## the index structure for redwood. interacts with ferret.
-require 'thread'
require 'fileutils'
require 'ferret'
+begin
+ require 'chronic'
+ $have_chronic = true
+rescue LoadError => e
+ Redwood::log "optional 'chronic' library not found (run 'gem install chronic' to install)"
+ $have_chronic = false
+end
module Redwood
include Singleton
attr_reader :index
+ alias ferret index
def initialize dir=BASE_DIR
@dir = dir
@sources = {}
@sources_dirty = false
wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
- sa = Ferret::Analysis::StandardAnalyzer.new Ferret::Analysis::FULL_ENGLISH_STOP_WORDS, true
+ sa = Ferret::Analysis::StandardAnalyzer.new [], true
@analyzer = Ferret::Analysis::PerFieldAnalyzer.new wsa
@analyzer[:body] = sa
@analyzer[:subject] = sa
- @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer
+ @qparser ||= Ferret::QueryParser.new :default_field => :body, :analyzer => @analyzer, :or_default => false
@lock = Lockfile.new lockfile, :retries => 0, :max_age => nil
self.class.i_am_the_instance self
end
def start_lock_update_thread
- @lock_update_thread = Redwood::reporting_thread do
+ @lock_update_thread = Redwood::reporting_thread("lock update") do
while true
sleep 30
@lock.touch_yourself
@lock_update_thread = nil
end
+ def possibly_pluralize number_of, kind
+ "#{number_of} #{kind}" +
+ if number_of == 1 then "" else "s" end
+ end
+
def fancy_lock_error_message_for e
- secs = Time.now - e.mtime
- mins = secs.to_i / 60
+ secs = (Time.now - e.mtime).to_i
+ mins = secs / 60
time =
if mins == 0
- "#{secs.to_i} seconds"
+ possibly_pluralize secs , "second"
else
- "#{mins} minutes"
+ possibly_pluralize mins, "minute"
end
<<EOS
def add_source source
raise "duplicate source!" if @sources.include? source
@sources_dirty = true
- source.id ||= @sources.size
- ##TODO: why was this necessary?
+ max = @sources.max_of { |id, s| s.is_a?(DraftLoader) || s.is_a?(SentLoader) ? 0 : id }
+ source.id ||= (max || 0) + 1
##source.id += 1 while @sources.member? source.id
@sources[source.id] = source
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 sources
+ ## favour the inbox by listing non-archived sources first
+ @sources.values.sort_by { |s| s.id }.partition { |s| !s.archived? }.flatten
+ end
+
+ def source_for uri; sources.find { |s| s.is_source_for? uri }; end
+ def usual_sources; sources.find_all { |s| s.usual? }; end
def load_index dir=File.join(@dir, "ferret")
if File.exists? dir
else
Redwood::log "creating index..."
field_infos = Ferret::Index::FieldInfos.new :store => :yes
- field_infos.add_field :message_id
+ field_infos.add_field :message_id, :index => :untokenized
field_infos.add_field :source_id
field_infos.add_field :source_info
field_infos.add_field :date, :index => :untokenized
- field_infos.add_field :body, :store => :no
+ field_infos.add_field :body
field_infos.add_field :label
+ field_infos.add_field :attachments
field_infos.add_field :subject
field_infos.add_field :from
field_infos.add_field :to
## 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
+ def sync_message m, docid=nil, entry=nil, opts={}
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
+ raise "trying to delete non-corresponding entry #{docid} with index message-id #{@index[docid][:message_id].inspect} and parameter message id #{m.id.inspect}" 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(" ")
+ snippet =
+ if m.snippet_contains_encrypted_content? && $config[:discard_snippets_from_encrypted_messages]
+ ""
+ else
+ m.snippet
+ end
+
+ ## write the new document to the index. if the entry already exists in the
+ ## index, reuse it (which avoids having to reload the entry from the source,
+ ## which can be quite expensive for e.g. large threads of IMAP actions.)
+ ##
+ ## exception: if the index entry belongs to an earlier version of the
+ ## message, use everything from the new message instead, but union the
+ ## flags. this allows messages sent to mailing lists to have their header
+ ## updated and to have flags set properly.
+ ##
+ ## minor hack: messages in sources with lower ids have priority over
+ ## messages in sources with higher ids. so messages in the inbox will
+ ## override everyone, and messages in the sent box will be overridden
+ ## by everyone else.
+ ##
+ ## written in this manner to support previous versions of the index which
+ ## did not keep around the entry body. upgrading is thus seamless.
+ entry ||= {}
+ labels = m.labels.uniq # override because this is the new state, unless...
+
+ ## if we are a later version of a message, ignore what's in the index,
+ ## but merge in the labels.
+ if entry[:source_id] && entry[:source_info] && entry[:label] &&
+ ((entry[:source_id].to_i > source_id) || (entry[:source_info].to_i < m.source_info))
+ labels = (entry[:label].split(/\s+/).map { |l| l.intern } + m.labels).uniq
+ #Redwood::log "found updated version of message #{m.id}: #{m.subj}"
+ #Redwood::log "previous version was at #{entry[:source_id].inspect}:#{entry[:source_info].inspect}, this version at #{source_id.inspect}:#{m.source_info.inspect}"
+ #Redwood::log "merged labels are #{labels.inspect} (index #{entry[:label].inspect}, message #{m.labels.inspect})"
+ entry = {}
+ end
+
+ ## if force_overwite is true, ignore what's in the index. this is used
+ ## primarily by sup-sync to force index updates.
+ entry = {} if opts[:force_overwrite]
+
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(" "),
+ :date => (entry[:date] || m.date.to_indexable_s),
+ :body => (entry[:body] || m.indexable_content),
+ :snippet => snippet, # always override
+ :label => labels.uniq.join(" "),
+ :attachments => (entry[:attachments] || m.attachments.uniq.join(" ")),
+ :from => (entry[:from] || (m.from ? m.from.indexable_content : "")),
+ :to => (entry[:to] || (m.to + m.cc + m.bcc).map { |x| x.indexable_content }.join(" ")),
+ :subject => (entry[:subject] || wrap_subj(Message.normalize_subj(m.subj))),
+ :refs => (entry[:refs] || (m.refs + m.replytos).uniq.join(" ")),
}
@index.delete docid if docid
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
+ raise "just added message #{m.id.inspect} 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
## 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.
+ ## only two options, :limit and :skip_killed. if :skip_killed is
+ ## true, stops loading any thread if a message with a :killed flag
+ ## is found.
SAME_SUBJECT_DATE_LIMIT = 7
+ MAX_CLAUSES = 1000
def each_message_in_thread_for m, opts={}
- Redwood::log "Building thread for #{m.id}: #{m.subj}"
+ #Redwood::log "Building thread for #{m.id}: #{m.subj}"
messages = {}
searched = {}
num_queries = 0
+ pending = [m.id]
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 = 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
- pending = [m.id]
+ p1 = @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] }
+ Redwood::log "found #{p1.size} results for subject query #{q}"
+
+ p2 = @index.search(q.to_s, :limit => :all).hits.map { |hit| @index[hit.doc][:message_id] }
+ Redwood::log "found #{p2.size} results in string form"
+
+ pending = (pending + p1 + p2).uniq
end
until pending.empty? || (opts[:limit] && messages.size >= opts[:limit])
- id = pending.pop
- next if searched.member? id
- searched[id] = true
q = Ferret::Search::BooleanQuery.new true
- q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
- q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
+ # this disappeared in newer ferrets... wtf.
+ # q.max_clause_count = 2048
+
+ lim = [MAX_CLAUSES / 2, pending.length].min
+ pending[0 ... lim].each do |id|
+ searched[id] = true
+ q.add_query Ferret::Search::TermQuery.new(:message_id, id), :should
+ q.add_query Ferret::Search::TermQuery.new(:refs, id), :should
+ end
+ pending = pending[lim .. -1]
- q = build_query :qobj => q, :load_killed => true
+ q = build_query :qobj => q
num_queries += 1
+ killed = false
@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]
+ if @index[docid][:label].split(/\s+/).include?("killed") && opts[:skip_killed]
+ killed = true
+ break
+ end
mid = @index[docid][:message_id]
unless messages.member?(mid)
- Redwood::log "got #{mid} as a child of #{id}"
+ #Redwood::log "got #{mid} as a child of #{id}"
messages[mid] ||= lambda { build_message docid }
refs = @index[docid][:refs].split(" ")
- pending += refs
+ pending += refs.select { |id| !searched[id] }
end
end
end
- 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 }
+
+ if killed
+ Redwood::log "thread for #{m.id} is killed, ignoring"
+ false
+ else
+ Redwood::log "ran #{num_queries} queries to build thread of #{messages.size + 1} messages for #{m.id}: #{m.subj}" if num_queries > 0
+ messages.each { |mid, builder| yield mid, builder }
+ true
+ end
end
## builds a message object from a ferret result
"date" => Time.at(doc[:date].to_i),
"subject" => unwrap_subj(doc[:subject]),
"from" => doc[:from],
- "to" => doc[:to],
+ "to" => doc[:to].split(/\s+/).join(", "), # reformat
"message-id" => doc[:message_id],
- "references" => doc[:refs],
+ "references" => doc[:refs].split(/\s+/).map { |x| "<#{x}>" }.join(" "),
}
Message.new :source => source, :source_info => doc[:source_info].to_i,
num = h[:num] || 20
@index.search_each(q, :sort => "date DESC", :limit => :all) do |docid, score|
break if contacts.size >= num
- #Redwood::log "got message with to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
+ #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
f = @index[docid][:from]
t = @index[docid][:to]
@sources_dirty = false
end
+ def has_any_from_source_with_label? source, label
+ q = Ferret::Search::BooleanQuery.new
+ q.add_query Ferret::Search::TermQuery.new("source_id", source.id.to_s), :must
+ q.add_query Ferret::Search::TermQuery.new("label", label.to_s), :must
+ index.search(q, :limit => 1).total_hits > 0
+ end
+
protected
- def parse_user_query_string str; @qparser.parse str; end
+ ## do any specialized parsing
+ ## returns nil and flashes error message if parsing failed
+ def parse_user_query_string s
+ extraopts = {}
+
+ subs = s.gsub(/\b(to|from):(\S+)\b/) do
+ field, name = $1, $2
+ if(p = ContactManager.contact_for(name))
+ [field, p.email]
+ elsif name == "me"
+ [field, "(" + AccountManager.user_emails.join("||") + ")"]
+ else
+ [field, name]
+ end.join(":")
+ end
+
+ ## if we see a label:deleted or a label:spam term anywhere in the query
+ ## string, we set the extra load_spam or load_deleted options to true.
+ ## bizarre? well, because the query allows arbitrary parenthesized boolean
+ ## expressions, without fully parsing the query, we can't tell whether
+ ## the user is explicitly directing us to search spam messages or not.
+ ## e.g. if the string is -(-(-(-(-label:spam)))), does the user want to
+ ## search spam messages or not?
+ ##
+ ## so, we rely on the fact that turning these extra options ON turns OFF
+ ## the adding of "-label:deleted" or "-label:spam" terms at the very
+ ## final stage of query processing. if the user wants to search spam
+ ## messages, not adding that is the right thing; if he doesn't want to
+ ## search spam messages, then not adding it won't have any effect.
+ extraopts[:load_spam] = true if subs =~ /\blabel:spam\b/
+ extraopts[:load_deleted] = true if subs =~ /\blabel:deleted\b/
+
+ ## gmail style "is" operator
+ subs = subs.gsub(/\b(is|has):(\S+)\b/) do
+ field, label = $1, $2
+ case label
+ when "read"
+ "-label:unread"
+ when "spam"
+ extraopts[:load_spam] = true
+ "label:spam"
+ when "deleted"
+ extraopts[:load_deleted] = true
+ "label:deleted"
+ else
+ "label:#{$2}"
+ end
+ end
+
+ ## gmail style attachments "filename" and "filetype" searches
+ subs = subs.gsub(/\b(filename|filetype):(\((.+?)\)\B|(\S+)\b)/) do
+ field, name = $1, ($3 || $4)
+ case field
+ when "filename"
+ Redwood::log "filename - translated #{field}:#{name} to attachments:(#{name.downcase})"
+ "attachments:(#{name.downcase})"
+ when "filetype"
+ Redwood::log "filetype - translated #{field}:#{name} to attachments:(*.#{name.downcase})"
+ "attachments:(*.#{name.downcase})"
+ end
+ end
+
+ if $have_chronic
+ chronic_failure = false
+ subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
+ break if chronic_failure
+ field, datestr = $1, ($3 || $4)
+ realdate = Chronic.parse(datestr, :guess => false, :context => :none)
+ if realdate
+ case field
+ when "after"
+ Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.end}"
+ "date:(>= #{sprintf "%012d", realdate.end.to_i})"
+ when "before"
+ Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate.begin}"
+ "date:(<= #{sprintf "%012d", realdate.begin.to_i})"
+ else
+ Redwood::log "chronic: translated #{field}:#{datestr} to #{realdate}"
+ "date:(<= #{sprintf "%012d", realdate.end.to_i}) date:(>= #{sprintf "%012d", realdate.begin.to_i})"
+ end
+ else
+ BufferManager.flash "Can't understand date #{datestr.inspect}!"
+ chronic_failure = true
+ end
+ end
+ subs = nil if chronic_failure
+ end
+
+ if subs
+ [@qparser.parse(subs), extraopts]
+ else
+ nil
+ end
+ end
+
def build_query opts
query = Ferret::Search::BooleanQuery.new
query.add_query opts[:qobj], :must if opts[:qobj]
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.add_query Ferret::Search::TermQuery.new("label", "killed"), :must_not if opts[:skip_killed]
query
end
File.chmod 0600, fn
FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(fn) == 0
end
- Redwood::save_yaml_obj @sources.values.sort_by { |s| s.id.to_i }, fn, true
+ Redwood::save_yaml_obj sources.sort_by { |s| s.id.to_i }, fn, true
File.chmod 0600, fn
end
@sources_dirty = false