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 def initialize dir=BASE_DIR
28 @index_mutex = Monitor.new
31 wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
32 sa = Ferret::Analysis::StandardAnalyzer.new [], true
33 @analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
35 @analyzer[:subject] = sa
36 @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
37 @lock = Lockfile.new lockfile, :retries => 0, :max_age => nil
39 self.class.i_am_the_instance self
42 def lockfile; File.join @dir, "lock" end
45 Redwood::log "locking #{lockfile}..."
48 rescue Lockfile::MaxTriesLockError
49 raise LockError, @lock.lockinfo_on_disk
53 def start_lock_update_thread
54 @lock_update_thread = Redwood::reporting_thread("lock update") do
62 def stop_lock_update_thread
63 @lock_update_thread.kill if @lock_update_thread
64 @lock_update_thread = nil
67 def possibly_pluralize number_of, kind
68 "#{number_of} #{kind}" +
69 if number_of == 1 then "" else "s" end
72 def fancy_lock_error_message_for e
73 secs = (Time.now - e.mtime).to_i
77 possibly_pluralize secs , "second"
79 possibly_pluralize mins, "minute"
83 Error: the sup index is locked by another process! User '#{e.user}' on
84 host '#{e.host}' is running #{e.pname} with pid #{e.pid}. The process was alive
93 $stderr.puts fancy_lock_error_message_for(e)
96 You can wait for the process to finish, or, if it crashed and left a
97 stale lock file behind, you can manually delete #{@lock.path}.
104 if @lock && @lock.locked?
105 Redwood::log "unlocking #{lockfile}..."
111 SourceManager.load_sources
116 Redwood::log "saving index and sources..."
117 FileUtils.mkdir_p @dir unless File.exists? @dir
118 SourceManager.save_sources
122 def load_index dir=File.join(@dir, "ferret")
124 Redwood::log "loading index..."
125 @index_mutex.synchronize do
126 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
127 Redwood::log "loaded index of #{@index.size} messages"
130 Redwood::log "creating index..."
131 @index_mutex.synchronize do
132 field_infos = Ferret::Index::FieldInfos.new :store => :yes
133 field_infos.add_field :message_id, :index => :untokenized
134 field_infos.add_field :source_id
135 field_infos.add_field :source_info
136 field_infos.add_field :date, :index => :untokenized
137 field_infos.add_field :body
138 field_infos.add_field :label
139 field_infos.add_field :attachments
140 field_infos.add_field :subject
141 field_infos.add_field :from
142 field_infos.add_field :to
143 field_infos.add_field :refs
144 field_infos.add_field :snippet, :index => :no, :term_vector => :no
145 field_infos.create_index dir
146 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
151 ## Syncs the message to the index, replacing any previous version. adding
152 ## either way. Index state will be determined by the message's #labels
154 def sync_message m, opts={}
157 raise "no source info for message #{m.id}" unless m.source && m.source_info
159 source_id = if m.source.is_a? Integer
162 m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
165 snippet = if m.snippet_contains_encrypted_content? && $config[:discard_snippets_from_encrypted_messages]
171 ## write the new document to the index. if the entry already exists in the
172 ## index, reuse it (which avoids having to reload the entry from the source,
173 ## which can be quite expensive for e.g. large threads of IMAP actions.)
175 ## exception: if the index entry belongs to an earlier version of the
176 ## message, use everything from the new message instead, but union the
177 ## flags. this allows messages sent to mailing lists to have their header
178 ## updated and to have flags set properly.
180 ## minor hack: messages in sources with lower ids have priority over
181 ## messages in sources with higher ids. so messages in the inbox will
182 ## override everyone, and messages in the sent box will be overridden
185 ## written in this manner to support previous versions of the index which
186 ## did not keep around the entry body. upgrading is thus seamless.
188 labels = m.labels.uniq # override because this is the new state, unless...
190 ## if we are a later version of a message, ignore what's in the index,
191 ## but merge in the labels.
192 if entry[:source_id] && entry[:source_info] && entry[:label] &&
193 ((entry[:source_id].to_i > source_id) || (entry[:source_info].to_i < m.source_info))
194 labels = (entry[:label].symbolistize + m.labels).uniq
195 #Redwood::log "found updated version of message #{m.id}: #{m.subj}"
196 #Redwood::log "previous version was at #{entry[:source_id].inspect}:#{entry[:source_info].inspect}, this version at #{source_id.inspect}:#{m.source_info.inspect}"
197 #Redwood::log "merged labels are #{labels.inspect} (index #{entry[:label].inspect}, message #{m.labels.inspect})"
201 ## if force_overwite is true, ignore what's in the index. this is used
202 ## primarily by sup-sync to force index updates.
203 entry = {} if opts[:force_overwrite]
207 :source_id => source_id,
208 :source_info => m.source_info,
209 :date => (entry[:date] || m.date.to_indexable_s),
210 :body => (entry[:body] || m.indexable_content),
211 :snippet => snippet, # always override
212 :label => labels.uniq.join(" "),
213 :attachments => (entry[:attachments] || m.attachments.uniq.join(" ")),
215 ## always override :from and :to.
216 ## older versions of Sup would often store the wrong thing in the index
217 ## (because they were canonicalizing email addresses, resulting in the
218 ## wrong name associated with each.) the correct address is read from
219 ## the original header when these messages are opened in thread-view-mode,
220 ## so this allows people to forcibly update the address in the index by
221 ## marking those threads for saving.
222 :from => (m.from ? m.from.indexable_content : ""),
223 :to => (m.to + m.cc + m.bcc).map { |x| x.indexable_content }.join(" "),
225 :subject => (entry[:subject] || wrap_subj(Message.normalize_subj(m.subj))),
226 :refs => (entry[:refs] || (m.refs + m.replytos).uniq.join(" ")),
229 @index_mutex.synchronize do
231 @index.add_document d
235 def save_index fn=File.join(@dir, "ferret")
236 # don't have to do anything, apparently
240 @index_mutex.synchronize { @index.search(Ferret::Search::TermQuery.new(:message_id, id)).total_hits > 0 }
242 def contains? m; contains_id? m.id end
243 def size; @index_mutex.synchronize { @index.size } end
244 def empty?; size == 0 end
246 ## you should probably not call this on a block that doesn't break
247 ## rather quickly because the results can be very large.
248 EACH_BY_DATE_NUM = 100
249 def each_id_by_date query={}
250 return if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
251 ferret_query = build_ferret_query query
254 limit = (query[:limit])? [EACH_BY_DATE_NUM, query[:limit] - offset].min : EACH_BY_DATE_NUM
255 results = @index_mutex.synchronize { @index.search ferret_query, :sort => "date DESC", :limit => limit, :offset => offset }
256 Redwood::log "got #{results.total_hits} results for query (offset #{offset}) #{ferret_query.inspect}"
257 results.hits.each do |hit|
258 yield @index_mutex.synchronize { @index[hit.doc][:message_id] }, lambda { build_message hit.doc }
260 break if query[:limit] and offset >= query[:limit] - limit
261 break if offset >= results.total_hits - limit
266 def num_results_for query={}
267 return 0 if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
269 ferret_query = build_ferret_query query
270 @index_mutex.synchronize { @index.search(ferret_query, :limit => 1).total_hits }
273 ## yield all messages in the thread containing 'm' by repeatedly
274 ## querying the index. yields pairs of message ids and
275 ## message-building lambdas, so that building an unwanted message
276 ## can be skipped in the block if desired.
278 ## only two options, :limit and :skip_killed. if :skip_killed is
279 ## true, stops loading any thread if a message with a :killed flag
281 SAME_SUBJECT_DATE_LIMIT = 7
283 def each_message_in_thread_for m, opts={}
284 #Redwood::log "Building thread for #{m.id}: #{m.subj}"
290 if $config[:thread_by_subject] # do subject queries
291 date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
292 date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
294 q = Ferret::Search::BooleanQuery.new true
295 sq = Ferret::Search::PhraseQuery.new(:subject)
296 wrap_subj(Message.normalize_subj(m.subj)).split.each do |t|
299 q.add_query sq, :must
300 q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
302 q = build_ferret_query :qobj => q
304 p1 = @index_mutex.synchronize { @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] } }
305 Redwood::log "found #{p1.size} results for subject query #{q}"
307 p2 = @index_mutex.synchronize { @index.search(q.to_s, :limit => :all).hits.map { |hit| @index[hit.doc][:message_id] } }
308 Redwood::log "found #{p2.size} results in string form"
310 pending = (pending + p1 + p2).uniq
313 until pending.empty? || (opts[:limit] && messages.size >= opts[:limit])
314 q = Ferret::Search::BooleanQuery.new true
315 # this disappeared in newer ferrets... wtf.
316 # q.max_clause_count = 2048
318 lim = [MAX_CLAUSES / 2, pending.length].min
319 pending[0 ... lim].each do |id|
321 q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
322 q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
324 pending = pending[lim .. -1]
326 q = build_ferret_query :qobj => q
330 @index_mutex.synchronize do
331 @index.search_each(q, :limit => :all) do |docid, score|
332 break if opts[:limit] && messages.size >= opts[:limit]
333 if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
337 mid = @index[docid][:message_id]
338 unless messages.member?(mid)
339 #Redwood::log "got #{mid} as a child of #{id}"
340 messages[mid] ||= lambda { build_message docid }
341 refs = @index[docid][:refs].split
342 pending += refs.select { |id| !searched[id] }
349 Redwood::log "thread for #{m.id} is killed, ignoring"
352 Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}: #{m.subj}" if num_queries > 0
353 messages.each { |mid, builder| yield mid, builder }
358 ## builds a message object from a ferret result
359 def build_message docid
360 @index_mutex.synchronize do
361 doc = @index[docid] or return
363 source = SourceManager[doc[:source_id].to_i]
364 raise "invalid source #{doc[:source_id]}" unless source
366 #puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
369 "date" => Time.at(doc[:date].to_i),
370 "subject" => unwrap_subj(doc[:subject]),
371 "from" => doc[:from],
372 "to" => doc[:to].split.join(", "), # reformat
373 "message-id" => doc[:message_id],
374 "references" => doc[:refs].split.map { |x| "<#{x}>" }.join(" "),
377 m = Message.new :source => source, :source_info => doc[:source_info].to_i,
378 :labels => doc[:label].symbolistize,
379 :snippet => doc[:snippet]
380 m.parse_header fake_header
385 def delete id; @index_mutex.synchronize { @index.delete id } end
387 def load_contacts emails, h={}
388 q = Ferret::Search::BooleanQuery.new true
390 qq = Ferret::Search::BooleanQuery.new true
391 qq.add_query Ferret::Search::TermQuery.new(:from, e), :should
392 qq.add_query Ferret::Search::TermQuery.new(:to, e), :should
395 q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
397 Redwood::log "contact search: #{q}"
400 @index_mutex.synchronize do
401 @index.search_each q, :sort => "date DESC", :limit => :all do |docid, score|
402 break if contacts.size >= num
403 #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
404 f = @index[docid][:from]
405 t = @index[docid][:to]
407 if AccountManager.is_account_email? f
408 t.split(" ").each { |e| contacts[Person.from_address(e)] = true }
410 contacts[Person.from_address(f)] = true
415 contacts.keys.compact
418 def each_docid query={}
419 ferret_query = build_ferret_query query
420 results = @index_mutex.synchronize { @index.search ferret_query, :limit => (query[:limit] || :all) }
421 results.hits.map { |hit| yield hit.doc }
424 def each_message query={}
425 each_docid query do |docid|
426 yield build_message(docid)
431 @index_mutex.synchronize { @index.optimize }
437 entry[:source_id].to_i
440 class ParseError < StandardError; end
442 ## parse a query string from the user. returns a query object
443 ## that can be passed to any index method with a 'query'
444 ## argument, as well as build_ferret_query.
446 ## raises a ParseError if something went wrong.
450 subs = s.gsub(/\b(to|from):(\S+)\b/) do
452 if(p = ContactManager.contact_for(name))
455 [field, "(" + AccountManager.user_emails.join("||") + ")"]
461 ## if we see a label:deleted or a label:spam term anywhere in the query
462 ## string, we set the extra load_spam or load_deleted options to true.
463 ## bizarre? well, because the query allows arbitrary parenthesized boolean
464 ## expressions, without fully parsing the query, we can't tell whether
465 ## the user is explicitly directing us to search spam messages or not.
466 ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
467 ## search spam messages or not?
469 ## so, we rely on the fact that turning these extra options ON turns OFF
470 ## the adding of "-label:deleted" or "-label:spam" terms at the very
471 ## final stage of query processing. if the user wants to search spam
472 ## messages, not adding that is the right thing; if he doesn't want to
473 ## search spam messages, then not adding it won't have any effect.
474 query[:load_spam] = true if subs =~ /\blabel:spam\b/
475 query[:load_deleted] = true if subs =~ /\blabel:deleted\b/
477 ## gmail style "is" operator
478 subs = subs.gsub(/\b(is|has):(\S+)\b/) do
479 field, label = $1, $2
484 query[:load_spam] = true
487 query[:load_deleted] = true
494 ## gmail style attachments "filename" and "filetype" searches
495 subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
496 field, name = $1, ($3 || $4)
499 Redwood::log "filename - translated #{field}:#{name} to attachments:(#{name.downcase})"
500 "attachments:(#{name.downcase})"
502 Redwood::log "filetype - translated #{field}:#{name} to attachments:(*.#{name.downcase})"
503 "attachments:(*.#{name.downcase})"
508 subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
509 field, datestr = $1, ($3 || $4)
510 realdate = Chronic.parse datestr, :guess => false, :context => :past
514 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
515 "date:(>= #{sprintf "%012d", realdate.end.to_i})"
517 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
518 "date:(<= #{sprintf "%012d", realdate.begin.to_i})"
520 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
521 "date:(<= #{sprintf "%012d", realdate.end.to_i}) date:(>= #{sprintf "%012d", realdate.begin.to_i})"
524 raise ParseError, "can't understand date #{datestr.inspect}"
529 ## limit:42 restrict the search to 42 results
530 subs = subs.gsub(/\blimit:(\S+)\b/) do
533 query[:limit] = lim.to_i
536 raise ParseError, "non-numeric limit #{lim.inspect}"
541 query[:qobj] = @qparser.parse(subs)
544 rescue Ferret::QueryParser::QueryParseException => e
545 raise ParseError, e.message
551 def build_ferret_query query
552 q = Ferret::Search::BooleanQuery.new
553 q.add_query query[:qobj], :must if query[:qobj]
554 labels = ([query[:label]] + (query[:labels] || [])).compact
555 labels.each { |t| q.add_query Ferret::Search::TermQuery.new("label", t.to_s), :must }
556 if query[:participants]
557 q2 = Ferret::Search::BooleanQuery.new
558 query[:participants].each do |p|
559 q2.add_query Ferret::Search::TermQuery.new("from", p.email), :should
560 q2.add_query Ferret::Search::TermQuery.new("to", p.email), :should
562 q.add_query q2, :must
565 q.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless query[:load_spam] || labels.include?(:spam)
566 q.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless query[:load_deleted] || labels.include?(:deleted)
567 q.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if query[:skip_killed]
569 q.add_query Ferret::Search::TermQuery.new("source_id", query[:source_id]), :must if query[:source_id]
573 def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
574 def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end