require 'thread'
require 'fileutils'
-require_gem 'ferret', ">= 0.10.13"
+require 'ferret'
module Redwood
-class IndexError < StandardError
- attr_reader :source
-
- def initialize source, s
- super s
- @source = source
- end
-end
-
class Index
include Singleton
- LOAD_THREAD_PETIT_DELAY = 0.1
- LOAD_THREAD_GRAND_DELAY = 5
-
- MESSAGES_AT_A_TIME = 10
-
- attr_reader :index # debugging only
-
+ attr_reader :index
def initialize dir=BASE_DIR
@dir = dir
- @mutex = Mutex.new
- @load_thread = nil # loads new messages
@sources = {}
@sources_dirty = false
+ wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
+ sa = Ferret::Analysis::StandardAnalyzer.new Ferret::Analysis::FULL_ENGLISH_STOP_WORDS, true
+ @analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
+ @analyzer[:body] = sa
+ @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer
+
self.class.i_am_the_instance self
end
end
def save
+ Redwood::log "saving index and sources..."
FileUtils.mkdir_p @dir unless File.exists? @dir
save_sources
save_index
raise "duplicate source!" if @sources.include? source
@sources_dirty = true
source.id ||= @sources.size
- source.id += 1 while @sources.member? source.id
+ ##TODO: why was this necessary?
+ ##source.id += 1 while @sources.member? source.id
@sources[source.id] = source
end
- def source_for name; @sources.values.find { |s| s.is_source_for? name }; end
+ def source_for uri; @sources.values.find { |s| s.is_source_for? uri }; end
def usual_sources; @sources.values.find_all { |s| s.usual? }; end
+ def sources; @sources.values; end
def load_index dir=File.join(@dir, "ferret")
- wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
- sa = Ferret::Analysis::StandardAnalyzer.new
- analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
- analyzer[:body] = sa
-
if File.exists? dir
- Redwood::log "loading index"
- @index = Ferret::Index::Index.new(:path => dir, :analyzer => analyzer)
+ Redwood::log "loading index..."
+ @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer)
+ Redwood::log "loaded index of #{@index.size} messages"
else
- Redwood::log "creating index"
+ Redwood::log "creating index..."
field_infos = Ferret::Index::FieldInfos.new :store => :yes
field_infos.add_field :message_id
field_infos.add_field :source_id
- field_infos.add_field :source_info, :index => :no, :term_vector => :no
+ field_infos.add_field :source_info
field_infos.add_field :date, :index => :untokenized
field_infos.add_field :body, :store => :no
field_infos.add_field :label
field_infos.add_field :refs
field_infos.add_field :snippet, :index => :no, :term_vector => :no
field_infos.create_index dir
- @index = Ferret::Index::Index.new(:path => dir, :analyzer => analyzer)
+ @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer)
end
end
- ## update the message by deleting and re-adding
- def update_message m, source=nil, source_info=nil
- docid, entry = load_entry_for_id m.id
- if entry
- source ||= entry[:source_id].to_i
- source_info ||= entry[:source_info].to_i
- end
- raise "no entry and no source info for message #{m.id}" unless source && source_info
+ ## Syncs the message to the index: deleting if it's already there,
+ ## and adding either way. Index state will be determined by m.labels.
+ ##
+ ## docid and entry can be specified if they're already known.
+ def sync_message m, docid=nil, entry=nil
+ docid, entry = load_entry_for_id m.id unless docid && entry
+
+ raise "no source info for message #{m.id}" unless m.source && m.source_info
+ raise "trying deleting non-corresponding entry #{docid}" if docid && @index[docid][:message_id] != m.id
+
+ source_id =
+ if m.source.is_a? Integer
+ raise "Debugging: integer source set"
+ m.source
+ else
+ m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
+ end
+
+ to = (m.to + m.cc + m.bcc).map { |x| x.email }.join(" ")
+ d = {
+ :message_id => m.id,
+ :source_id => source_id,
+ :source_info => m.source_info,
+ :date => m.date.to_indexable_s,
+ :body => m.content,
+ :snippet => m.snippet,
+ :label => m.labels.join(" "),
+ :from => m.from ? m.from.email : "",
+ :to => (m.to + m.cc + m.bcc).map { |x| x.email }.join(" "),
+ :subject => wrap_subj(Message.normalize_subj(m.subj)),
+ :refs => (m.refs + m.replytos).uniq.join(" "),
+ }
- raise "deleting non-corresponding entry #{docid}" unless @index[docid][:message_id] == m.id
- @index.delete docid
- add_message m
+ @index.delete docid if docid
+ @index.add_document d
+
+ docid, entry = load_entry_for_id m.id
+ ## this hasn't been triggered in a long time. TODO: decide whether it's still a problem.
+ raise "just added message #{m.id} but couldn't find it in a search" unless docid
+ true
end
def save_index fn=File.join(@dir, "ferret")
- # don't have to do anything apparently
+ # don't have to do anything, apparently
end
def contains_id? id
def size; @index.size; end
## you should probably not call this on a block that doesn't break
- ## rather quickly because the results will probably be, as we say
- ## in scotland, frikkin' huuuge.
+ ## rather quickly because the results can be very large.
EACH_BY_DATE_NUM = 100
def each_id_by_date opts={}
- return if @index.size == 0 # otherwise ferret barfs
+ return if @index.size == 0 # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
query = build_query opts
offset = 0
while true
end
def num_results_for opts={}
- query = build_query opts
- x = @index.search(query).total_hits
- Redwood::log "num_results_for: have #{x} for query #{query}"
- x
+ return 0 if @index.size == 0 # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
+ q = build_query opts
+ index.search(q).total_hits
end
+ ## yield all messages in the thread containing 'm' by repeatedly
+ ## querying the index. yields pairs of message ids and
+ ## message-building lambdas, so that building an unwanted message
+ ## can be skipped in the block if desired.
+ ##
+ ## stops loading any thread if a message with a :killed flag is found.
SAME_SUBJECT_DATE_LIMIT = 7
def each_message_in_thread_for m, opts={}
+ Redwood::log "Building thread for #{m.id}: #{m.subj}"
messages = {}
searched = {}
num_queries = 0
- ## temporarily disabling subject searching because it's a
- ## significant slowdown.
- ##
- ## TODO: make this configurable, i guess
- if false
+ if $config[:thread_by_subject] # do subject queries
date_min = m.date - (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
date_max = m.date + (SAME_SUBJECT_DATE_LIMIT * 12 * 3600)
q.add_query sq, :must
q.add_query Ferret::Search::RangeQuery.new(:date, :>= => date_min.to_indexable_s, :<= => date_max.to_indexable_s), :must
+ q = build_query :qobj => q
+
pending = @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] }
Redwood::log "found #{pending.size} results for subject query #{q}"
else
q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
+ q = build_query :qobj => q, :load_killed => true
+
num_queries += 1
@index.search_each(q, :limit => :all) do |docid, score|
break if opts[:limit] && messages.size >= opts[:limit]
+ break if @index[docid][:label].split(/\s+/).include? "killed" unless opts[:load_killed]
mid = @index[docid][:message_id]
- unless messages.member? mid
+ unless id == mid || messages.member?(mid)
+ Redwood::log "got #{mid} as a child of #{id}"
messages[mid] ||= lambda { build_message docid }
refs = @index[docid][:refs].split(" ")
pending += refs
end
end
end
- Redwood::log "ran #{num_queries} queries to build thread of #{messages.size} messages for #{m.id}"
+ Redwood::log "ran #{num_queries} queries to build thread of #{messages.size + 1} messages for #{m.id}" if num_queries > 0
messages.each { |mid, builder| yield mid, builder }
end
source = @sources[doc[:source_id].to_i]
#puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
raise "invalid source #{doc[:source_id]}" unless source
- begin
- raise "no snippet" unless doc[:snippet]
- Message.new source, doc[:source_info].to_i,
- doc[:label].split(" ").map { |s| s.intern },
- doc[:snippet]
- rescue MessageFormatError => e
- raise IndexError.new(source, "error building message #{doc[:message_id]} at #{source}/#{doc[:source_info]}: #{e.message}")
- nil
- end
- end
- def start_load_thread
- return if @load_thread
- @load_thread = true
- @load_thread = ::Thread.new do
- while @load_thread
- load_some_entries ENTRIES_AT_A_TIME, LOAD_THREAD_PETIT_DELAY, LOAD_THREAD_GRAND_DELAY
- end
- end
+ fake_header = {
+ "date" => Time.at(doc[:date].to_i),
+ "subject" => unwrap_subj(doc[:subject]),
+ "from" => doc[:from],
+ "to" => doc[:to],
+ "message-id" => doc[:message_id],
+ "references" => doc[:refs],
+ }
+
+ Message.new :source => source, :source_info => doc[:source_info].to_i,
+ :labels => doc[:label].split(" ").map { |s| s.intern },
+ :snippet => doc[:snippet], :header => fake_header
end
- def end_load_thread; @load_thread = nil; end
def fresh_thread_id; @next_thread_id += 1; end
-
def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
-
- def add_message m
- return false if contains? m
-
- source_id =
- if m.source.is_a? Integer
- m.source
- else
- m.source.id or raise "unregistered source #{m.source}"
- end
-
- to = (m.to + m.cc + m.bcc).map { |x| x.email }.join(" ")
- d = {
- :message_id => m.id,
- :source_id => source_id,
- :source_info => m.source_info,
- :date => m.date.to_indexable_s,
- :body => m.content,
- :snippet => m.snippet,
- :label => m.labels.join(" "),
- :from => m.from ? m.from.email : "",
- :to => (m.to + m.cc + m.bcc).map { |x| x.email }.join(" "),
- :subject => wrap_subj(Message.normalize_subj(m.subj)),
- :refs => (m.refs + m.replytos).join(" "),
- }
-
- @index.add_document d
-
- ## TODO: figure out why this is sometimes triggered
- #docid, entry = load_entry_for_id m.id
- #raise "just added message #{m.id} but couldn't find it in a search" unless docid
- true
- end
+ def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end
def drop_entry docno; @index.delete docno; end
contacts.keys.compact
end
+ def load_sources fn=Redwood::SOURCE_FN
+ source_array = (Redwood::load_yaml_obj(fn) || []).map { |o| Recoverable.new o }
+ @sources = Hash[*(source_array).map { |s| [s.id, s] }.flatten]
+ @sources_dirty = false
+ end
+
protected
- ## TODO: convert this to query objects rather than strings
+ def parse_user_query_string str; @qparser.parse str; end
def build_query opts
- query = ""
- query += opts[:labels].map { |t| "+label:#{t}" }.join(" ") if opts[:labels]
- query += " +label:#{opts[:label]}" if opts[:label]
- query += " #{opts[:content]}" if opts[:content]
+ query = Ferret::Search::BooleanQuery.new
+ query.add_query opts[:qobj], :must if opts[:qobj]
+ labels = ([opts[:label]] + (opts[:labels] || [])).compact
+ labels.each { |t| query.add_query Ferret::Search::TermQuery.new("label", t.to_s), :must }
if opts[:participants]
- query += "+(" +
- opts[:participants].map { |p| "from:#{p.email} OR to:#{p.email}" }.join(" OR ") + ")"
+ q2 = Ferret::Search::BooleanQuery.new
+ opts[:participants].each do |p|
+ q2.add_query Ferret::Search::TermQuery.new("from", p.email), :should
+ q2.add_query Ferret::Search::TermQuery.new("to", p.email), :should
+ end
+ query.add_query q2, :must
end
- query += " -label:spam" unless opts[:load_spam] || opts[:labels] == :spam ||
- (opts[:labels] && opts[:labels].include?(:spam))
- query += " -label:killed" unless opts[:load_killed] || opts[:labels] == :killed ||
- (opts[:labels] && opts[:labels].include?(:killed))
+ query.add_query Ferret::Search::TermQuery.new("label", "spam"), :must_not unless opts[:load_spam] || labels.include?(:spam)
+ query.add_query Ferret::Search::TermQuery.new("label", "deleted"), :must_not unless opts[:load_deleted] || labels.include?(:deleted)
+ query.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not unless opts[:load_killed] || labels.include?(:killed)
query
end
- def load_sources fn=Redwood::SOURCE_FN
- @sources = Hash[*(Redwood::load_yaml_obj(fn) || []).map { |s| [s.id, s] }.flatten]
- @sources_dirty = false
- end
-
def save_sources fn=Redwood::SOURCE_FN
if @sources_dirty || @sources.any? { |id, s| s.dirty? }
- FileUtils.mv fn, fn + ".bak", :force => true if File.exists? fn
- Redwood::save_yaml_obj @sources.values, fn
- end
- @sources_dirty = false
- end
-
- def load_some_entries max=ENTRIES_AT_A_TIME, delay1=nil, delay2=nil
- num = 0
- begin
- @sources.each_with_index do |source, source_id|
- next if source.done? || num >= max
- source.each do |source_info, label|
- begin
- m = Message.new(source, source_info, label + [:inbox])
- add_message m unless contains_id? m.id
- puts m.content.inspect
- num += 1
- rescue MessageFormatError => e
- $stderr.puts "ignoring erroneous message at #{source}##{source_info}: #{e.message}"
- end
- break if num >= max
- sleep delay1 if delay1
- end
- Redwood::log "loaded #{num} entries from #{source}"
- sleep delay2 if delay2
+ bakfn = fn + ".bak"
+ if File.exists? fn
+ File.chmod 0600, fn
+ FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(bakfn) > File.size(fn)
end
- ensure
- save_sources
- save_index
+ Redwood::save_yaml_obj @sources.values, fn
+ File.chmod 0600, fn
end
- num
+ @sources_dirty = false
end
end
projects / sup / blobdiff