6 # This index implementation uses Xapian for searching and GDBM for storage. It
7 # tends to be slightly faster than Ferret for indexing and significantly faster
8 # for searching due to precomputing thread membership.
9 class XapianIndex < BaseIndex
10 STEM_LANGUAGE = "english"
13 ## dates are converted to integers for xapian, and are used for document ids,
14 ## so we must ensure they're reasonably valid. this typically only affect
17 MAX_DATE = Time.at(2**31-1)
19 def initialize dir=BASE_DIR
22 @index_mutex = Monitor.new
26 path = File.join(@dir, 'xapian')
28 @xapian = Xapian::WritableDatabase.new(path, Xapian::DB_OPEN)
29 db_version = @xapian.get_metadata 'version'
30 db_version = '0' if db_version.empty?
31 if db_version != INDEX_VERSION
32 fail "This Sup version expects a v#{INDEX_VERSION} index, but you have an existing v#{db_version} index. Please downgrade to your previous version and dump your labels before upgrading to this version (then run sup-sync --restore)."
35 @xapian = Xapian::WritableDatabase.new(path, Xapian::DB_CREATE)
36 @xapian.set_metadata 'version', INDEX_VERSION
38 @term_generator = Xapian::TermGenerator.new()
39 @term_generator.stemmer = Xapian::Stem.new(STEM_LANGUAGE)
40 @enquire = Xapian::Enquire.new @xapian
41 @enquire.weighting_scheme = Xapian::BoolWeight.new
42 @enquire.docid_order = Xapian::Enquire::ASCENDING
52 synchronize { @xapian.doccount }
56 synchronize { find_docid(id) && true }
60 synchronize { get_entry(id)[:source_id] }
64 synchronize { @xapian.delete_document mkterm(:msgid, id) }
68 entry = synchronize { get_entry id }
71 source = SourceManager[entry[:source_id]]
72 raise "invalid source #{entry[:source_id]}" unless source
74 mk_addrs = lambda { |l| l.map { |e,n| "#{n} <#{e}>" } * ', ' }
75 mk_refs = lambda { |l| l.map { |r| "<#{r}>" } * ' ' }
77 'message-id' => entry[:message_id],
78 'date' => Time.at(entry[:date]),
79 'subject' => entry[:subject],
80 'from' => mk_addrs[[entry[:from]]],
81 'to' => mk_addrs[entry[:to]],
82 'cc' => mk_addrs[entry[:cc]],
83 'bcc' => mk_addrs[entry[:bcc]],
84 'reply-tos' => mk_refs[entry[:replytos]],
85 'references' => mk_refs[entry[:refs]],
88 m = Message.new :source => source, :source_info => entry[:source_info],
89 :labels => entry[:labels],
90 :snippet => entry[:snippet]
91 m.parse_header fake_header
95 def sync_message m, opts={}
96 entry = synchronize { get_entry m.id }
100 entry = {} if opts[:force_overwrite]
104 :source_id => m.source.id,
105 :source_info => m.source_info,
106 :date => (entry[:date] || m.date),
108 :labels => labels.uniq,
109 :from => (entry[:from] || [m.from.email, m.from.name]),
110 :to => (entry[:to] || m.to.map { |p| [p.email, p.name] }),
111 :cc => (entry[:cc] || m.cc.map { |p| [p.email, p.name] }),
112 :bcc => (entry[:bcc] || m.bcc.map { |p| [p.email, p.name] }),
114 :refs => (entry[:refs] || m.refs),
115 :replytos => (entry[:replytos] || m.replytos),
118 m.labels.each { |l| LabelManager << l }
121 index_message m, d, opts
126 def num_results_for query={}
127 xapian_query = build_xapian_query query
128 matchset = run_query xapian_query, 0, 0, 100
129 matchset.matches_estimated
137 xapian_query = build_xapian_query query
139 ids = run_query_ids xapian_query, offset, (offset+page)
140 ids.each { |id| yield id }
141 break if ids.size < page
146 def each_id_by_date query={}
147 each_id(query) { |id| yield id, lambda { build_message id } }
150 def each_message_in_thread_for m, opts={}
151 # TODO thread by subject
152 # TODO handle killed threads
153 return unless doc = find_doc(m.id)
154 queue = doc.value(THREAD_VALUENO).split(',')
156 seen_threads = Set.new
157 seen_messages = Set.new [m.id]
158 while not queue.empty?
159 thread_id = queue.pop
160 next if seen_threads.member? thread_id
161 return false if thread_killed? thread_id
162 seen_threads << thread_id
163 docs = term_docids(mkterm(:thread, thread_id)).map { |x| @xapian.document x }
165 msgid = doc.value MSGID_VALUENO
166 next if seen_messages.member? msgid
168 seen_messages << msgid
169 queue.concat doc.value(THREAD_VALUENO).split(',')
172 msgids.each { |id| yield id, lambda { build_message id } }
176 def load_contacts emails, opts={}
178 num = opts[:num] || 20
179 each_id_by_date :participants => emails do |id,b|
180 break if contacts.size >= num
182 ([m.from]+m.to+m.cc+m.bcc).compact.each { |p| contacts << [p.name, p.email] }
184 contacts.to_a.compact.map { |n,e| Person.new n, e }[0...num]
187 # TODO share code with the Ferret index
191 subs = s.gsub(/\b(to|from):(\S+)\b/) do
193 if(p = ContactManager.contact_for(name))
196 [field, "(" + AccountManager.user_emails.join("||") + ")"]
202 ## if we see a label:deleted or a label:spam term anywhere in the query
203 ## string, we set the extra load_spam or load_deleted options to true.
204 ## bizarre? well, because the query allows arbitrary parenthesized boolean
205 ## expressions, without fully parsing the query, we can't tell whether
206 ## the user is explicitly directing us to search spam messages or not.
207 ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
208 ## search spam messages or not?
210 ## so, we rely on the fact that turning these extra options ON turns OFF
211 ## the adding of "-label:deleted" or "-label:spam" terms at the very
212 ## final stage of query processing. if the user wants to search spam
213 ## messages, not adding that is the right thing; if he doesn't want to
214 ## search spam messages, then not adding it won't have any effect.
215 query[:load_spam] = true if subs =~ /\blabel:spam\b/
216 query[:load_deleted] = true if subs =~ /\blabel:deleted\b/
218 ## gmail style "is" operator
219 subs = subs.gsub(/\b(is|has):(\S+)\b/) do
220 field, label = $1, $2
225 query[:load_spam] = true
228 query[:load_deleted] = true
235 ## gmail style attachments "filename" and "filetype" searches
236 subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
237 field, name = $1, ($3 || $4)
240 Redwood::log "filename - translated #{field}:#{name} to attachment:\"#{name.downcase}\""
241 "attachment:\"#{name.downcase}\""
243 Redwood::log "filetype - translated #{field}:#{name} to attachment_extension:#{name.downcase}"
244 "attachment_extension:#{name.downcase}"
251 subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
252 field, datestr = $1, ($3 || $4)
253 realdate = Chronic.parse datestr, :guess => false, :context => :past
257 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
258 "date:#{realdate.end.to_i}..#{lastdate}"
260 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
261 "date:#{firstdate}..#{realdate.end.to_i}"
263 Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
264 "date:#{realdate.begin.to_i}..#{realdate.end.to_i}"
267 raise ParseError, "can't understand date #{datestr.inspect}"
272 ## limit:42 restrict the search to 42 results
273 subs = subs.gsub(/\blimit:(\S+)\b/) do
276 query[:limit] = lim.to_i
279 raise ParseError, "non-numeric limit #{lim.inspect}"
283 qp = Xapian::QueryParser.new
284 qp.database = @xapian
285 qp.stemmer = Xapian::Stem.new(STEM_LANGUAGE)
286 qp.stemming_strategy = Xapian::QueryParser::STEM_SOME
287 qp.default_op = Xapian::Query::OP_AND
288 qp.add_valuerangeprocessor(Xapian::NumberValueRangeProcessor.new(DATE_VALUENO, 'date:', true))
289 NORMAL_PREFIX.each { |k,v| qp.add_prefix k, v }
290 BOOLEAN_PREFIX.each { |k,v| qp.add_boolean_prefix k, v }
291 xapian_query = qp.parse_query(subs, Xapian::QueryParser::FLAG_PHRASE|Xapian::QueryParser::FLAG_BOOLEAN|Xapian::QueryParser::FLAG_LOVEHATE|Xapian::QueryParser::FLAG_WILDCARD, PREFIX['body'])
293 raise ParseError if xapian_query.nil? or xapian_query.empty?
294 query[:qobj] = xapian_query
314 'from_email' => 'FE',
320 'attachment_extension' => 'O',
326 PREFIX = NORMAL_PREFIX.merge BOOLEAN_PREFIX
332 MAX_TERM_LENGTH = 245
334 # Xapian can very efficiently sort in ascending docid order. Sup always wants
335 # to sort by descending date, so this method maps between them. In order to
336 # handle multiple messages per second, we use a logistic curve centered
337 # around MIDDLE_DATE so that the slope (docid/s) is greatest in this time
338 # period. A docid collision is not an error - the code will pick the next
339 # smallest unused one.
340 DOCID_SCALE = 2.0**32
342 MIDDLE_DATE = Time.gm(2011)
343 def assign_docid m, truncated_date
344 t = (truncated_date.to_i - MIDDLE_DATE.to_i).to_f
345 docid = (DOCID_SCALE - DOCID_SCALE/(Math::E**(-(t/TIME_SCALE)) + 1)).to_i
346 while docid > 0 and docid_exists? docid
349 docid > 0 ? docid : nil
352 # XXX is there a better way?
353 def docid_exists? docid
355 @xapian.doclength docid
357 rescue RuntimeError #Xapian::DocNotFoundError
358 raise unless $!.message =~ /DocNotFoundError/
364 @xapian.postlist(term).map { |x| x.docid }
368 term_docids(mkterm(:msgid,id)).tap { |x| fail unless x.size <= 1 }.first
372 return unless docid = find_docid(id)
373 @xapian.document docid
377 return unless doc = @xapian.document(docid)
378 doc.value MSGID_VALUENO
382 return unless doc = find_doc(id)
383 Marshal.load doc.data
386 def thread_killed? thread_id
387 not run_query(Q.new(Q::OP_AND, mkterm(:thread, thread_id), mkterm(:label, :Killed)), 0, 1).empty?
391 @index_mutex.synchronize &b
394 def run_query xapian_query, offset, limit, checkatleast=0
396 @enquire.query = xapian_query
397 @enquire.mset(offset, limit-offset, checkatleast)
401 def run_query_ids xapian_query, offset, limit
402 matchset = run_query xapian_query, offset, limit
403 matchset.matches.map { |r| r.document.value MSGID_VALUENO }
407 def build_xapian_query opts
408 labels = ([opts[:label]] + (opts[:labels] || [])).compact
409 neglabels = [:spam, :deleted, :killed].reject { |l| (labels.include? l) || opts.member?("load_#{l}".intern) }
410 pos_terms, neg_terms = [], []
412 pos_terms << mkterm(:type, 'mail')
413 pos_terms.concat(labels.map { |l| mkterm(:label,l) })
414 pos_terms << opts[:qobj] if opts[:qobj]
415 pos_terms << mkterm(:source_id, opts[:source_id]) if opts[:source_id]
417 if opts[:participants]
418 participant_terms = opts[:participants].map { |p| mkterm(:email,:any, (Redwood::Person === p) ? p.email : p) }
419 pos_terms << Q.new(Q::OP_OR, participant_terms)
422 neg_terms.concat(neglabels.map { |l| mkterm(:label,l) })
424 pos_query = Q.new(Q::OP_AND, pos_terms)
425 neg_query = Q.new(Q::OP_OR, neg_terms)
430 Q.new(Q::OP_AND_NOT, [pos_query, neg_query])
434 def index_message m, entry, opts
438 subject_text = m.indexable_subject
439 body_text = m.indexable_body
441 # Person names are indexed with several prefixes
442 person_termer = lambda do |d|
444 ["#{d}_name", "name", "body"].each do |x|
445 text << [p.name, PREFIX[x]]
447 [d, :any].each { |x| terms << mkterm(:email, x, p.email) }
451 person_termer[:from][m.from] if m.from
452 (m.to+m.cc+m.bcc).each(&(person_termer[:to]))
454 terms << mkterm(:date,m.date) if m.date
455 m.labels.each { |t| terms << mkterm(:label,t) }
456 terms << mkterm(:type, 'mail')
457 terms << mkterm(:msgid, m.id)
458 terms << mkterm(:source_id, m.source.id)
459 m.attachments.each do |a|
460 a =~ /\.(\w+)$/ or next
461 t = mkterm(:attachment_extension, $1)
466 children = term_docids(mkterm(:ref, m.id)).map { |docid| @xapian.document docid }
467 parent_ids = m.refs + m.replytos
468 parents = parent_ids.map { |id| find_doc id }.compact
469 thread_members = SavingHash.new { [] }
470 (children + parents).each do |doc2|
471 thread_ids = doc2.value(THREAD_VALUENO).split ','
472 thread_ids.each { |thread_id| thread_members[thread_id] << doc2 }
475 thread_ids = thread_members.empty? ? [m.id] : thread_members.keys
477 thread_ids.each { |thread_id| terms << mkterm(:thread, thread_id) }
478 parent_ids.each do |ref|
479 terms << mkterm(:ref, ref)
482 # Full text search content
483 text << [subject_text, PREFIX['subject']]
484 text << [subject_text, PREFIX['body']]
485 text << [body_text, PREFIX['body']]
486 m.attachments.each { |a| text << [a, PREFIX['attachment']] }
488 truncated_date = if m.date < MIN_DATE
489 Redwood::log "warning: adjusting too-low date #{m.date} for indexing"
491 elsif m.date > MAX_DATE
492 Redwood::log "warning: adjusting too-high date #{m.date} for indexing"
498 # Date value for range queries
500 Xapian.sortable_serialise truncated_date.to_i
502 Xapian.sortable_serialise 0
506 unless doc = find_doc(m.id)
507 doc = Xapian::Document.new
508 if not docid = assign_docid(m, truncated_date)
509 # Could be triggered by spam
510 Redwood::log "warning: docid underflow, dropping #{m.id.inspect}"
519 @term_generator.document = doc
520 text.each { |text,prefix| @term_generator.index_text text, 1, prefix }
521 terms.each { |term| doc.add_term term if term.length <= MAX_TERM_LENGTH }
522 doc.add_value MSGID_VALUENO, m.id
523 doc.add_value THREAD_VALUENO, (thread_ids * ',')
524 doc.add_value DATE_VALUENO, date_value
525 doc.data = Marshal.dump entry
527 @xapian.replace_document docid, doc
530 # Construct a Xapian term
531 def mkterm type, *args
534 PREFIX['label'] + args[0].to_s.downcase
536 PREFIX['type'] + args[0].to_s.downcase
538 PREFIX['date'] + args[0].getutc.strftime("%Y%m%d%H%M%S")
541 when :from then PREFIX['from_email']
542 when :to then PREFIX['to_email']
543 when :any then PREFIX['email']
544 else raise "Invalid email term type #{args[0]}"
545 end + args[1].to_s.downcase
547 PREFIX['source_id'] + args[0].to_s.downcase
548 when :attachment_extension
549 PREFIX['attachment_extension'] + args[0].to_s.downcase
550 when :msgid, :ref, :thread
551 PREFIX[type.to_s] + args[0][0...(MAX_TERM_LENGTH-1)]
553 raise "Invalid term type #{type}"