require 'net/imap'
require 'stringio'
require 'time'
+require 'rmail'
+require 'cgi'
+require 'set'
-## fucking imap fucking sucks. what the FUCK kind of committee of
-## dunces designed this shit.
+## TODO: remove synchronized method protector calls; use a Monitor instead
+## (ruby's reentrant mutex)
+## fucking imap fucking sucks. what the FUCK kind of committee of dunces
+## designed this shit.
+##
## imap talks about 'unique ids' for messages, to be used for
-## cross-session identification. great---just what sup needs! except
-## it turns out the uids can be invalidated every time the
-## 'uidvalidity' value changes on the server, and 'uidvalidity' can
-## change without restriction. it can change any time you log in. it
-## can change EVERY time you log in. of course the imap spec "strongly
-## recommends" that it never change, but there's nothing to stop
-## people from just setting it to the current timestamp, and in fact
-## that's exactly what the one imap server i have at my disposal
-## does. thus the so-called uids are absolutely useless and imap
-## provides no cross-session way of uniquely identifying a
-## message. but thanks for the "strong recommendation", guys!
-
+## cross-session identification. great---just what sup needs! except it
+## turns out the uids can be invalidated every time the 'uidvalidity'
+## value changes on the server, and 'uidvalidity' can change without
+## restriction. it can change any time you log in. it can change EVERY
+## time you log in. of course the imap spec "strongly recommends" that it
+## never change, but there's nothing to stop people from just setting it
+## to the current timestamp, and in fact that's EXACTLY what the one imap
+## server i have at my disposal does. thus the so-called uids are
+## absolutely useless and imap provides no cross-session way of uniquely
+## identifying a message. but thanks for the "strong recommendation",
+## guys!
+##
## so right now i'm using the 'internal date' and the size of each
## message to uniquely identify it, and i scan over the entire mailbox
## each time i open it to map those things to message ids. that can be
-## slow for large mailboxes, and we'll just have to hope that there
-## are no collisions. ho ho! a perfectly reasonable solution!
-
-## fuck you, imap committee. you managed to design something as shitty
-## as mbox but goddamn THIRTY YEARS LATER.
-
+## slow for large mailboxes, and we'll just have to hope that there are
+## no collisions. ho ho! a perfectly reasonable solution!
+##
+## and here's another thing. check out RFC2060 2.2.2 paragraph 5:
+##
+## A client MUST be prepared to accept any server response at all
+## times. This includes server data that was not requested.
+##
+## yeah. that totally makes a lot of sense. and once again, the idiocy of
+## the spec actually happens in practice. you'll request flags for one
+## message, and get it interspersed with a random bunch of flags for some
+## other messages, including a different set of flags for the same
+## message! totally ok by the imap spec. totally retarded by any other
+## metric.
+##
+## fuck you, imap committee. you managed to design something nearly as
+## shitty as mbox but goddamn THIRTY YEARS LATER.
module Redwood
class IMAP < Source
- attr_reader_cloned :labels
+ include SerializeLabelsNicely
+ SCAN_INTERVAL = 60 # seconds
+
+ ## upon these errors we'll try to rereconnect a few times
+ RECOVERABLE_ERRORS = [ Errno::EPIPE, Errno::ETIMEDOUT, OpenSSL::SSL::SSLError ]
+
attr_accessor :username, :password
+ yaml_properties :uri, :username, :password, :cur_offset, :usual,
+ :archived, :id, :labels
- def initialize uri, username, password, last_idate=nil, usual=true, archived=false, id=nil
+ def initialize uri, username, password, last_idate=nil, usual=true, archived=false, id=nil, labels=[]
raise ArgumentError, "username and password must be specified" unless username && password
raise ArgumentError, "not an imap uri" unless uri =~ %r!imaps?://!
@username = username
@password = password
@imap = nil
- @imap_ids = {}
+ @imap_state = {}
@ids = []
- @labels = [:unread]
- @labels << :inbox unless archived?
- @labels << mailbox.intern unless mailbox =~ /inbox/i
+ @last_scan = nil
+ @labels = Set.new((labels || []) - LabelManager::RESERVED_LABELS)
+ @say_id = nil
@mutex = Mutex.new
+ end
- @recover_msg = "It is likely that messages have been deleted from this IMAP mailbox. Please run sup-import --rebuild #{to_s} to correct this problem."
+ def self.suggest_labels_for path
+ path =~ /([^\/]*inbox[^\/]*)/i ? [$1.downcase.intern] : []
end
- def say s
- @say_id = BufferManager.say s, @say_id if BufferManager.instantiated?
- Redwood::log s
+ def host; @parsed_uri.host; end
+ def port; @parsed_uri.port || (ssl? ? 993 : 143); end
+ def mailbox
+ x = @parsed_uri.path[1..-1]
+ (x.nil? || x.empty?) ? 'INBOX' : CGI.unescape(x)
end
- def shutup
- BufferManager.clear @say_id if BufferManager.instantiated?
- @say_id = nil
+ def ssl?; @parsed_uri.scheme == 'imaps' end
+
+ def check; end # do nothing because anything we do will be too slow,
+ # and we'll catch the errors later.
+
+ ## is this necessary? TODO: remove maybe
+ def == o; o.is_a?(IMAP) && o.uri == self.uri && o.username == self.username; end
+
+ def load_header id
+ parse_raw_email_header StringIO.new(raw_header(id))
+ end
+
+ def load_message id
+ RMail::Parser.read raw_message(id)
+ end
+
+ def each_raw_message_line id
+ StringIO.new(raw_message(id)).each { |l| yield l }
+ end
+
+ def raw_header id
+ unsynchronized_scan_mailbox
+ header, flags = get_imap_fields id, 'RFC822.HEADER'
+ header.gsub(/\r\n/, "\n")
+ end
+ synchronized :raw_header
+
+ def store_message date, from_email, &block
+ message = StringIO.new
+ yield message
+ message.string.gsub! /\n/, "\r\n"
+
+ safely { @imap.append mailbox, message.string, [:Seen], Time.now }
+ end
+
+ def raw_message id
+ unsynchronized_scan_mailbox
+ get_imap_fields(id, 'RFC822').first.gsub(/\r\n/, "\n")
+ end
+ synchronized :raw_message
+
+ def mark_as_deleted ids
+ ids = [ids].flatten # accept single arguments
+ unsynchronized_scan_mailbox
+ imap_ids = ids.map { |i| @imap_state[i] && @imap_state[i][:id] }.compact
+ return if imap_ids.empty?
+ @imap.store imap_ids, "+FLAGS", [:Deleted]
+ end
+ synchronized :mark_as_deleted
+
+ def expunge
+ @imap.expunge
+ unsynchronized_scan_mailbox true
+ true
end
- private :say, :shutup
+ synchronized :expunge
def connect
- return false if broken?
- return true if @imap
+ return if @imap
+ safely { } # do nothing!
+ end
+ synchronized :connect
- ## ok, this is FUCKING ANNOYING.
- ##
- ## what imap.rb likes to do is, if an exception occurs, catch it
- ## and re-raise it on the calling thread. seems reasonable. but
- ## what that REALLY means is that the only way to reasonably
- ## initialize imap is in its own thread, because otherwise, you
- ## will never be able to catch the exception it raises on the
- ## calling thread, and the backtrace will not make any sense at
- ## all, and you will waste HOURS of your life on this fucking
- ## problem.
- ##
- ## FUCK!!!!!!!!!
+ def scan_mailbox force=false
+ return if !force && @last_scan && (Time.now - @last_scan) < SCAN_INTERVAL
+ last_id = safely do
+ @imap.examine mailbox
+ @imap.responses["EXISTS"].last
+ end
+ @last_scan = Time.now
+
+ @ids = [] if force
+ return if last_id == @ids.length
+
+ range = (@ids.length + 1) .. last_id
+ Redwood::log "fetching IMAP headers #{range}"
+ fetch(range, ['RFC822.SIZE', 'INTERNALDATE', 'FLAGS']).each do |v|
+ id = make_id v
+ @ids << id
+ @imap_state[id] = { :id => v.seqno, :flags => v.attr["FLAGS"] }
+ end
+ Redwood::log "done fetching IMAP headers"
+ end
+ synchronized :scan_mailbox
+ def each
+ return unless start_offset
+
+ ids =
+ @mutex.synchronize do
+ unsynchronized_scan_mailbox
+ @ids
+ end
+
+ start = ids.index(cur_offset || start_offset) or raise OutOfSyncSourceError, "Unknown message id #{cur_offset || start_offset}."
+
+ start.upto(ids.length - 1) do |i|
+ id = ids[i]
+ state = @mutex.synchronize { @imap_state[id] } or next
+ self.cur_offset = id
+ labels = { :Flagged => :starred,
+ :Deleted => :deleted
+ }.inject(@labels) do |cur, (imap, sup)|
+ cur + (state[:flags].include?(imap) ? [sup] : [])
+ end
+
+ labels += [:unread] unless state[:flags].include?(:Seen)
+
+ yield id, labels
+ end
+ end
+
+ def start_offset
+ unsynchronized_scan_mailbox
+ @ids.first
+ end
+ synchronized :start_offset
+
+ def end_offset
+ unsynchronized_scan_mailbox
+ @ids.last + 1
+ end
+ synchronized :end_offset
+
+ def pct_done; 100.0 * (@ids.index(cur_offset) || 0).to_f / (@ids.length - 1).to_f; end
+
+private
+
+ def fetch ids, fields
+ results = safely { @imap.fetch ids, fields }
+ good_results =
+ if ids.respond_to? :member?
+ results.find_all { |r| ids.member?(r.seqno) && fields.all? { |f| r.attr.member?(f) } }
+ else
+ results.find_all { |r| ids == r.seqno && fields.all? { |f| r.attr.member?(f) } }
+ end
+
+ if good_results.empty?
+ raise FatalSourceError, "no IMAP response for #{ids} containing all fields #{fields.join(', ')} (got #{results.size} results)"
+ elsif good_results.size < results.size
+ Redwood::log "Your IMAP server sucks. It sent #{results.size} results for a request for #{good_results.size} messages. What are you using, Binc?"
+ end
+
+ good_results
+ end
+
+ def unsafe_connect
say "Connecting to IMAP server #{host}:#{port}..."
- Redwood::reporting_thread do
+ ## apparently imap.rb does a lot of threaded stuff internally and if
+ ## an exception occurs, it will catch it and re-raise it on the
+ ## calling thread. but i can't seem to catch that exception, so i've
+ ## resorted to initializing it in its own thread. surely there's a
+ ## better way.
+ exception = nil
+ ::Thread.new do
begin
#raise Net::IMAP::ByeResponseError, "simulated imap failure"
@imap = Net::IMAP.new host, port, ssl?
say "Logging in..."
- @imap.authenticate 'LOGIN', @username, @password
- say "Sizing mailbox..."
- @imap.examine mailbox
- last_id = @imap.responses["EXISTS"][-1]
-
- say "Reading headers (because IMAP sucks)..."
- values = @imap.fetch(1 .. last_id, ['RFC822.SIZE', 'INTERNALDATE'])
- say "Successfully connected to #{@parsed_uri}."
-
- values.each do |v|
- id = make_id v
- @ids << id
- @imap_ids[id] = v.seqno
+
+ ## although RFC1730 claims that "If an AUTHENTICATE command fails
+ ## with a NO response, the client may try another", in practice
+ ## it seems like they can also send a BAD response.
+ begin
+ raise Net::IMAP::NoResponseError unless @imap.capability().member? "AUTH=CRAM-MD5"
+ @imap.authenticate 'CRAM-MD5', @username, @password
+ rescue Net::IMAP::BadResponseError, Net::IMAP::NoResponseError => e
+ Redwood::log "CRAM-MD5 authentication failed: #{e.class}. Trying LOGIN auth..."
+ begin
+ raise Net::IMAP::NoResponseError unless @imap.capability().member? "AUTH=LOGIN"
+ @imap.authenticate 'LOGIN', @username, @password
+ rescue Net::IMAP::BadResponseError, Net::IMAP::NoResponseError => e
+ Redwood::log "LOGIN authentication failed: #{e.class}. Trying plain-text LOGIN..."
+ @imap.login @username, @password
+ end
end
- rescue SocketError, Net::IMAP::Error, SourceError => e
- self.broken_msg = e.message.chomp # fucking chomp! fuck!!!
- @imap = nil
- msg = "error connecting to IMAP server: #{self.broken_msg}"
- Redwood::log msg
- BufferManager.flash msg
+ say "Successfully connected to #{@parsed_uri}."
+ rescue Exception => e
+ exception = e
ensure
shutup
end
end.join
- !!@imap
+ raise exception if exception
+ end
+
+ def say s
+ @say_id = BufferManager.say s, @say_id if BufferManager.instantiated?
+ Redwood::log s
+ end
+
+ def shutup
+ BufferManager.clear @say_id if BufferManager.instantiated?
+ @say_id = nil
end
- private :connect
- ## build a fake unique id
def make_id imap_stuff
# use 7 digits for the size. why 7? seems nice.
+ %w(RFC822.SIZE INTERNALDATE).each do |w|
+ raise FatalSourceError, "requested data not in IMAP response: #{w}" unless imap_stuff.attr[w]
+ end
+
msize, mdate = imap_stuff.attr['RFC822.SIZE'] % 10000000, Time.parse(imap_stuff.attr["INTERNALDATE"])
sprintf("%d%07d", mdate.to_i, msize).to_i
end
- private :make_id
- def host; @parsed_uri.host; end
- def port; @parsed_uri.port || (ssl? ? 993 : 143); end
- def mailbox
- x = @parsed_uri.path[1..-1]
- x.nil? || x.empty? ? 'INBOX' : x
- end
- def ssl?; @parsed_uri.scheme == 'imaps' end
+ def get_imap_fields id, *fields
+ raise OutOfSyncSourceError, "Unknown message id #{id}" unless @imap_state[id]
- def load_header id
- MBox::read_header StringIO.new(raw_header(id))
- end
+ imap_id = @imap_state[id][:id]
+ result = fetch(imap_id, (fields + ['RFC822.SIZE', 'INTERNALDATE']).uniq).first
+ got_id = make_id result
- def load_message id
- RMail::Parser.read raw_full_message(id)
- end
+ ## I've turned off the following sanity check because Microsoft
+ ## Exchange fails it. Exchange actually reports two different
+ ## INTERNALDATEs for the exact same message when queried at different
+ ## points in time.
+ ##
+ ## RFC2060 defines the semantics of INTERNALDATE for messages that
+ ## arrive via SMTP for via various IMAP commands, but states that
+ ## "All other cases are implementation defined.". Great, thanks guys,
+ ## yet another useless field.
+ ##
+ ## Of course no OTHER imap server I've encountered returns DIFFERENT
+ ## values for the SAME message. But it's Microsoft; what do you
+ ## expect? If their programmers were any good they'd be working at
+ ## Google.
- def raw_header id
- @mutex.synchronize do
- connect or raise SourceError, broken_msg
- header, flags = get_imap_fields id, 'RFC822.HEADER', 'FLAGS'
- header = "Status: RO\n" + header if flags.include? :Seen # fake an mbox-style read header
- header.gsub(/\r\n/, "\n")
- end
- end
+ # raise OutOfSyncSourceError, "IMAP message mismatch: requested #{id}, got #{got_id}." unless got_id == id
- def raw_full_message id
- @mutex.synchronize do
- connect or raise SourceError, broken_msg
- get_imap_fields(id, 'RFC822').first.gsub(/\r\n/, "\n")
- end
+ fields.map { |f| result.attr[f] or raise FatalSourceError, "empty response from IMAP server: #{f}" }
end
- def get_imap_fields id, *fields
+ ## execute a block, connected if unconnected, re-connected up to 3
+ ## times if a recoverable error occurs, and properly dying if an
+ ## unrecoverable error occurs.
+ def safely
retries = 0
- f = nil
- imap_id = @imap_ids[id] or raise SourceError, "Unknown message id #{id}. #@recover_msg"
begin
- f = @imap.fetch imap_id, (fields + ['RFC822.SIZE', 'INTERNALDATE']).uniq
- got_id = make_id f[0]
- raise SourceError, "IMAP message mismatch: requested #{id}, got #{got_id}. #@recover_msg" unless got_id == id
- rescue Net::IMAP::Error => e
- raise SourceError, e.message
- rescue Errno::EPIPE
- if (retries += 1) <= 3
- @imap = nil
- connect
- retry
+ begin
+ unsafe_connect unless @imap
+ yield
+ rescue *RECOVERABLE_ERRORS => e
+ if (retries += 1) <= 3
+ @imap = nil
+ Redwood::log "got #{e.class.name}: #{e.message.inspect}"
+ sleep 2
+ retry
+ end
+ raise
end
- end
- raise SourceError, "null IMAP field '#{field}' for message with id #{id} imap id #{imap_id}" if f.nil?
-
- fields.map { |field| f[0].attr[field] }
- end
- private :get_imap_fields
-
- def each
- @mutex.synchronize { connect or raise SourceError, broken_msg }
-
- start = @ids.index(cur_offset || start_offset)
- if start.nil? # couldn't find the most recent email
- self.broken_msg = "Unknown message id #{cur_offset || start_offset}. #@recover_msg"
- raise SourceError, broken_msg
- end
- start.upto(@ids.length - 1) do |i|
- id = @ids[i]
- self.cur_offset = id
- yield id, labels
+ rescue SocketError, Net::IMAP::Error, SystemCallError, IOError, OpenSSL::SSL::SSLError => e
+ raise FatalSourceError, "While communicating with IMAP server (type #{e.class.name}): #{e.message.inspect}"
end
end
- def start_offset
- @mutex.synchronize { connect or raise SourceError, broken_msg }
- @ids.first
- end
-
- def end_offset
- @mutex.synchronize { connect or raise SourceError, broken_msg }
- @ids.last
- end
end
-Redwood::register_yaml(IMAP, %w(uri username password cur_offset usual archived id))
-
end
projects / sup / blobdiff