[WDTK login link]")

        # Remove things from censor rules
        self.info_request.apply_censor_rules_to_text!(text)
    end


    # Remove quoted sections from emails (eventually the aim would be for this
    # to do as good a job as GMail does) XXX bet it needs a proper parser
    # XXX and this FOLDED_QUOTED_SECTION stuff is a mess
    def self.remove_quoted_sections(text, replacement = "FOLDED_QUOTED_SECTION")
        text = text.dup
        replacement = "\n" + replacement + "\n"

        # First do this peculiar form of quoting, as the > single line quoting
        # further below messes with it. Note the carriage return where it wraps -
        # this can happen anywhere according to length of the name/email. e.g.
        # >>> D K Elwell <[email address]> 17/03/2008
        # 01:51:50 >>>
        # http://www.whatdotheyknow.com/request/71/response/108
        # http://www.whatdotheyknow.com/request/police_powers_to_inform_car_insu
        # http://www.whatdotheyknow.com/request/secured_convictions_aided_by_cct
        multiline_original_message = '(' + '''>>>.* \d\d/\d\d/\d\d\d\d\s+\d\d:\d\d(?::\d\d)?\s*>>>''' + ')'
        text.gsub!(/^(#{multiline_original_message}\n.*)$/ms, replacement)

        # Single line sections
        text.gsub!(/^(>.*\n)/, replacement)
        text.gsub!(/^(On .+ (wrote|said):\n)/, replacement)

        ['-', '_', '*', '#'].each do |scorechar|
            score = /(?:[#{scorechar}]\s*){8,}/
            text.sub!(/(Disclaimer\s+)?  # appears just before
                        (
                            \s*#{score}\n(?:(?!#{score}\n).)*? # top line
                            (disclaimer:\n|confidential|received\sthis\semail\sin\serror|virus|intended\s+recipient|monitored\s+centrally|intended\s+(for\s+|only\s+for\s+use\s+by\s+)the\s+addressee|routinely\s+monitored|MessageLabs|unauthorised\s+use)
                            .*?(?:#{score}|\z) # bottom line OR end of whole string (for ones with no terminator XXX risky)
                        )
                       /imx, replacement)
        end

        # Special paragraphs
        # http://www.whatdotheyknow.com/request/identity_card_scheme_expenditure
        text.gsub!(/^[^\n]+Government\s+Secure\s+Intranet\s+virus\s+scanning
                    .*?
                    virus\sfree\.
                    /imx, replacement)
        text.gsub!(/^Communications\s+via\s+the\s+GSi\s+
                    .*?
                    legal\spurposes\.
                    /imx, replacement)
        # http://www.whatdotheyknow.com/request/net_promoter_value_scores_for_bb
        text.gsub!(/^http:\/\/www.bbc.co.uk
                    .*?
                    Further\s+communication\s+will\s+signify\s+your\s+consent\s+to\s+this\.
                    /imx, replacement)


        # To end of message sections
        # http://www.whatdotheyknow.com/request/123/response/192
        # http://www.whatdotheyknow.com/request/235/response/513
        # http://www.whatdotheyknow.com/request/445/response/743
        original_message =
            '(' + '''----* This is a copy of the message, including all the headers. ----*''' +
            '|' + '''----*\s*Original Message\s*----*''' +
            '|' + '''----*\s*Forwarded message.+----*''' +
            '|' + '''----*\s*Forwarded by.+----*''' +
            ')'
        # Could have a ^ at start here, but see messed up formatting here:
        # http://www.whatdotheyknow.com/request/refuse_and_recycling_collection#incoming-842
        text.gsub!(/(#{original_message}\n.*)$/mi, replacement)


        # Some silly Microsoft XML gets into parts marked as plain text.
        # e.g. http://www.whatdotheyknow.com/request/are_traffic_wardens_paid_commiss#incoming-401
        # Don't replace with "replacement" as it's pretty messy
        text.gsub!(/<\?xml:namespace[^>]*\/>/, " ")

        return text
    end

    # Internal function
    def _get_part_file_name(mail)
        part_file_name = TMail::Mail.get_part_file_name(mail)
        if part_file_name.nil?
            return nil
        end
        part_file_name = part_file_name.dup
        return part_file_name
    end

    # (This risks losing info if the unchosen alternative is the only one to contain
    # useful info, but let's worry about that another time)
    def get_attachment_leaves
        force = true
        return _get_attachment_leaves_recursive(self.mail(force))
    end
    def _get_attachment_leaves_recursive(curr_mail, within_rfc822_attachment = nil)
        leaves_found = []
        if curr_mail.multipart?
            if curr_mail.parts.size == 0
                raise "no parts on multipart mail"
            end

            if curr_mail.sub_type == 'alternative'
                # Choose best part from alternatives
                best_part = nil
                # Take the last text/plain one, or else the first one
                curr_mail.parts.each do |m|
                    if not best_part
                        best_part = m
                    elsif m.content_type == 'text/plain'
                        best_part = m
                    end
                end
                # Take an HTML one as even higher priority. (They tend
                # to render better than text/plain, e.g. don't wrap links here:
                # http://www.whatdotheyknow.com/request/amount_and_cost_of_freedom_of_in#incoming-72238 )
                curr_mail.parts.each do |m|
                    if m.content_type == 'text/html'
                        best_part = m
                    end
                end
                leaves_found += _get_attachment_leaves_recursive(best_part, within_rfc822_attachment)
            else
                # Add all parts
                curr_mail.parts.each do |m|
                    leaves_found += _get_attachment_leaves_recursive(m, within_rfc822_attachment)
                end
            end
        else
            # XXX Yuck. this section alters various content_type's. That puts
            # it into conflict with ensure_parts_counted which it has to be
            # called both before and after.  It will fail with cases of
            # attachments of attachments etc.
            charset = curr_mail.charset # save this, because overwriting content_type also resets charset
            # Don't allow nil content_types
            if curr_mail.content_type.nil?
                curr_mail.content_type = 'application/octet-stream'
            end
            # PDFs often come with this mime type, fix it up for view code
            if curr_mail.content_type == 'application/octet-stream'
                part_file_name = self._get_part_file_name(curr_mail)
                calc_mime = AlaveteliFileTypes.filename_and_content_to_mimetype(part_file_name, curr_mail.body)
                if calc_mime
                    curr_mail.content_type = calc_mime
                end
            end

            # Use standard content types for Word documents etc.
            curr_mail.content_type = normalise_content_type(curr_mail.content_type)
            if curr_mail.content_type == 'message/rfc822'
                ensure_parts_counted # fills in rfc822_attachment variable
                if curr_mail.rfc822_attachment.nil?
                    # Attached mail didn't parse, so treat as text
                    curr_mail.content_type = 'text/plain'
                end
            end
            if curr_mail.content_type == 'application/vnd.ms-outlook' || curr_mail.content_type == 'application/ms-tnef'
                ensure_parts_counted # fills in rfc822_attachment variable
                if curr_mail.rfc822_attachment.nil?
                    # Attached mail didn't parse, so treat as binary
                    curr_mail.content_type = 'application/octet-stream'
                end
            end
            # If the part is an attachment of email
            if curr_mail.content_type == 'message/rfc822' || curr_mail.content_type == 'application/vnd.ms-outlook' || curr_mail.content_type == 'application/ms-tnef'
                ensure_parts_counted # fills in rfc822_attachment variable
                leaves_found += _get_attachment_leaves_recursive(curr_mail.rfc822_attachment, curr_mail.rfc822_attachment)
            else
                # Store leaf
                curr_mail.within_rfc822_attachment = within_rfc822_attachment
                leaves_found += [curr_mail]
            end
            # restore original charset
            curr_mail.charset = charset
        end
        return leaves_found
    end

    # Removes anything cached about the object in the database, and saves
    def clear_in_database_caches!
        self.cached_attachment_text_clipped = nil
        self.cached_main_body_text_unfolded = nil
        self.cached_main_body_text_folded = nil
        self.save!
    end

    # Internal function to cache two sorts of main body text.
    # Cached as loading raw_email can be quite huge, and need this for just
    # search results
    def _cache_main_body_text
        text = self.get_main_body_text_internal
        # Strip the uudecode parts from main text
        # - this also effectively does a .dup as well, so text mods don't alter original
        text = text.split(/^begin.+^`\n^end\n/sm).join(" ")

        if text.size > 1000000 # 1 MB ish
            raise "main body text more than 1 MB, need to implement clipping like for attachment text, or there is some other MIME decoding problem or similar"
        end

        # remove emails for privacy/anti-spam reasons
        self.mask_special_emails!(text)
        self.remove_privacy_sensitive_things!(text)

        # Remove existing quoted sections
        folded_quoted_text = self.remove_lotus_quoting(text, 'FOLDED_QUOTED_SECTION')
        folded_quoted_text = IncomingMessage.remove_quoted_sections(text, "FOLDED_QUOTED_SECTION")
        self.cached_main_body_text_unfolded = text
        self.cached_main_body_text_folded = folded_quoted_text
        self.save!
    end
    # Returns body text from main text part of email, converted to UTF-8, with uudecode removed,
    # emails and privacy sensitive things remove, censored, and folded to remove excess quoted text
    # (marked with FOLDED_QUOTED_SECTION)
    # XXX returns a .dup of the text, so calling functions can in place modify it
    def get_main_body_text_folded
        if self.cached_main_body_text_folded.nil?
            self._cache_main_body_text
        end
        return self.cached_main_body_text_folded
    end
    def get_main_body_text_unfolded
        if self.cached_main_body_text_unfolded.nil?
            self._cache_main_body_text
        end
        return self.cached_main_body_text_unfolded
    end
    # Returns body text from main text part of email, converted to UTF-8
    def get_main_body_text_internal
        parse_raw_email!
        main_part = get_main_body_text_part
        return _convert_part_body_to_text(main_part)
    end

    # Given a main text part, converts it to text
    def _convert_part_body_to_text(part)
        if part.nil?
            text = "[ Email has no body, please see attachments ]"
            source_charset = "utf-8"
        else
            text = part.body # by default, TMail converts to UTF8 in this call
            source_charset = part.charset
            if part.content_type == 'text/html'
                # e.g. http://www.whatdotheyknow.com/request/35/response/177
                # XXX This is a bit of a hack as it is calling a
                # convert to text routine.  Could instead call a
                # sanitize HTML one.

                # If the text isn't UTF8, it means TMail had a problem
                # converting it (invalid characters, etc), and we
                # should instead tell elinks to respect the source
                # charset
                use_charset = "utf-8"
                begin
                    text = Iconv.conv('utf-8', 'utf-8', text)
                rescue Iconv::IllegalSequence
                    use_charset = source_charset
                end
                text = self.class._get_attachment_text_internal_one_file(part.content_type, text, use_charset)
            end
        end

        # If TMail can't convert text, it just returns it, so we sanitise it.
        begin
            # Test if it's good UTF-8
            text = Iconv.conv('utf-8', 'utf-8', text)
        rescue Iconv::IllegalSequence
            # Text looks like unlabelled nonsense,
            # strip out anything that isn't UTF-8
            begin
                text = Iconv.conv('utf-8//IGNORE', source_charset, text) +
                    _("\n\n[ {{site_name}} note: The above text was badly encoded, and has had strange characters removed. ]",
                      :site_name => MySociety::Config.get('SITE_NAME', 'Alaveteli'))
            rescue Iconv::InvalidEncoding, Iconv::IllegalSequence
                if source_charset != "utf-8"
                    source_charset = "utf-8"
                    retry
                end
            end
        end

        # Fix DOS style linefeeds to Unix style ones (or other later regexps won't work)
        # Needed for e.g. http://www.whatdotheyknow.com/request/60/response/98
        text = text.gsub(/\r\n/, "\n")

        # Compress extra spaces down to save space, and to stop regular expressions
        # breaking in strange extreme cases. e.g. for
        # http://www.whatdotheyknow.com/request/spending_on_consultants
        text = text.gsub(/ +/, " ")

        return text
    end
    # Returns part which contains main body text, or nil if there isn't one
    def get_main_body_text_part
        leaves = self.foi_attachments

        # Find first part which is text/plain or text/html
        # (We have to include HTML, as increasingly there are mail clients that
        # include no text alternative for the main part, and we don't want to
        # instead use the first text attachment
        # e.g. http://www.whatdotheyknow.com/request/list_of_public_authorties)
        leaves.each do |p|
            if p.content_type == 'text/plain' or p.content_type == 'text/html'
                return p
            end
        end

        # Otherwise first part which is any sort of text
        leaves.each do |p|
            if p.content_type.match(/^text/)
                return p
            end
        end

        # ... or if none, consider first part
        p = leaves[0]
        # if it is a known type then don't use it, return no body (nil)
        if !p.nil? && AlaveteliFileTypes.mimetype_to_extension(p.content_type)
            # this is guess of case where there are only attachments, no body text
            # e.g. http://www.whatdotheyknow.com/request/cost_benefit_analysis_for_real_n
            return nil
        end
        # otherwise return it assuming it is text (sometimes you get things
        # like binary/octet-stream, or the like, which are really text - XXX if
        # you find an example, put URL here - perhaps we should be always returning
        # nil in this case)
        return p
    end
    # Returns attachments that are uuencoded in main body part
    def _uudecode_and_save_attachments(text)
        # Find any uudecoded things buried in it, yeuchly
        uus = text.scan(/^begin.+^`\n^end\n/sm)
        attachments = []
        for uu in uus
            # Decode the string
            content = nil
            tempfile = Tempfile.new('foiuu')
            tempfile.print uu
            tempfile.flush
            content = AlaveteliExternalCommand.run("uudecode", "-o", "/dev/stdout", tempfile.path)
            tempfile.close
            # Make attachment type from it, working out filename and mime type
            filename = uu.match(/^begin\s+[0-9]+\s+(.*)$/)[1]
            calc_mime = AlaveteliFileTypes.filename_and_content_to_mimetype(filename, content)
            if calc_mime
                calc_mime = normalise_content_type(calc_mime)
                content_type = calc_mime
            else
                content_type = 'application/octet-stream'
            end
            hexdigest = Digest::MD5.hexdigest(content)
            attachment = self.foi_attachments.find_or_create_by_hexdigest(:hexdigest => hexdigest)
            attachment.update_attributes(:filename => filename,
                                         :content_type => content_type,
                                         :body => content,
                                         :display_size => "0K")
            attachment.save!
            attachments << attachment
        end
        return attachments
    end

    def get_attachments_for_display
        parse_raw_email!
        # return what user would consider attachments, i.e. not the main body
        main_part = get_main_body_text_part
        attachments = []
        for attachment in self.foi_attachments
            attachments << attachment if attachment != main_part
        end
        return attachments
    end

    def extract_attachments!
        leaves = get_attachment_leaves # XXX check where else this is called from
        # XXX we have to call ensure_parts_counted after get_attachment_leaves
        # which is really messy.
        ensure_parts_counted
        attachments = []
        for leaf in leaves
            body = leaf.body
            # As leaf.body causes MIME decoding which uses lots of RAM, do garbage collection here
            # to prevent excess memory use. XXX not really sure if this helps reduce
            # peak RAM use overall. Anyway, maybe there is something better to do than this.
            GC.start
            if leaf.within_rfc822_attachment
                within_rfc822_subject = leaf.within_rfc822_attachment.subject
                # Test to see if we are in the first part of the attached
                # RFC822 message and it is text, if so add headers.
                # XXX should probably use hunting algorithm to find main text part, rather than
                # just expect it to be first. This will do for now though.
                # Example request that needs this:
                # http://www.whatdotheyknow.com/request/2923/response/7013/attach/2/Cycle%20Path%20Bank.txt
                if leaf.within_rfc822_attachment == leaf && leaf.content_type == 'text/plain'
                    headers = ""
                    for header in [ 'Date', 'Subject', 'From', 'To', 'Cc' ]
                        if leaf.within_rfc822_attachment.header.include?(header.downcase)
                            header_value = leaf.within_rfc822_attachment.header[header.downcase]
                            # Example message which has a blank Date header:
                            # http://www.whatdotheyknow.com/request/30747/response/80253/attach/html/17/Common%20Purpose%20Advisory%20Group%20Meeting%20Tuesday%202nd%20March.txt.html
                            if !header_value.blank?
                                headers = headers + header + ": " + header_value.to_s + "\n"
                            end
                        end
                    end
                    # XXX call _convert_part_body_to_text here, but need to get charset somehow
                    # e.g. http://www.whatdotheyknow.com/request/1593/response/3088/attach/4/Freedom%20of%20Information%20request%20-%20car%20oval%20sticker:%20Article%2020,%20Convention%20on%20Road%20Traffic%201949.txt
                    body = headers + "\n" + body

                    # This is quick way of getting all headers, but instead we only add some a) to
                    # make it more usable, b) as at least one authority accidentally leaked security
                    # information into a header.
                    #attachment.body = leaf.within_rfc822_attachment.port.to_s
                end
            end
            hexdigest = Digest::MD5.hexdigest(body)
            attachment = self.foi_attachments.find_or_create_by_hexdigest(:hexdigest => hexdigest)
            attachment.update_attributes(:url_part_number => leaf.url_part_number,
                                         :content_type => leaf.content_type,
                                         :filename => _get_part_file_name(leaf),
                                         :charset => leaf.charset,
                                         :within_rfc822_subject => within_rfc822_subject,
                                         :body => body)
            attachment.save!
            attachments << attachment.id
        end
        main_part = get_main_body_text_part
        # we don't use get_main_body_text_internal, as we want to avoid charset
        # conversions, since /usr/bin/uudecode needs to deal with those.
        # e.g. for https://secure.mysociety.org/admin/foi/request/show_raw_email/24550
        if !main_part.nil?
            uudecoded_attachments = _uudecode_and_save_attachments(main_part.body)
            c = @count_first_uudecode_count
            for uudecode_attachment in uudecoded_attachments
                c += 1
                uudecode_attachment.url_part_number = c
                uudecode_attachment.save!
                attachments << uudecode_attachment.id
            end
        end

        # now get rid of any attachments we no longer have
        FoiAttachment.destroy_all("id NOT IN (#{attachments.join(',')}) AND incoming_message_id = #{self.id}")
   end

    # Returns body text as HTML with quotes flattened, and emails removed.
    def get_body_for_html_display(collapse_quoted_sections = true)
        # Find the body text and remove emails for privacy/anti-spam reasons
        text = get_main_body_text_unfolded
        folded_quoted_text = get_main_body_text_folded

        # Remove quoted sections, adding HTML. XXX The FOLDED_QUOTED_SECTION is
        # a nasty hack so we can escape other HTML before adding the unfold
        # links, without escaping them. Rather than using some proper parser
        # making a tree structure (I don't know of one that is to hand, that
        # works well in this kind of situation, such as with regexps).
        if collapse_quoted_sections
            text = folded_quoted_text
        end
        text = MySociety::Format.simplify_angle_bracketed_urls(text)
        text = CGI.escapeHTML(text)
        text = MySociety::Format.make_clickable(text, :contract => 1)
        text.gsub!(/\[(email address|mobile number)\]/, '[<a href="/help/officers#mobiles">\1</a>]')
        if collapse_quoted_sections
            text = text.gsub(/(\s*FOLDED_QUOTED_SECTION\s*)+/m, "FOLDED_QUOTED_SECTION")
            text.strip!
            # if there is nothing but quoted stuff, then show the subject
            if text == "FOLDED_QUOTED_SECTION"
                text = "[Subject only] " + CGI.escapeHTML(self.subject) + text
            end
            # and display link for quoted stuff
            text = text.gsub(/FOLDED_QUOTED_SECTION/, "\n\n" + '<span class="unfold_link"><a href="?unfold=1#incoming-'+self.id.to_s+'">'+_("show quoted sections")+'</a></span>' + "\n\n")
        else
            if folded_quoted_text.include?('FOLDED_QUOTED_SECTION')
                text = text + "\n\n" + '<span class="unfold_link"><a href="?#incoming-'+self.id.to_s+'">'+_("hide quoted sections")+'</a></span>'
            end
        end
        text.strip!

        text = text.gsub(/\n/, '<br>')
        text = text.gsub(/(?:<br>\s*){2,}/, '<br><br>') # remove excess linebreaks that unnecessarily space it out
        return text
    end


    # Returns text of email for using in quoted section when replying
    def get_body_for_quoting
        # Get the body text with emails and quoted sections removed
        text = get_main_body_text_folded
        text.gsub!("FOLDED_QUOTED_SECTION", " ")
        text.strip!
        raise "internal error" if text.nil?
        return text
    end

    MAX_ATTACHMENT_TEXT_CLIPPED = 1000000 # 1Mb ish

    # Returns text version of attachment text
    def get_attachment_text_full
        text = self._get_attachment_text_internal
        self.mask_special_emails!(text)
        self.remove_privacy_sensitive_things!(text)
        # This can be useful for memory debugging
        #STDOUT.puts 'xxx '+ MySociety::DebugHelpers::allocated_string_size_around_gc

        # Save clipped version for snippets
        if self.cached_attachment_text_clipped.nil?
            self.cached_attachment_text_clipped = text[0..MAX_ATTACHMENT_TEXT_CLIPPED]
            self.save!
        end

        return text
    end
    # Returns a version reduced to a sensible maximum size - this
    # is for performance reasons when showing snippets in search results.
    def get_attachment_text_clipped
        if self.cached_attachment_text_clipped.nil?
            # As side effect, get_attachment_text_full makes snippet text
            attachment_text = self.get_attachment_text_full
            raise "internal error" if self.cached_attachment_text_clipped.nil?
        end

        return self.cached_attachment_text_clipped
    end
    def IncomingMessage._get_attachment_text_internal_one_file(content_type, body, charset = 'utf-8')
        # note re. charset: TMail always tries to convert email bodies
        # to UTF8 by default, so normally it should already be that.
        text = ''
        # XXX - tell all these command line tools to return utf-8
        if content_type == 'text/plain'
            text += body + "\n\n"
        else
            tempfile = Tempfile.new('foiextract')
            tempfile.print body
            tempfile.flush
            if content_type == 'application/vnd.ms-word'
                AlaveteliExternalCommand.run("wvText", tempfile.path, tempfile.path + ".txt")
                # Try catdoc if we get into trouble (e.g. for InfoRequestEvent 2701)
                if not File.exists?(tempfile.path + ".txt")
                    AlaveteliExternalCommand.run("catdoc", tempfile.path, :append_to => text)
                else
                    text += File.read(tempfile.path + ".txt") + "\n\n"
                    File.unlink(tempfile.path + ".txt")
                end
            elsif content_type == 'application/rtf'
                # catdoc on RTF prodcues less comments and extra bumf than --text option to unrtf
                AlaveteliExternalCommand.run("catdoc", tempfile.path, :append_to => text)
            elsif content_type == 'text/html'
                # lynx wordwraps links in its output, which then don't
                # get formatted properly by Alaveteli. We use elinks
                # instead, which doesn't do that.
                AlaveteliExternalCommand.run("elinks", "-eval", "set document.codepage.assume = \"#{charset}\"", "-eval", "set document.codepage.force_assumed = 1", "-dump-charset", "utf-8", "-force-html", "-dump",
                    tempfile.path, :append_to => text, :env => {"LANG" => "C"})
            elsif content_type == 'application/vnd.ms-excel'
                # Bit crazy using /usr/bin/strings - but xls2csv, xlhtml and
                # py_xls2txt only extract text from cells, not from floating
                # notes. catdoc may be fooled by weird character sets, but will
                # probably do for UK FOI requests.
                AlaveteliExternalCommand.run("/usr/bin/strings", tempfile.path, :append_to => text)
            elsif content_type == 'application/vnd.ms-powerpoint'
                # ppthtml seems to catch more text, but only outputs HTML when
                # we want text, so just use catppt for now
                AlaveteliExternalCommand.run("catppt", tempfile.path, :append_to => text)
            elsif content_type == 'application/pdf'
                AlaveteliExternalCommand.run("pdftotext", tempfile.path, "-", :append_to => text)
            elsif content_type == 'application/vnd.openxmlformats-officedocument.wordprocessingml.document'
                # This is Microsoft's XML office document format.
                # Just pull out the main XML file, and strip it of text.
                xml = AlaveteliExternalCommand.run("/usr/bin/unzip", "-qq", "-c", tempfile.path, "word/document.xml")
                if !xml.nil?
                    doc = REXML::Document.new(xml)
                    text += doc.each_element( './/text()' ){}.join(" ")
                end
            elsif content_type == 'application/zip'
                # recurse into zip files
                begin
                    zip_file = Zip::ZipFile.open(tempfile.path)
                    text += IncomingMessage._get_attachment_text_from_zip_file(zip_file)
                    zip_file.close()
                rescue
                    $stderr.puts("Error processing zip file: #{$!.inspect}")
                end
            end
            tempfile.close
        end

        return text
    end
    def IncomingMessage._get_attachment_text_from_zip_file(zip_file)
        text = ""
        for entry in zip_file
            if entry.file?
                filename = entry.to_s
                begin
                    body = entry.get_input_stream.read
                rescue
                    # move to next attachment silently if there were problems
                    # XXX really should reduce this to specific exceptions?
                    # e.g. password protected
                    next
                end
                calc_mime = AlaveteliFileTypes.filename_to_mimetype(filename)
                if calc_mime
                    content_type = calc_mime
                else
                    content_type = 'application/octet-stream'
                end

                text += _get_attachment_text_internal_one_file(content_type, body)
            end
        end
        return text
    end
    def _get_attachment_text_internal
        # Extract text from each attachment
        text = ''
        attachments = self.get_attachments_for_display
        for attachment in attachments
            text += IncomingMessage._get_attachment_text_internal_one_file(attachment.content_type, attachment.body, attachment.charset)
        end
        # Remove any bad characters
        text = Iconv.conv('utf-8//IGNORE', 'utf-8', text)
        return text
    end


    # Returns text for indexing
    def get_text_for_indexing_full
        return get_body_for_quoting + "\n\n" + get_attachment_text_full
    end
    # Used for excerpts in search results, when loading full text would be too slow
    def get_text_for_indexing_clipped
        return get_body_for_quoting + "\n\n" + get_attachment_text_clipped
    end

    # Has message arrived "recently"?
    def recently_arrived
        (Time.now - self.created_at) <= 3.days
    end

    def fully_destroy
        ActiveRecord::Base.transaction do
            for o in self.outgoing_message_followups
                o.incoming_message_followup = nil
                o.save!
            end
            info_request_event = InfoRequestEvent.find_by_incoming_message_id(self.id)
            info_request_event.track_things_sent_emails.each { |a| a.destroy }
            info_request_event.user_info_request_sent_alerts.each { |a| a.destroy }
            info_request_event.destroy
            self.raw_email.destroy_file_representation!
            self.destroy
        end
    end

    # Search all info requests for
    def IncomingMessage.find_all_unknown_mime_types
        for incoming_message in IncomingMessage.find(:all)
            for attachment in incoming_message.get_attachments_for_display
                raise "internal error incoming_message " + incoming_message.id.to_s if attachment.content_type.nil?
                if AlaveteliFileTypes.mimetype_to_extension(attachment.content_type).nil?
                    $stderr.puts "Unknown type for /request/" + incoming_message.info_request.id.to_s + "#incoming-"+incoming_message.id.to_s
                    $stderr.puts " " + attachment.filename.to_s + " " + attachment.content_type.to_s
                end
            end
        end

        return nil
    end

    # Returns space separated list of file extensions of attachments to this message. Defaults to
    # the normal extension for known mime type, otherwise uses other extensions.
    def get_present_file_extensions
        ret = {}
        for attachment in self.get_attachments_for_display
            ext = AlaveteliFileTypes.mimetype_to_extension(attachment.content_type)
            ext = File.extname(attachment.filename).gsub(/^[.]/, "") if ext.nil? && !attachment.filename.nil?
            ret[ext] = 1 if !ext.nil?
        end
        return ret.keys.join(" ")
    end
    # Return space separated list of all file extensions known
    def IncomingMessage.get_all_file_extensions
        return AlaveteliFileTypes.all_extensions.join(" ")
    end

    # Return false if for some reason this is a message that we shouldn't let them reply to
    def valid_to_reply_to?
        # check validity of email
        if self.mail.from_addrs.nil? || self.mail.from_addrs.size == 0
            return false
        end
        email = self.mail.from_addrs[0].spec
        if !MySociety::Validate.is_valid_email(email)
            return false
        end

        # reject postmaster - authorities seem to nearly always not respond to
        # email to postmaster, and it tends to only happen after delivery failure.
        # likewise Mailer-Daemon, Auto_Reply...
        prefix = email
        prefix =~ /^(.*)@/
        prefix = $1
        if !prefix.nil? && prefix.downcase.match(/^(postmaster|mailer-daemon|auto_reply|do.?not.?reply|no.reply)$/)
            return false
        end
        if !self.mail['return-path'].nil? && self.mail['return-path'].addr == "<>"
            return false
        end
        if !self.mail['auto-submitted'].nil?
            return false
        end
        return true
    end

    def normalise_content_type(content_type)
        # e.g. http://www.whatdotheyknow.com/request/93/response/250
        if content_type == 'application/excel' or content_type == 'application/msexcel' or content_type == 'application/x-ms-excel'
            content_type = 'application/vnd.ms-excel'
        end
        if content_type == 'application/mspowerpoint' or content_type == 'application/x-ms-powerpoint'
            content_type = 'application/vnd.ms-powerpoint'
        end
        if content_type == 'application/msword' or content_type == 'application/x-ms-word'
            content_type = 'application/vnd.ms-word'
        end
        if content_type == 'application/x-zip-compressed'
            content_type = 'application/zip'
        end

        # e.g. http://www.whatdotheyknow.com/request/copy_of_current_swessex_scr_opt#incoming-9928
        if content_type == 'application/acrobat'
            content_type = 'application/pdf'
        end

        return content_type
    end

  def for_admin_column
    self.class.content_columns.each do |column|
      yield(column.human_name, self.send(column.name), column.type.to_s, column.name)
    end
  end

  private :normalise_content_type

end