1 ## the index structure for redwood. interacts with ferret.
10 Redwood::log "optional 'chronic' library not found (run 'gem install chronic' to install)"
17 class LockError < StandardError
22 def method_missing m; @h[m.to_s] end
27 HookManager.register "custom-search", <<EOS
28 Executes before a string search is applied to the index,
29 returning a new search string.
31 subs: The string being searched.
34 ## these two accessors should ONLY be used by single-threaded programs.
35 ## otherwise you will have a naughty ferret on your hands.
39 def initialize dir=BASE_DIR
40 @index_mutex = Monitor.new
44 @sources_dirty = false
45 @source_mutex = Monitor.new
47 wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
48 sa = Ferret::Analysis::StandardAnalyzer.new [], true
49 @analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
51 @analyzer[:subject] = sa
52 @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
53 @lock = Lockfile.new lockfile, :retries => 0, :max_age => nil
55 self.class.i_am_the_instance self
58 def lockfile; File.join @dir, "lock" end
61 Redwood::log "locking #{lockfile}..."
64 rescue Lockfile::MaxTriesLockError
65 raise LockError, @lock.lockinfo_on_disk
69 def start_lock_update_thread
70 @lock_update_thread = Redwood::reporting_thread("lock update") do
78 def stop_lock_update_thread
79 @lock_update_thread.kill if @lock_update_thread
80 @lock_update_thread = nil
83 def possibly_pluralize number_of, kind
84 "#{number_of} #{kind}" +
85 if number_of == 1 then "" else "s" end
88 def fancy_lock_error_message_for e
89 secs = (Time.now - e.mtime).to_i
93 possibly_pluralize secs , "second"
95 possibly_pluralize mins, "minute"
99 Error: the sup index is locked by another process! User '#{e.user}' on
100 host '#{e.host}' is running #{e.pname} with pid #{e.pid}. The process was alive
108 rescue LockError => e
109 $stderr.puts fancy_lock_error_message_for(e)
112 You can wait for the process to finish, or, if it crashed and left a
113 stale lock file behind, you can manually delete #{@lock.path}.
120 if @lock && @lock.locked?
121 Redwood::log "unlocking #{lockfile}..."
132 Redwood::log "saving index and sources..."
133 FileUtils.mkdir_p @dir unless File.exists? @dir
138 def add_source source
139 @source_mutex.synchronize do
140 raise "duplicate source!" if @sources.include? source
141 @sources_dirty = true
142 max = @sources.max_of { |id, s| s.is_a?(DraftLoader) || s.is_a?(SentLoader) ? 0 : id }
143 source.id ||= (max || 0) + 1
144 ##source.id += 1 while @sources.member? source.id
145 @sources[source.id] = source
150 ## favour the inbox by listing non-archived sources first
151 @source_mutex.synchronize { @sources.values }.sort_by { |s| s.id }.partition { |s| !s.archived? }.flatten
154 def source_for uri; sources.find { |s| s.is_source_for? uri }; end
155 def usual_sources; sources.find_all { |s| s.usual? }; end
157 def load_index dir=File.join(@dir, "ferret")
159 Redwood::log "loading index..."
160 @index_mutex.synchronize do
161 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer)
162 Redwood::log "loaded index of #{@index.size} messages"
165 Redwood::log "creating index..."
166 @index_mutex.synchronize do
167 field_infos = Ferret::Index::FieldInfos.new :store => :yes
168 field_infos.add_field :message_id, :index => :untokenized
169 field_infos.add_field :source_id
170 field_infos.add_field :source_info
171 field_infos.add_field :date, :index => :untokenized
172 field_infos.add_field :body
173 field_infos.add_field :label
174 field_infos.add_field :attachments
175 field_infos.add_field :subject
176 field_infos.add_field :from
177 field_infos.add_field :to
178 field_infos.add_field :refs
179 field_infos.add_field :snippet, :index => :no, :term_vector => :no
180 field_infos.create_index dir
181 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer)
186 ## Syncs the message to the index, replacing any previous version. adding
187 ## either way. Index state will be determined by the message's #labels
190 ## if need_load is false, docid and entry are assumed to be set to the
191 ## result of load_entry_for_id (which can be nil).
192 def sync_message m, need_load=true, docid=nil, entry=nil, opts={}
193 docid, entry = load_entry_for_id m.id if need_load
195 raise "no source info for message #{m.id}" unless m.source && m.source_info
196 @index_mutex.synchronize do
197 raise "trying to delete non-corresponding entry #{docid} with index message-id #{@index[docid][:message_id].inspect} and parameter message id #{m.id.inspect}" if docid && @index[docid][:message_id] != m.id
200 source_id = if m.source.is_a? Integer
203 m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
206 snippet = if m.snippet_contains_encrypted_content? && $config[:discard_snippets_from_encrypted_messages]
212 ## write the new document to the index. if the entry already exists in the
213 ## index, reuse it (which avoids having to reload the entry from the source,
214 ## which can be quite expensive for e.g. large threads of IMAP actions.)
216 ## exception: if the index entry belongs to an earlier version of the
217 ## message, use everything from the new message instead, but union the
218 ## flags. this allows messages sent to mailing lists to have their header
219 ## updated and to have flags set properly.
221 ## minor hack: messages in sources with lower ids have priority over
222 ## messages in sources with higher ids. so messages in the inbox will
223 ## override everyone, and messages in the sent box will be overridden
226 ## written in this manner to support previous versions of the index which
227 ## did not keep around the entry body. upgrading is thus seamless.
229 labels = m.labels.uniq # override because this is the new state, unless...
231 ## if we are a later version of a message, ignore what's in the index,
232 ## but merge in the labels.
233 if entry[:source_id] && entry[:source_info] && entry[:label] &&
234 ((entry[:source_id].to_i > source_id) || (entry[:source_info].to_i < m.source_info))
235 labels = (entry[:label].symbolistize + m.labels).uniq
236 #Redwood::log "found updated version of message #{m.id}: #{m.subj}"
237 #Redwood::log "previous version was at #{entry[:source_id].inspect}:#{entry[:source_info].inspect}, this version at #{source_id.inspect}:#{m.source_info.inspect}"
238 #Redwood::log "merged labels are #{labels.inspect} (index #{entry[:label].inspect}, message #{m.labels.inspect})"
242 ## if force_overwite is true, ignore what's in the index. this is used
243 ## primarily by sup-sync to force index updates.
244 entry = {} if opts[:force_overwrite]
248 :source_id => source_id,
249 :source_info => m.source_info,
250 :date => (entry[:date] || m.date.to_indexable_s),
251 :body => (entry[:body] || m.indexable_content),
252 :snippet => snippet, # always override
253 :label => labels.uniq.join(" "),
254 :attachments => (entry[:attachments] || m.attachments.uniq.join(" ")),
256 ## always override :from and :to.
257 ## older versions of Sup would often store the wrong thing in the index
258 ## (because they were canonicalizing email addresses, resulting in the
259 ## wrong name associated with each.) the correct address is read from
260 ## the original header when these messages are opened in thread-view-mode,
261 ## so this allows people to forcibly update the address in the index by
262 ## marking those threads for saving.
263 :from => (m.from ? m.from.indexable_content : ""),
264 :to => (m.to + m.cc + m.bcc).map { |x| x.indexable_content }.join(" "),
266 :subject => (entry[:subject] || wrap_subj(Message.normalize_subj(m.subj))),
267 :refs => (entry[:refs] || (m.refs + m.replytos).uniq.join(" ")),
270 @index_mutex.synchronize do
271 @index.delete docid if docid
272 @index.add_document d
275 ## this hasn't been triggered in a long time.
276 ## docid, entry = load_entry_for_id m.id
277 ## raise "just added message #{m.id.inspect} but couldn't find it in a search" unless docid
280 def save_index fn=File.join(@dir, "ferret")
281 # don't have to do anything, apparently
285 @index_mutex.synchronize { @index.search(Ferret::Search::TermQuery.new(:message_id, id)).total_hits > 0 }
287 def contains? m; contains_id? m.id end
288 def size; @index_mutex.synchronize { @index.size } end
289 def empty?; size == 0 end
291 ## you should probably not call this on a block that doesn't break
292 ## rather quickly because the results can be very large.
293 EACH_BY_DATE_NUM = 100
294 def each_id_by_date opts={}
295 return if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
296 query = build_query opts
299 limit = (opts[:limit])? [EACH_BY_DATE_NUM, opts[:limit] - offset].min : EACH_BY_DATE_NUM
300 results = @index_mutex.synchronize { @index.search query, :sort => "date DESC", :limit => limit, :offset => offset }
301 Redwood::log "got #{results.total_hits} results for query (offset #{offset}) #{query.inspect}"
302 results.hits.each do |hit|
303 yield @index_mutex.synchronize { @index[hit.doc][:message_id] }, lambda { build_message hit.doc }
305 break if opts[:limit] and offset >= opts[:limit] - limit
306 break if offset >= results.total_hits - limit
311 def num_results_for opts={}
312 return 0 if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
315 @index_mutex.synchronize { @index.search(q, :limit => 1).total_hits }
318 ## yield all messages in the thread containing 'm' by repeatedly
319 ## querying the index. yields pairs of message ids and
320 ## message-building lambdas, so that building an unwanted message
321 ## can be skipped in the block if desired.
323 ## only two options, :limit and :skip_killed. if :skip_killed is
324 ## true, stops loading any thread if a message with a :killed flag
326 SAME_SUBJECT_DATE_LIMIT = 7
328 def each_message_in_thread_for m, opts={}
329 #Redwood::log "Building thread for #{m.id}: #{m.subj}"
335 if $config[:thread_by_subject] # do subject queries
336 date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
337 date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
339 q = Ferret::Search::BooleanQuery.new true
340 sq = Ferret::Search::PhraseQuery.new(:subject)
341 wrap_subj(Message.normalize_subj(m.subj)).split.each do |t|
344 q.add_query sq, :must
345 q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
347 q = build_query :qobj => q
349 p1 = @index_mutex.synchronize { @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] } }
350 Redwood::log "found #{p1.size} results for subject query #{q}"
352 p2 = @index_mutex.synchronize { @index.search(q.to_s, :limit => :all).hits.map { |hit| @index[hit.doc][:message_id] } }
353 Redwood::log "found #{p2.size} results in string form"
355 pending = (pending + p1 + p2).uniq
358 until pending.empty? || (opts[:limit] && messages.size >= opts[:limit])
359 q = Ferret::Search::BooleanQuery.new true
360 # this disappeared in newer ferrets... wtf.
361 # q.max_clause_count = 2048
363 lim = [MAX_CLAUSES / 2, pending.length].min
364 pending[0 ... lim].each do |id|
366 q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
367 q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
369 pending = pending[lim .. -1]
371 q = build_query :qobj => q
375 @index_mutex.synchronize do
376 @index.search_each(q, :limit => :all) do |docid, score|
377 break if opts[:limit] && messages.size >= opts[:limit]
378 if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
382 mid = @index[docid][:message_id]
383 unless messages.member?(mid)
384 #Redwood::log "got #{mid} as a child of #{id}"
385 messages[mid] ||= lambda { build_message docid }
386 refs = @index[docid][:refs].split
387 pending += refs.select { |id| !searched[id] }
394 Redwood::log "thread for #{m.id} is killed, ignoring"
397 Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}: #{m.subj}" if num_queries > 0
398 messages.each { |mid, builder| yield mid, builder }
403 ## builds a message object from a ferret result
404 def build_message docid
405 @index_mutex.synchronize do
408 source = @source_mutex.synchronize { @sources[doc[:source_id].to_i] }
409 raise "invalid source #{doc[:source_id]}" unless source
411 #puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
414 "date" => Time.at(doc[:date].to_i),
415 "subject" => unwrap_subj(doc[:subject]),
416 "from" => doc[:from],
417 "to" => doc[:to].split.join(", "), # reformat
418 "message-id" => doc[:message_id],
419 "references" => doc[:refs].split.map { |x| "<#{x}>" }.join(" "),
422 m = Message.new :source => source, :source_info => doc[:source_info].to_i,
423 :labels => doc[:label].symbolistize,
424 :snippet => doc[:snippet]
425 m.parse_header fake_header
430 def fresh_thread_id; @next_thread_id += 1; end
431 def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
432 def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end
434 def drop_entry docno; @index_mutex.synchronize { @index.delete docno } end
436 def load_entry_for_id mid
437 @index_mutex.synchronize do
438 results = @index.search Ferret::Search::TermQuery.new(:message_id, mid)
439 return if results.total_hits == 0
440 docid = results.hits[0].doc
441 entry = @index[docid]
442 entry_dup = entry.fields.inject({}) { |h, f| h[f] = entry[f]; h }
447 def load_contacts emails, h={}
448 q = Ferret::Search::BooleanQuery.new true
450 qq = Ferret::Search::BooleanQuery.new true
451 qq.add_query Ferret::Search::TermQuery.new(:from, e), :should
452 qq.add_query Ferret::Search::TermQuery.new(:to, e), :should
455 q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
457 Redwood::log "contact search: #{q}"
460 @index_mutex.synchronize do
461 @index.search_each q, :sort => "date DESC", :limit => :all do |docid, score|
462 break if contacts.size >= num
463 #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
464 f = @index[docid][:from]
465 t = @index[docid][:to]
467 if AccountManager.is_account_email? f
468 t.split(" ").each { |e| contacts[Person.from_address(e)] = true }
470 contacts[Person.from_address(f)] = true
475 contacts.keys.compact
478 def load_sources fn=Redwood::SOURCE_FN
479 source_array = (Redwood::load_yaml_obj(fn) || []).map { |o| Recoverable.new o }
480 @source_mutex.synchronize do
481 @sources = Hash[*(source_array).map { |s| [s.id, s] }.flatten]
482 @sources_dirty = false
486 def has_any_from_source_with_label? source, label
487 q = Ferret::Search::BooleanQuery.new
488 q.add_query Ferret::Search::TermQuery.new("source_id", source.id.to_s), :must
489 q.add_query Ferret::Search::TermQuery.new("label", label.to_s), :must
490 @index_mutex.synchronize { @index.search(q, :limit => 1).total_hits > 0 }
493 ## takes a user query string and returns the list of docids for messages
494 ## that match the query.
496 ## messages can then be loaded from the index with #build_message.
498 ## raises a ParseError if the parsing failed.
500 qobj, opts = Redwood::Index.parse_user_query_string query
501 query = Redwood::Index.build_query opts.merge(:qobj => qobj)
502 results = @index.search query, :limit => (opts[:limit] || :all)
503 results.hits.map { |hit| hit.doc }
508 class ParseError < StandardError; end
510 ## parse a query string from the user. returns a query object and a set of
511 ## extra flags; both of these are meant to be passed to #build_query.
513 ## raises a ParseError if something went wrong.
514 def parse_user_query_string s
517 subs = HookManager.run("custom-search", :subs => s) || s
519 subs = subs.gsub(/\b(to|from):(\S+)\b/) do
521 if(p = ContactManager.contact_for(name))
524 [field, "(" + AccountManager.user_emails.join("||") + ")"]
530 ## if we see a label:deleted or a label:spam term anywhere in the query
531 ## string, we set the extra load_spam or load_deleted options to true.
532 ## bizarre? well, because the query allows arbitrary parenthesized boolean
533 ## expressions, without fully parsing the query, we can't tell whether
534 ## the user is explicitly directing us to search spam messages or not.
535 ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
536 ## search spam messages or not?
538 ## so, we rely on the fact that turning these extra options ON turns OFF
539 ## the adding of "-label:deleted" or "-label:spam" terms at the very
540 ## final stage of query processing. if the user wants to search spam
541 ## messages, not adding that is the right thing; if he doesn't want to
542 ## search spam messages, then not adding it won't have any effect.
543 extraopts[:load_spam] = true if subs =~ /\blabel:spam\b/
544 extraopts[:load_deleted] = true if subs =~ /\blabel:deleted\b/
546 ## gmail style "is" operator
547 subs = subs.gsub(/\b(is|has):(\S+)\b/) do
548 field, label = $1, $2
553 extraopts[:load_spam] = true
556 extraopts[:load_deleted] = true
563 ## gmail style attachments "filename" and "filetype" searches
564 subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
565 field, name = $1, ($3 || $4)
568 Redwood::log "filename - translated #{field}:#{name} to attachments:(#{name.downcase})"
569 "attachments:(#{name.downcase})"
571 Redwood::log "filetype - translated #{field}:#{name} to attachments:(*.#{name.downcase})"
572 "attachments:(*.#{name.downcase})"
577 subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
578 field, datestr = $1, ($3 || $4)
579 realdate = Chronic.parse datestr, :guess => false, :context => :past
583 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
584 "date:(>= #{sprintf "%012d", realdate.end.to_i})"
586 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
587 "date:(<= #{sprintf "%012d", realdate.begin.to_i})"
589 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
590 "date:(<= #{sprintf "%012d", realdate.end.to_i}) date:(>= #{sprintf "%012d", realdate.begin.to_i})"
593 raise ParseError, "can't understand date #{datestr.inspect}"
598 ## limit:42 restrict the search to 42 results
599 subs = subs.gsub(/\blimit:(\S+)\b/) do
602 extraopts[:limit] = lim.to_i
605 raise ParseError, "non-numeric limit #{lim.inspect}"
610 [@qparser.parse(subs), extraopts]
611 rescue Ferret::QueryParser::QueryParseException => e
612 raise ParseError, e.message
617 query = Ferret::Search::BooleanQuery.new
618 query.add_query opts[:qobj], :must if opts[:qobj]
619 labels = ([opts[:label]] + (opts[:labels] || [])).compact
620 labels.each { |t| query.add_query Ferret::Search::TermQuery.new("label", t.to_s), :must }
621 if opts[:participants]
622 q2 = Ferret::Search::BooleanQuery.new
623 opts[:participants].each do |p|
624 q2.add_query Ferret::Search::TermQuery.new("from", p.email), :should
625 q2.add_query Ferret::Search::TermQuery.new("to", p.email), :should
627 query.add_query q2, :must
630 query.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless opts[:load_spam] || labels.include?(:spam)
631 query.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless opts[:load_deleted] || labels.include?(:deleted)
632 query.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if opts[:skip_killed]
636 def save_sources fn=Redwood::SOURCE_FN
637 @source_mutex.synchronize do
638 if @sources_dirty || @sources.any? { |id, s| s.dirty? }
642 FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(fn) == 0
644 Redwood::save_yaml_obj sources.sort_by { |s| s.id.to_i }, fn, true
647 @sources_dirty = false