5 class FerretIndex < BaseIndex
7 def initialize dir=BASE_DIR
10 @index_mutex = Monitor.new
11 wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
12 sa = Ferret::Analysis::StandardAnalyzer.new [], true
13 @analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
15 @analyzer[:subject] = sa
16 @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
19 def load_index dir=File.join(@dir, "ferret")
21 Redwood::log "loading index..."
22 @index_mutex.synchronize do
23 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
24 Redwood::log "loaded index of #{@index.size} messages"
27 Redwood::log "creating index..."
28 @index_mutex.synchronize do
29 field_infos = Ferret::Index::FieldInfos.new :store => :yes
30 field_infos.add_field :message_id, :index => :untokenized
31 field_infos.add_field :source_id
32 field_infos.add_field :source_info
33 field_infos.add_field :date, :index => :untokenized
34 field_infos.add_field :body
35 field_infos.add_field :label
36 field_infos.add_field :attachments
37 field_infos.add_field :subject
38 field_infos.add_field :from
39 field_infos.add_field :to
40 field_infos.add_field :refs
41 field_infos.add_field :snippet, :index => :no, :term_vector => :no
42 field_infos.create_index dir
43 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
48 def sync_message m, opts={}
51 raise "no source info for message #{m.id}" unless m.source && m.source_info
53 source_id = if m.source.is_a? Integer
56 m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
59 snippet = if m.snippet_contains_encrypted_content? && $config[:discard_snippets_from_encrypted_messages]
65 ## write the new document to the index. if the entry already exists in the
66 ## index, reuse it (which avoids having to reload the entry from the source,
67 ## which can be quite expensive for e.g. large threads of IMAP actions.)
69 ## exception: if the index entry belongs to an earlier version of the
70 ## message, use everything from the new message instead, but union the
71 ## flags. this allows messages sent to mailing lists to have their header
72 ## updated and to have flags set properly.
74 ## minor hack: messages in sources with lower ids have priority over
75 ## messages in sources with higher ids. so messages in the inbox will
76 ## override everyone, and messages in the sent box will be overridden
79 ## written in this manner to support previous versions of the index which
80 ## did not keep around the entry body. upgrading is thus seamless.
82 labels = m.labels.uniq # override because this is the new state, unless...
84 ## if we are a later version of a message, ignore what's in the index,
85 ## but merge in the labels.
86 if entry[:source_id] && entry[:source_info] && entry[:label] &&
87 ((entry[:source_id].to_i > source_id) || (entry[:source_info].to_i < m.source_info))
88 labels = (entry[:label].symbolistize + m.labels).uniq
89 #Redwood::log "found updated version of message #{m.id}: #{m.subj}"
90 #Redwood::log "previous version was at #{entry[:source_id].inspect}:#{entry[:source_info].inspect}, this version at #{source_id.inspect}:#{m.source_info.inspect}"
91 #Redwood::log "merged labels are #{labels.inspect} (index #{entry[:label].inspect}, message #{m.labels.inspect})"
95 ## if force_overwite is true, ignore what's in the index. this is used
96 ## primarily by sup-sync to force index updates.
97 entry = {} if opts[:force_overwrite]
101 :source_id => source_id,
102 :source_info => m.source_info,
103 :date => (entry[:date] || m.date.to_indexable_s),
104 :body => (entry[:body] || m.indexable_content),
105 :snippet => snippet, # always override
106 :label => labels.uniq.join(" "),
107 :attachments => (entry[:attachments] || m.attachments.uniq.join(" ")),
109 ## always override :from and :to.
110 ## older versions of Sup would often store the wrong thing in the index
111 ## (because they were canonicalizing email addresses, resulting in the
112 ## wrong name associated with each.) the correct address is read from
113 ## the original header when these messages are opened in thread-view-mode,
114 ## so this allows people to forcibly update the address in the index by
115 ## marking those threads for saving.
116 :from => (m.from ? m.from.indexable_content : ""),
117 :to => (m.to + m.cc + m.bcc).map { |x| x.indexable_content }.join(" "),
119 :subject => (entry[:subject] || wrap_subj(Message.normalize_subj(m.subj))),
120 :refs => (entry[:refs] || (m.refs + m.replytos).uniq.join(" ")),
123 @index_mutex.synchronize do
125 @index.add_document d
129 def save_index fn=File.join(@dir, "ferret")
130 # don't have to do anything, apparently
134 @index_mutex.synchronize { @index.search(Ferret::Search::TermQuery.new(:message_id, id)).total_hits > 0 }
138 @index_mutex.synchronize { @index.size }
141 EACH_BY_DATE_NUM = 100
142 def each_id_by_date query={}
143 return if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
144 ferret_query = build_ferret_query query
147 limit = (query[:limit])? [EACH_BY_DATE_NUM, query[:limit] - offset].min : EACH_BY_DATE_NUM
148 results = @index_mutex.synchronize { @index.search ferret_query, :sort => "date DESC", :limit => limit, :offset => offset }
149 Redwood::log "got #{results.total_hits} results for query (offset #{offset}) #{ferret_query.inspect}"
150 results.hits.each do |hit|
151 yield @index_mutex.synchronize { @index[hit.doc][:message_id] }, lambda { build_message hit.doc }
153 break if query[:limit] and offset >= query[:limit] - limit
154 break if offset >= results.total_hits - limit
159 def num_results_for query={}
160 return 0 if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
161 ferret_query = build_ferret_query query
162 @index_mutex.synchronize { @index.search(ferret_query, :limit => 1).total_hits }
165 SAME_SUBJECT_DATE_LIMIT = 7
167 def each_message_in_thread_for m, opts={}
168 #Redwood::log "Building thread for #{m.id}: #{m.subj}"
174 if $config[:thread_by_subject] # do subject queries
175 date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
176 date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
178 q = Ferret::Search::BooleanQuery.new true
179 sq = Ferret::Search::PhraseQuery.new(:subject)
180 wrap_subj(Message.normalize_subj(m.subj)).split.each do |t|
183 q.add_query sq, :must
184 q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
186 q = build_ferret_query :qobj => q
188 p1 = @index_mutex.synchronize { @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] } }
189 Redwood::log "found #{p1.size} results for subject query #{q}"
191 p2 = @index_mutex.synchronize { @index.search(q.to_s, :limit => :all).hits.map { |hit| @index[hit.doc][:message_id] } }
192 Redwood::log "found #{p2.size} results in string form"
194 pending = (pending + p1 + p2).uniq
197 until pending.empty? || (opts[:limit] && messages.size >= opts[:limit])
198 q = Ferret::Search::BooleanQuery.new true
199 # this disappeared in newer ferrets... wtf.
200 # q.max_clause_count = 2048
202 lim = [MAX_CLAUSES / 2, pending.length].min
203 pending[0 ... lim].each do |id|
205 q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
206 q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
208 pending = pending[lim .. -1]
210 q = build_ferret_query :qobj => q
214 @index_mutex.synchronize do
215 @index.search_each(q, :limit => :all) do |docid, score|
216 break if opts[:limit] && messages.size >= opts[:limit]
217 if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
221 mid = @index[docid][:message_id]
222 unless messages.member?(mid)
223 #Redwood::log "got #{mid} as a child of #{id}"
224 messages[mid] ||= lambda { build_message docid }
225 refs = @index[docid][:refs].split
226 pending += refs.select { |id| !searched[id] }
233 Redwood::log "thread for #{m.id} is killed, ignoring"
236 Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}: #{m.subj}" if num_queries > 0
237 messages.each { |mid, builder| yield mid, builder }
242 ## builds a message object from a ferret result
243 def build_message docid
244 @index_mutex.synchronize do
245 doc = @index[docid] or return
247 source = SourceManager[doc[:source_id].to_i]
248 raise "invalid source #{doc[:source_id]}" unless source
250 #puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
253 "date" => Time.at(doc[:date].to_i),
254 "subject" => unwrap_subj(doc[:subject]),
255 "from" => doc[:from],
256 "to" => doc[:to].split.join(", "), # reformat
257 "message-id" => doc[:message_id],
258 "references" => doc[:refs].split.map { |x| "<#{x}>" }.join(" "),
261 m = Message.new :source => source, :source_info => doc[:source_info].to_i,
262 :labels => doc[:label].symbolistize,
263 :snippet => doc[:snippet]
264 m.parse_header fake_header
270 @index_mutex.synchronize { @index.delete id }
273 def load_contacts emails, h={}
274 q = Ferret::Search::BooleanQuery.new true
276 qq = Ferret::Search::BooleanQuery.new true
277 qq.add_query Ferret::Search::TermQuery.new(:from, e), :should
278 qq.add_query Ferret::Search::TermQuery.new(:to, e), :should
281 q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
283 Redwood::log "contact search: #{q}"
286 @index_mutex.synchronize do
287 @index.search_each q, :sort => "date DESC", :limit => :all do |docid, score|
288 break if contacts.size >= num
289 #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
290 f = @index[docid][:from]
291 t = @index[docid][:to]
293 if AccountManager.is_account_email? f
294 t.split(" ").each { |e| contacts[Person.from_address(e)] = true }
296 contacts[Person.from_address(f)] = true
301 contacts.keys.compact
305 ferret_query = build_ferret_query query
306 results = @index_mutex.synchronize { @index.search ferret_query, :limit => (query[:limit] || :all) }
307 results.hits.map { |hit| yield @index[hit.doc][:message_id] }
311 @index_mutex.synchronize { @index.optimize }
317 entry[:source_id].to_i
320 class ParseError < StandardError; end
322 ## parse a query string from the user. returns a query object
323 ## that can be passed to any index method with a 'query'
324 ## argument, as well as build_ferret_query.
326 ## raises a ParseError if something went wrong.
330 subs = s.gsub(/\b(to|from):(\S+)\b/) do
332 if(p = ContactManager.contact_for(name))
335 [field, "(" + AccountManager.user_emails.join("||") + ")"]
341 ## if we see a label:deleted or a label:spam term anywhere in the query
342 ## string, we set the extra load_spam or load_deleted options to true.
343 ## bizarre? well, because the query allows arbitrary parenthesized boolean
344 ## expressions, without fully parsing the query, we can't tell whether
345 ## the user is explicitly directing us to search spam messages or not.
346 ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
347 ## search spam messages or not?
349 ## so, we rely on the fact that turning these extra options ON turns OFF
350 ## the adding of "-label:deleted" or "-label:spam" terms at the very
351 ## final stage of query processing. if the user wants to search spam
352 ## messages, not adding that is the right thing; if he doesn't want to
353 ## search spam messages, then not adding it won't have any effect.
354 query[:load_spam] = true if subs =~ /\blabel:spam\b/
355 query[:load_deleted] = true if subs =~ /\blabel:deleted\b/
357 ## gmail style "is" operator
358 subs = subs.gsub(/\b(is|has):(\S+)\b/) do
359 field, label = $1, $2
364 query[:load_spam] = true
367 query[:load_deleted] = true
374 ## gmail style attachments "filename" and "filetype" searches
375 subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
376 field, name = $1, ($3 || $4)
379 Redwood::log "filename - translated #{field}:#{name} to attachments:(#{name.downcase})"
380 "attachments:(#{name.downcase})"
382 Redwood::log "filetype - translated #{field}:#{name} to attachments:(*.#{name.downcase})"
383 "attachments:(*.#{name.downcase})"
388 subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
389 field, datestr = $1, ($3 || $4)
390 realdate = Chronic.parse datestr, :guess => false, :context => :past
394 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
395 "date:(>= #{sprintf "%012d", realdate.end.to_i})"
397 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
398 "date:(<= #{sprintf "%012d", realdate.begin.to_i})"
400 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
401 "date:(<= #{sprintf "%012d", realdate.end.to_i}) date:(>= #{sprintf "%012d", realdate.begin.to_i})"
404 raise ParseError, "can't understand date #{datestr.inspect}"
409 ## limit:42 restrict the search to 42 results
410 subs = subs.gsub(/\blimit:(\S+)\b/) do
413 query[:limit] = lim.to_i
416 raise ParseError, "non-numeric limit #{lim.inspect}"
421 query[:qobj] = @qparser.parse(subs)
424 rescue Ferret::QueryParser::QueryParseException => e
425 raise ParseError, e.message
431 def build_ferret_query query
432 q = Ferret::Search::BooleanQuery.new
433 q.add_query Ferret::Search::MatchAllQuery.new, :must
434 q.add_query query[:qobj], :must if query[:qobj]
435 labels = ([query[:label]] + (query[:labels] || [])).compact
436 labels.each { |t| q.add_query Ferret::Search::TermQuery.new("label", t.to_s), :must }
437 if query[:participants]
438 q2 = Ferret::Search::BooleanQuery.new
439 query[:participants].each do |p|
440 q2.add_query Ferret::Search::TermQuery.new("from", p.email), :should
441 q2.add_query Ferret::Search::TermQuery.new("to", p.email), :should
443 q.add_query q2, :must
446 q.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless query[:load_spam] || labels.include?(:spam)
447 q.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless query[:load_deleted] || labels.include?(:deleted)
448 q.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if query[:skip_killed]
450 q.add_query Ferret::Search::TermQuery.new("source_id", query[:source_id]), :must if query[:source_id]
454 def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
455 def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end