5 class FerretIndex < BaseIndex
7 HookManager.register "custom-search", <<EOS
8 Executes before a string search is applied to the index,
9 returning a new search string.
11 subs: The string being searched. Be careful about shadowing:
12 this variable is actually a method, so use a temporary variable
13 or explicitly call self.subs; the substitutions in index.rb
17 def initialize dir=BASE_DIR
20 @index_mutex = Monitor.new
21 wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
22 sa = Ferret::Analysis::StandardAnalyzer.new [], true
23 @analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
25 @analyzer[:subject] = sa
26 @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
29 def load_index dir=File.join(@dir, "ferret")
31 Redwood::log "loading index..."
32 @index_mutex.synchronize do
33 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
34 Redwood::log "loaded index of #{@index.size} messages"
37 Redwood::log "creating index..."
38 @index_mutex.synchronize do
39 field_infos = Ferret::Index::FieldInfos.new :store => :yes
40 field_infos.add_field :message_id, :index => :untokenized
41 field_infos.add_field :source_id
42 field_infos.add_field :source_info
43 field_infos.add_field :date, :index => :untokenized
44 field_infos.add_field :body
45 field_infos.add_field :label
46 field_infos.add_field :attachments
47 field_infos.add_field :subject
48 field_infos.add_field :from
49 field_infos.add_field :to
50 field_infos.add_field :refs
51 field_infos.add_field :snippet, :index => :no, :term_vector => :no
52 field_infos.create_index dir
53 @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer, :id_field => 'message_id')
58 def sync_message m, opts={}
61 raise "no source info for message #{m.id}" unless m.source && m.source_info
63 source_id = if m.source.is_a? Integer
66 m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
69 snippet = if m.snippet_contains_encrypted_content? && $config[:discard_snippets_from_encrypted_messages]
75 ## write the new document to the index. if the entry already exists in the
76 ## index, reuse it (which avoids having to reload the entry from the source,
77 ## which can be quite expensive for e.g. large threads of IMAP actions.)
79 ## exception: if the index entry belongs to an earlier version of the
80 ## message, use everything from the new message instead, but union the
81 ## flags. this allows messages sent to mailing lists to have their header
82 ## updated and to have flags set properly.
84 ## minor hack: messages in sources with lower ids have priority over
85 ## messages in sources with higher ids. so messages in the inbox will
86 ## override everyone, and messages in the sent box will be overridden
89 ## written in this manner to support previous versions of the index which
90 ## did not keep around the entry body. upgrading is thus seamless.
92 labels = m.labels.uniq # override because this is the new state, unless...
94 ## if we are a later version of a message, ignore what's in the index,
95 ## but merge in the labels.
96 if entry[:source_id] && entry[:source_info] && entry[:label] &&
97 ((entry[:source_id].to_i > source_id) || (entry[:source_info].to_i < m.source_info))
98 labels = (entry[:label].symbolistize + m.labels).uniq
99 #Redwood::log "found updated version of message #{m.id}: #{m.subj}"
100 #Redwood::log "previous version was at #{entry[:source_id].inspect}:#{entry[:source_info].inspect}, this version at #{source_id.inspect}:#{m.source_info.inspect}"
101 #Redwood::log "merged labels are #{labels.inspect} (index #{entry[:label].inspect}, message #{m.labels.inspect})"
105 ## if force_overwite is true, ignore what's in the index. this is used
106 ## primarily by sup-sync to force index updates.
107 entry = {} if opts[:force_overwrite]
111 :source_id => source_id,
112 :source_info => m.source_info,
113 :date => (entry[:date] || m.date.to_indexable_s),
114 :body => (entry[:body] || m.indexable_content),
115 :snippet => snippet, # always override
116 :label => labels.uniq.join(" "),
117 :attachments => (entry[:attachments] || m.attachments.uniq.join(" ")),
119 ## always override :from and :to.
120 ## older versions of Sup would often store the wrong thing in the index
121 ## (because they were canonicalizing email addresses, resulting in the
122 ## wrong name associated with each.) the correct address is read from
123 ## the original header when these messages are opened in thread-view-mode,
124 ## so this allows people to forcibly update the address in the index by
125 ## marking those threads for saving.
126 :from => (m.from ? m.from.indexable_content : ""),
127 :to => (m.to + m.cc + m.bcc).map { |x| x.indexable_content }.join(" "),
129 :subject => (entry[:subject] || wrap_subj(Message.normalize_subj(m.subj))),
130 :refs => (entry[:refs] || (m.refs + m.replytos).uniq.join(" ")),
133 @index_mutex.synchronize do
135 @index.add_document d
139 def save_index fn=File.join(@dir, "ferret")
140 # don't have to do anything, apparently
144 @index_mutex.synchronize { @index.search(Ferret::Search::TermQuery.new(:message_id, id)).total_hits > 0 }
148 @index_mutex.synchronize { @index.size }
151 EACH_BY_DATE_NUM = 100
152 def each_id_by_date query={}
153 return if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
154 ferret_query = build_ferret_query query
157 limit = (query[:limit])? [EACH_BY_DATE_NUM, query[:limit] - offset].min : EACH_BY_DATE_NUM
158 results = @index_mutex.synchronize { @index.search ferret_query, :sort => "date DESC", :limit => limit, :offset => offset }
159 Redwood::log "got #{results.total_hits} results for query (offset #{offset}) #{ferret_query.inspect}"
160 results.hits.each do |hit|
161 yield @index_mutex.synchronize { @index[hit.doc][:message_id] }, lambda { build_message hit.doc }
163 break if query[:limit] and offset >= query[:limit] - limit
164 break if offset >= results.total_hits - limit
169 def num_results_for query={}
170 return 0 if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
171 ferret_query = build_ferret_query query
172 @index_mutex.synchronize { @index.search(ferret_query, :limit => 1).total_hits }
175 SAME_SUBJECT_DATE_LIMIT = 7
177 def each_message_in_thread_for m, opts={}
178 #Redwood::log "Building thread for #{m.id}: #{m.subj}"
184 if $config[:thread_by_subject] # do subject queries
185 date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
186 date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
188 q = Ferret::Search::BooleanQuery.new true
189 sq = Ferret::Search::PhraseQuery.new(:subject)
190 wrap_subj(Message.normalize_subj(m.subj)).split.each do |t|
193 q.add_query sq, :must
194 q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
196 q = build_ferret_query :qobj => q
198 p1 = @index_mutex.synchronize { @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] } }
199 Redwood::log "found #{p1.size} results for subject query #{q}"
201 p2 = @index_mutex.synchronize { @index.search(q.to_s, :limit => :all).hits.map { |hit| @index[hit.doc][:message_id] } }
202 Redwood::log "found #{p2.size} results in string form"
204 pending = (pending + p1 + p2).uniq
207 until pending.empty? || (opts[:limit] && messages.size >= opts[:limit])
208 q = Ferret::Search::BooleanQuery.new true
209 # this disappeared in newer ferrets... wtf.
210 # q.max_clause_count = 2048
212 lim = [MAX_CLAUSES / 2, pending.length].min
213 pending[0 ... lim].each do |id|
215 q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
216 q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
218 pending = pending[lim .. -1]
220 q = build_ferret_query :qobj => q
224 @index_mutex.synchronize do
225 @index.search_each(q, :limit => :all) do |docid, score|
226 break if opts[:limit] && messages.size >= opts[:limit]
227 if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
231 mid = @index[docid][:message_id]
232 unless messages.member?(mid)
233 #Redwood::log "got #{mid} as a child of #{id}"
234 messages[mid] ||= lambda { build_message docid }
235 refs = @index[docid][:refs].split
236 pending += refs.select { |id| !searched[id] }
243 #Redwood::log "thread for #{m.id} is killed, ignoring"
246 #Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}: #{m.subj}" if num_queries > 0
247 messages.each { |mid, builder| yield mid, builder }
252 ## builds a message object from a ferret result
253 def build_message docid
254 @index_mutex.synchronize do
255 doc = @index[docid] or return
257 source = SourceManager[doc[:source_id].to_i]
258 raise "invalid source #{doc[:source_id]}" unless source
260 #puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
263 "date" => Time.at(doc[:date].to_i),
264 "subject" => unwrap_subj(doc[:subject]),
265 "from" => doc[:from],
266 "to" => doc[:to].split.join(", "), # reformat
267 "message-id" => doc[:message_id],
268 "references" => doc[:refs].split.map { |x| "<#{x}>" }.join(" "),
271 m = Message.new :source => source, :source_info => doc[:source_info].to_i,
272 :labels => doc[:label].symbolistize,
273 :snippet => doc[:snippet]
274 m.parse_header fake_header
280 @index_mutex.synchronize { @index.delete id }
283 def load_contacts emails, h={}
284 q = Ferret::Search::BooleanQuery.new true
286 qq = Ferret::Search::BooleanQuery.new true
287 qq.add_query Ferret::Search::TermQuery.new(:from, e), :should
288 qq.add_query Ferret::Search::TermQuery.new(:to, e), :should
291 q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
293 Redwood::log "contact search: #{q}"
296 @index_mutex.synchronize do
297 @index.search_each q, :sort => "date DESC", :limit => :all do |docid, score|
298 break if contacts.size >= num
299 #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
300 f = @index[docid][:from]
301 t = @index[docid][:to]
303 if AccountManager.is_account_email? f
304 t.split(" ").each { |e| contacts[Person.from_address(e)] = true }
306 contacts[Person.from_address(f)] = true
311 contacts.keys.compact
315 ferret_query = build_ferret_query query
316 results = @index_mutex.synchronize { @index.search ferret_query, :limit => (query[:limit] || :all) }
317 results.hits.map { |hit| yield @index[hit.doc][:message_id] }
321 @index_mutex.synchronize { @index.optimize }
327 entry[:source_id].to_i
330 class ParseError < StandardError; end
332 ## parse a query string from the user. returns a query object
333 ## that can be passed to any index method with a 'query'
334 ## argument, as well as build_ferret_query.
336 ## raises a ParseError if something went wrong.
340 subs = HookManager.run("custom-search", :subs => s) || s
342 subs = subs.gsub(/\b(to|from):(\S+)\b/) do
344 if(p = ContactManager.contact_for(name))
347 [field, "(" + AccountManager.user_emails.join("||") + ")"]
353 ## if we see a label:deleted or a label:spam term anywhere in the query
354 ## string, we set the extra load_spam or load_deleted options to true.
355 ## bizarre? well, because the query allows arbitrary parenthesized boolean
356 ## expressions, without fully parsing the query, we can't tell whether
357 ## the user is explicitly directing us to search spam messages or not.
358 ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
359 ## search spam messages or not?
361 ## so, we rely on the fact that turning these extra options ON turns OFF
362 ## the adding of "-label:deleted" or "-label:spam" terms at the very
363 ## final stage of query processing. if the user wants to search spam
364 ## messages, not adding that is the right thing; if he doesn't want to
365 ## search spam messages, then not adding it won't have any effect.
366 query[:load_spam] = true if subs =~ /\blabel:spam\b/
367 query[:load_deleted] = true if subs =~ /\blabel:deleted\b/
369 ## gmail style "is" operator
370 subs = subs.gsub(/\b(is|has):(\S+)\b/) do
371 field, label = $1, $2
376 query[:load_spam] = true
379 query[:load_deleted] = true
386 ## gmail style attachments "filename" and "filetype" searches
387 subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
388 field, name = $1, ($3 || $4)
391 Redwood::log "filename - translated #{field}:#{name} to attachments:(#{name.downcase})"
392 "attachments:(#{name.downcase})"
394 Redwood::log "filetype - translated #{field}:#{name} to attachments:(*.#{name.downcase})"
395 "attachments:(*.#{name.downcase})"
400 subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
401 field, datestr = $1, ($3 || $4)
402 realdate = Chronic.parse datestr, :guess => false, :context => :past
406 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
407 "date:(>= #{sprintf "%012d", realdate.end.to_i})"
409 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
410 "date:(<= #{sprintf "%012d", realdate.begin.to_i})"
412 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
413 "date:(<= #{sprintf "%012d", realdate.end.to_i}) date:(>= #{sprintf "%012d", realdate.begin.to_i})"
416 raise ParseError, "can't understand date #{datestr.inspect}"
421 ## limit:42 restrict the search to 42 results
422 subs = subs.gsub(/\blimit:(\S+)\b/) do
425 query[:limit] = lim.to_i
428 raise ParseError, "non-numeric limit #{lim.inspect}"
433 query[:qobj] = @qparser.parse(subs)
436 rescue Ferret::QueryParser::QueryParseException => e
437 raise ParseError, e.message
443 def build_ferret_query query
444 q = Ferret::Search::BooleanQuery.new
445 q.add_query Ferret::Search::MatchAllQuery.new, :must
446 q.add_query query[:qobj], :must if query[:qobj]
447 labels = ([query[:label]] + (query[:labels] || [])).compact
448 labels.each { |t| q.add_query Ferret::Search::TermQuery.new("label", t.to_s), :must }
449 if query[:participants]
450 q2 = Ferret::Search::BooleanQuery.new
451 query[:participants].each do |p|
452 q2.add_query Ferret::Search::TermQuery.new("from", p.email), :should
453 q2.add_query Ferret::Search::TermQuery.new("to", p.email), :should
455 q.add_query q2, :must
458 q.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless query[:load_spam] || labels.include?(:spam)
459 q.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless query[:load_deleted] || labels.include?(:deleted)
460 q.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if query[:skip_killed]
462 q.add_query Ferret::Search::TermQuery.new("source_id", query[:source_id]), :must if query[:source_id]
466 def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
467 def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end