-## the index structure for redwood. interacts with ferret.
+## Index interface, subclassed by Ferret indexer.
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)"
+ debug "optional 'chronic' library not found; date-time query restrictions disabled"
$have_chronic = false
end
module Redwood
-class Index
+class BaseIndex
+ include InteractiveLock
+
class LockError < StandardError
def initialize h
@h = h
include Singleton
- ## these two accessors should ONLY be used by single-threaded programs.
- ## otherwise you will have a naughty ferret on your hands.
- attr_reader :index
- alias ferret index
-
def initialize dir=BASE_DIR
- @index_mutex = Monitor.new
-
@dir = dir
- @sources = {}
- @sources_dirty = false
- @source_mutex = Monitor.new
-
- wsa = Ferret::Analysis::WhiteSpaceAnalyzer.new false
- 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, :or_default => false
@lock = Lockfile.new lockfile, :retries => 0, :max_age => nil
-
- self.class.i_am_the_instance self
end
def lockfile; File.join @dir, "lock" end
def lock
- Redwood::log "locking #{lockfile}..."
+ debug "locking #{lockfile}..."
begin
@lock.lock
rescue Lockfile::MaxTriesLockError
@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).to_i
- mins = secs / 60
- time =
- if mins == 0
- possibly_pluralize secs , "second"
- else
- possibly_pluralize mins, "minute"
- end
-
- <<EOS
-Error: the sup index is locked by another process! User '#{e.user}' on
-host '#{e.host}' is running #{e.pname} with pid #{e.pid}. The process was alive
-as of #{time} ago.
-EOS
- end
-
- def lock_or_die
- begin
- lock
- rescue LockError => e
- $stderr.puts fancy_lock_error_message_for(e)
- $stderr.puts <<EOS
-
-You can wait for the process to finish, or, if it crashed and left a
-stale lock file behind, you can manually delete #{@lock.path}.
-EOS
- exit
- end
- end
-
def unlock
if @lock && @lock.locked?
- Redwood::log "unlocking #{lockfile}..."
+ debug "unlocking #{lockfile}..."
@lock.unlock
end
end
def load
- load_sources
+ SourceManager.load_sources
load_index
end
def save
- Redwood::log "saving index and sources..."
+ debug "saving index and sources..."
FileUtils.mkdir_p @dir unless File.exists? @dir
- save_sources
+ SourceManager.save_sources
save_index
end
- def add_source source
- @source_mutex.synchronize do
- raise "duplicate source!" if @sources.include? source
- @sources_dirty = true
- 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
- end
-
- def sources
- ## favour the inbox by listing non-archived sources first
- @source_mutex.synchronize { @sources.values }.sort_by { |s| s.id }.partition { |s| !s.archived? }.flatten
+ def load_index
+ unimplemented
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 add_message m; unimplemented end
+ def update_message m; unimplemented end
+ def update_message_state m; unimplemented end
- def load_index dir=File.join(@dir, "ferret")
- if File.exists? dir
- Redwood::log "loading index..."
- @index_mutex.synchronize do
- @index = Ferret::Index::Index.new(:path => dir, :analyzer => @analyzer)
- Redwood::log "loaded index of #{@index.size} messages"
- end
- else
- Redwood::log "creating index..."
- @index_mutex.synchronize do
- field_infos = Ferret::Index::FieldInfos.new :store => :yes
- 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
- 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
- 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)
- end
- end
+ def save_index fn
+ unimplemented
end
- ## Syncs the message to the index, replacing any previous version. adding
- ## either way. Index state will be determined by the message's #labels
- ## accessor.
- ##
- ## if need_load is false, docid and entry are assumed to be set to the
- ## result of load_entry_for_id (which can be nil).
- def sync_message m, need_load=true, docid=nil, entry=nil, opts={}
- docid, entry = load_entry_for_id m.id if need_load
-
- raise "no source info for message #{m.id}" unless m.source && m.source_info
- @index_mutex.synchronize do
- 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
- end
-
- source_id = if m.source.is_a? Integer
- m.source
- else
- m.source.id or raise "unregistered source #{m.source} (id #{m.source.id.inspect})"
- end
-
- 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].symbolistize + 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 => (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(" ")),
-
- ## always override :from and :to.
- ## older versions of Sup would often store the wrong thing in the index
- ## (because they were canonicalizing email addresses, resulting in the
- ## wrong name associated with each.) the correct address is read from
- ## the original header when these messages are opened in thread-view-mode,
- ## so this allows people to forcibly update the address in the index by
- ## marking those threads for saving.
- :from => (m.from ? m.from.indexable_content : ""),
- :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_mutex.synchronize do
- @index.delete docid if docid
- @index.add_document d
- end
-
- ## this hasn't been triggered in a long time.
- ## docid, entry = load_entry_for_id m.id
- ## raise "just added message #{m.id.inspect} but couldn't find it in a search" unless docid
+ def contains_id? id
+ unimplemented
end
- def save_index fn=File.join(@dir, "ferret")
- # don't have to do anything, apparently
- end
+ def contains? m; contains_id? m.id end
- def contains_id? id
- @index_mutex.synchronize { @index.search(Ferret::Search::TermQuery.new(:message_id, id)).total_hits > 0 }
+ def size
+ unimplemented
end
- def contains? m; contains_id? m.id end
- def size; @index_mutex.synchronize { @index.size } end
+
def empty?; size == 0 end
- ## you should probably not call this on a block that doesn't break
+ ## Yields a message-id and message-building lambda for each
+ ## message that matches the given query, in descending date order.
+ ## You should probably not call this on a block that doesn't break
## rather quickly because the results can be very large.
- EACH_BY_DATE_NUM = 100
- def each_id_by_date opts={}
- return if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
- query = build_query opts
- offset = 0
- while true
- limit = (opts[:limit])? [EACH_BY_DATE_NUM, opts[:limit] - offset].min : EACH_BY_DATE_NUM
- results = @index_mutex.synchronize { @index.search query, :sort => "date DESC", :limit => limit, :offset => offset }
- Redwood::log "got #{results.total_hits} results for query (offset #{offset}) #{query.inspect}"
- results.hits.each do |hit|
- yield @index_mutex.synchronize { @index[hit.doc][:message_id] }, lambda { build_message hit.doc }
- end
- break if opts[:limit] and offset >= opts[:limit] - limit
- break if offset >= results.total_hits - limit
- offset += limit
- end
+ def each_id_by_date query={}
+ unimplemented
end
- def num_results_for opts={}
- return 0 if empty? # otherwise ferret barfs ###TODO: remove this once my ferret patch is accepted
-
- q = build_query opts
- @index_mutex.synchronize { @index.search(q, :limit => 1).total_hits }
+ ## Return the number of matches for query in the index
+ def num_results_for query={}
+ unimplemented
end
## yield all messages in the thread containing 'm' by repeatedly
## 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}"
- 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 = Ferret::Search::BooleanQuery.new true
- sq = Ferret::Search::PhraseQuery.new(:subject)
- wrap_subj(Message.normalize_subj(m.subj)).split.each do |t|
- sq.add_term t
- end
- 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
-
- p1 = @index_mutex.synchronize { @index.search(q).hits.map { |hit| @index[hit.doc][:message_id] } }
- Redwood::log "found #{p1.size} results for subject query #{q}"
-
- p2 = @index_mutex.synchronize { @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])
- q = Ferret::Search::BooleanQuery.new true
- # 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
-
- num_queries += 1
- killed = false
- @index_mutex.synchronize do
- @index.search_each(q, :limit => :all) do |docid, score|
- break if opts[:limit] && messages.size >= opts[:limit]
- 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}"
- messages[mid] ||= lambda { build_message docid }
- refs = @index[docid][:refs].split
- pending += refs.select { |id| !searched[id] }
- end
- end
- end
- end
-
- 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} messages for #{m.id}: #{m.subj}" if num_queries > 0
- messages.each { |mid, builder| yield mid, builder }
- true
- end
+ unimplemented
end
- ## builds a message object from a ferret result
- def build_message docid
- @index_mutex.synchronize do
- doc = @index[docid]
-
- source = @source_mutex.synchronize { @sources[doc[:source_id].to_i] }
- raise "invalid source #{doc[:source_id]}" unless source
-
- #puts "building message #{doc[:message_id]} (#{source}##{doc[:source_info]})"
-
- fake_header = {
- "date" => Time.at(doc[:date].to_i),
- "subject" => unwrap_subj(doc[:subject]),
- "from" => doc[:from],
- "to" => doc[:to].split.join(", "), # reformat
- "message-id" => doc[:message_id],
- "references" => doc[:refs].split.map { |x| "<#{x}>" }.join(" "),
- }
-
- m = Message.new :source => source, :source_info => doc[:source_info].to_i,
- :labels => doc[:label].symbolistize,
- :snippet => doc[:snippet]
- m.parse_header fake_header
- m
- end
+ ## Load message with the given message-id from the index
+ def build_message id
+ unimplemented
end
- def fresh_thread_id; @next_thread_id += 1; end
- def wrap_subj subj; "__START_SUBJECT__ #{subj} __END_SUBJECT__"; end
- def unwrap_subj subj; subj =~ /__START_SUBJECT__ (.*?) __END_SUBJECT__/ && $1; end
-
- def drop_entry docno; @index_mutex.synchronize { @index.delete docno } end
-
- def load_entry_for_id mid
- @index_mutex.synchronize do
- results = @index.search Ferret::Search::TermQuery.new(:message_id, mid)
- return if results.total_hits == 0
- docid = results.hits[0].doc
- entry = @index[docid]
- entry_dup = entry.fields.inject({}) { |h, f| h[f] = entry[f]; h }
- [docid, entry_dup]
- end
+ ## Delete message with the given message-id from the index
+ def delete id
+ unimplemented
end
- def load_contacts emails, h={}
- q = Ferret::Search::BooleanQuery.new true
- emails.each do |e|
- qq = Ferret::Search::BooleanQuery.new true
- qq.add_query Ferret::Search::TermQuery.new(:from, e), :should
- qq.add_query Ferret::Search::TermQuery.new(:to, e), :should
- q.add_query qq
- end
- q.add_query Ferret::Search::TermQuery.new(:label, "spam"), :must_not
-
- Redwood::log "contact search: #{q}"
- contacts = {}
- num = h[:num] || 20
- @index_mutex.synchronize do
- @index.search_each q, :sort => "date DESC", :limit => :all do |docid, score|
- break if contacts.size >= num
- #Redwood::log "got message #{docid} to: #{@index[docid][:to].inspect} and from: #{@index[docid][:from].inspect}"
- f = @index[docid][:from]
- t = @index[docid][:to]
-
- if AccountManager.is_account_email? f
- t.split(" ").each { |e| contacts[Person.from_address(e)] = true }
- else
- contacts[Person.from_address(f)] = true
- end
- end
- end
+ ## Given an array of email addresses, return an array of Person objects that
+ ## have sent mail to or received mail from any of the given addresses.
+ def load_contacts email_addresses, h={}
+ unimplemented
+ end
- contacts.keys.compact
+ ## Yield each message-id matching query
+ def each_id query={}
+ unimplemented
end
- def load_sources fn=Redwood::SOURCE_FN
- source_array = (Redwood::load_yaml_obj(fn) || []).map { |o| Recoverable.new o }
- @source_mutex.synchronize do
- @sources = Hash[*(source_array).map { |s| [s.id, s] }.flatten]
- @sources_dirty = false
+ ## Yield each message matching query
+ def each_message query={}, &b
+ each_id query do |id|
+ yield build_message(id)
end
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_mutex.synchronize { @index.search(q, :limit => 1).total_hits > 0 }
+ ## Implementation-specific optimization step
+ def optimize
+ unimplemented
end
- ## takes a user query string and returns the list of docids for messages
- ## that match the query.
- ##
- ## messages can then be loaded from the index with #build_message.
- ##
- ## raises a ParseError if the parsing failed.
- def run_query query
- qobj, opts = Redwood::Index.parse_user_query_string query
- query = Redwood::Index.build_query opts.merge(:qobj => qobj)
- results = @index.search query, :limit => (opts[:limit] || :all)
- results.hits.map { |hit| hit.doc }
+ ## Return the id source of the source the message with the given message-id
+ ## was synced from
+ def source_for_id id
+ unimplemented
end
-protected
-
class ParseError < StandardError; end
- ## parse a query string from the user. returns a query object and a set of
- ## extra flags; both of these are meant to be passed to #build_query.
+ ## parse a query string from the user. returns a query object
+ ## that can be passed to any index method with a 'query'
+ ## argument.
##
## raises a ParseError if something went wrong.
- 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
- subs = subs.gsub(/\b(before|on|in|during|after):(\((.+?)\)\B|(\S+)\b)/) do
- field, datestr = $1, ($3 || $4)
- realdate = Chronic.parse datestr, :guess => false, :context => :past
- 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
- raise ParseError, "can't understand date #{datestr.inspect}"
- end
- end
- end
-
- ## limit:42 restrict the search to 42 results
- subs = subs.gsub(/\blimit:(\S+)\b/) do
- lim = $1
- if lim =~ /^\d+$/
- extraopts[:limit] = lim.to_i
- ''
- else
- raise ParseError, "non-numeric limit #{lim.inspect}"
- end
- end
-
- begin
- [@qparser.parse(subs), extraopts]
- rescue Ferret::QueryParser::QueryParseException => e
- raise ParseError, e.message
- end
- end
-
- def build_query opts
- 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]
- 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.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 if opts[:skip_killed]
- query
+ def parse_query s
+ unimplemented
end
+end
- def save_sources fn=Redwood::SOURCE_FN
- @source_mutex.synchronize do
- if @sources_dirty || @sources.any? { |id, s| s.dirty? }
- bakfn = fn + ".bak"
- if File.exists? fn
- File.chmod 0600, fn
- FileUtils.mv fn, bakfn, :force => true unless File.exists?(bakfn) && File.size(fn) == 0
- end
- Redwood::save_yaml_obj sources.sort_by { |s| s.id.to_i }, fn, true
- File.chmod 0600, fn
- end
- @sources_dirty = false
- end
- end
+index_name = ENV['SUP_INDEX'] || $config[:index] || DEFAULT_INDEX
+case index_name
+ when "xapian"; require "sup/xapian_index"
+ when "ferret"; require "sup/ferret_index"
+ else fail "unknown index type #{index_name.inspect}"
end
+Index = Redwood.const_get "#{index_name.capitalize}Index"
+debug "using index #{Index.name}"
end
projects / sup / blobdiff