#-- encoding: UTF-8 #-- copyright # OpenProject is a project management system. # Copyright (C) 2012-2017 the OpenProject Foundation (OPF) # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License version 3. # # OpenProject is a fork of ChiliProject, which is a fork of Redmine. The copyright follows: # Copyright (C) 2006-2017 Jean-Philippe Lang # Copyright (C) 2010-2013 the ChiliProject Team # # This program is free software; you can redistribute it and/or # modify it under the terms of the GNU General Public License # as published by the Free Software Foundation; either version 2 # of the License, or (at your option) any later version. # # This program is distributed in the hope that it will be useful, # but WITHOUT ANY WARRANTY; without even the implied warranty of # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the # GNU General Public License for more details. # # You should have received a copy of the GNU General Public License # along with this program; if not, write to the Free Software # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. # # See doc/COPYRIGHT.rdoc for more details. #++ # Similar to regular Journals, but under the following circumstances journals are aggregated: # * they are in temporal proximity # * they belong to the same resource # * they were created by the same user (i.e. the same user edited the journable) # * no other user has an own journal on the same object between the aggregated ones # When a user commented (added a note) twice within a short time, the second comment will # "open" a new aggregation, since we do not want to merge comments in any way. # The term "aggregation" means the following when applied to our journaling: # * ignore/hide old journal rows (since every journal row contains a full copy of the journaled # object, dropping intermediate rows will just increase the diff of the following journal) # * in case an older row had notes, take the notes from the older row, since they shall not # be dropped class Journal::AggregatedJournal class << self # Returns the aggregated journal that contains the specified (vanilla/pure) journal. def for_journal(pure_journal) raw = Journal::AggregatedJournal.query_aggregated_journals(journable: pure_journal.journable) .where("#{version_projection} >= ?", pure_journal.version) .first raw ? Journal::AggregatedJournal.new(raw) : nil end # Returns the aggregated journal that contains the vanilla/pure journal with the specified id. def with_notes_id(notes_id) # We need to limit the journal aggregation as soon as possible for performance reasons. # Therefore we have to provide the notes_id to the aggregation on top of it being used # in the where clause to pick the desired AggregatedJournal. raw_journal = query_aggregated_journals(journal_id: notes_id) .where("#{table_name}.id = #{notes_id}") .first raw_journal ? Journal::AggregatedJournal.new(raw_journal) : nil end ## # The +journable+ parameter allows to filter for aggregated journals of a given journable. # # The +until_version+ parameter can be used in conjunction with the +journable+ parameter # to see the aggregated journals as if no versions were known after the specified version. def aggregated_journals(journable: nil, until_version: nil, includes: []) raw_journals = query_aggregated_journals(journable: journable, until_version: until_version) predecessors = {} raw_journals.each do |journal| journable_key = [journal.journable_type, journal.journable_id] predecessors[journable_key] = [nil] unless predecessors[journable_key] predecessors[journable_key] << journal end aggregated_journals = raw_journals.map { |journal| journable_key = [journal.journable_type, journal.journable_id] Journal::AggregatedJournal.new(journal, predecessor: predecessors[journable_key].shift) } preload_associations(journable, aggregated_journals, includes) aggregated_journals end def query_aggregated_journals(journable: nil, until_version: nil, journal_id: nil) # Using the roughly aggregated groups from :sql_rough_group we need to merge journals # where an entry with empty notes follows an entry containing notes, so that the notes # from the main entry are taken, while the remaining information is taken from the # more recent entry. We therefore join the rough groups with itself # _wherever a merge would be valid_. # Since the results are already pre-merged, this can only happen if Our first entry (master) # had a comment and its successor (addition) had no comment, but can be merged. # This alone would, however, leave the addition in the result set, leaving a "no change" # journal entry back. By an additional self-join towards the predecessor, we can make sure # that our own row (master) would not already have been merged by its predecessor. If it is # (that means if we can find a valid predecessor), we drop our current row, because it will # already be present (in a merged form) in the row of our predecessor. Journal.from("(#{sql_rough_group(1, journable, until_version, journal_id)}) #{table_name}") .joins("LEFT OUTER JOIN (#{sql_rough_group(2, journable, until_version, journal_id)}) addition ON #{sql_on_groups_belong_condition(table_name, 'addition')}") .joins("LEFT OUTER JOIN (#{sql_rough_group(3, journable, until_version, journal_id)}) predecessor ON #{sql_on_groups_belong_condition('predecessor', table_name)}") .where('predecessor.id IS NULL') .order("COALESCE(addition.created_at, #{table_name}.created_at) ASC") .order("#{version_projection} ASC") .select("#{table_name}.journable_id, #{table_name}.journable_type, #{table_name}.user_id, #{table_name}.notes, #{table_name}.id \"notes_id\", #{table_name}.version \"notes_version\", #{table_name}.activity_type, COALESCE(addition.created_at, #{table_name}.created_at) \"created_at\", COALESCE(addition.id, #{table_name}.id) \"id\", #{version_projection} \"version\"") end # Returns whether "notification-hiding" should be assumed for the given journal pair. # This leads to an aggregated journal effectively blocking notifications of an earlier journal, # because it "steals" the addition from its predecessor. See the specs section under # "mail suppressing aggregation" (for EnqueueWorkPackageNotificationJob) for more details def hides_notifications?(successor, predecessor) return false unless successor && predecessor timeout = Setting.journal_aggregation_time_minutes.to_i.minutes if successor.journable_type != predecessor.journable_type || successor.journable_id != predecessor.journable_id || successor.user_id != predecessor.user_id || (successor.created_at - predecessor.created_at) <= timeout return false end # imaginary state in which the successor never existed # if this makes the predecessor disappear, the successor must have taken journals # from it (that now became part of the predecessor again). !Journal::AggregatedJournal .query_aggregated_journals( journable: successor.journable, until_version: successor.version - 1) .where("#{version_projection} = #{predecessor.version}") .exists? end def table_name Journal.table_name end def version_projection "COALESCE(addition.version, #{table_name}.version)" end private # Provides a full SQL statement that returns journals that are aggregated on a basic level: # * a row is dropped as soon as its successor is eligible to be merged with it # * rows with a comment are never dropped (we _might_ need the comment later) # Thereby the result already has aggregation performed, but will still have too many rows: # Changes without notes after changes containing notes (even if both were performed by # the same user). Those need to be filtered out later. # To be able to self-join results of this statement, we add an additional column called # "group_number" to the result. This allows to compare a group resulting from this query with # its predecessor and successor. def sql_rough_group(uid, journable, until_version, journal_id) if until_version && !journable raise 'need to provide a journable, when specifying a version limit' elsif journable && journable.id.nil? raise 'journable has no id' end conditions = additional_conditions(journable, until_version, journal_id) "SELECT predecessor.*, #{sql_group_counter(uid)} AS group_number FROM #{sql_rough_group_from_clause(uid)} #{sql_rough_group_join(conditions[:join_conditions])} #{sql_rough_group_where(conditions[:where_conditions])}" end def additional_conditions(journable, until_version, journal_id) where_conditions = '' join_conditions = '' if journable where_conditions += " AND predecessor.journable_type = '#{journable.class.name}' AND predecessor.journable_id = #{journable.id}" if until_version where_conditions += " AND predecessor.version <= #{until_version}" join_conditions += "AND successor.version <= #{until_version}" end end if journal_id where_conditions += "AND predecessor.id IN ( SELECT id_key.id FROM #{table_name} id_key JOIN #{table_name} journable_key ON id_key.journable_id = journable_key.journable_id AND id_key.journable_type = journable_key.journable_type AND journable_key.id = #{journal_id})" end { where_conditions: where_conditions, join_conditions: join_conditions } end def sql_rough_group_join(additional_conditions) "LEFT OUTER JOIN #{table_name} successor ON predecessor.version + 1 = successor.version AND predecessor.journable_type = successor.journable_type AND predecessor.journable_id = successor.journable_id #{additional_conditions}" end def sql_rough_group_where(additional_conditions) "WHERE (predecessor.user_id != successor.user_id OR (predecessor.notes != '' AND predecessor.notes IS NOT NULL) OR #{sql_beyond_aggregation_time?('predecessor', 'successor')} OR successor.id IS NULL) #{additional_conditions}" end # The "group_number" required in :sql_rough_group has to be generated differently depending on # the DBMS used. This method returns the appropriate statement to be used inside a SELECT to # obtain the current group number. # The :uid parameter allows to define non-conflicting variable names (for MySQL). def sql_group_counter(uid) if OpenProject::Database.mysql? group_counter = mysql_group_count_variable(uid) "(#{group_counter} := #{group_counter} + 1)" else 'row_number() OVER (ORDER BY predecessor.version ASC)' end end # MySQL requires some initialization to be performed before being able to count the groups. # This method allows to inject further FROM sources to achieve that in a single SQL statement. # Sadly MySQL requires the whole statement to be wrapped in parenthesis, while PostgreSQL # prohibits that. def sql_rough_group_from_clause(uid) if OpenProject::Database.mysql? "(journals predecessor, (SELECT #{mysql_group_count_variable(uid)}:=0) number_initializer)" else 'journals predecessor' end end def mysql_group_count_variable(uid) "@aggregated_journal_row_counter_#{uid}" end # Similar to the WHERE statement used in :sql_rough_group. However, this condition will # match (return true) for all pairs where a merge/aggregation IS possible. def sql_on_groups_belong_condition(predecessor, successor) "#{predecessor}.group_number + 1 = #{successor}.group_number AND (NOT #{sql_beyond_aggregation_time?(predecessor, successor)} AND #{predecessor}.user_id = #{successor}.user_id AND #{successor}.journable_type = #{predecessor}.journable_type AND #{successor}.journable_id = #{predecessor}.journable_id AND NOT ((#{predecessor}.notes != '' AND #{predecessor}.notes IS NOT NULL) AND (#{successor}.notes != '' AND #{successor}.notes IS NOT NULL)))" end # Returns a SQL condition that will determine whether two entries are too far apart (temporal) # to be considered for aggregation. This takes the current instance settings for temporal # proximity into account. def sql_beyond_aggregation_time?(predecessor, successor) aggregation_time_seconds = Setting.journal_aggregation_time_minutes.to_i.minutes if aggregation_time_seconds == 0 # if aggregation is disabled, we consider everything to be beyond aggregation time # even if creation dates are exactly equal return '(true = true)' end if OpenProject::Database.mysql? difference = "TIMESTAMPDIFF(second, #{predecessor}.created_at, #{successor}.created_at)" threshold = aggregation_time_seconds else difference = "(#{successor}.created_at - #{predecessor}.created_at)" threshold = "interval '#{aggregation_time_seconds} second'" end "(#{difference} > #{threshold})" end def preload_associations(journable, aggregated_journals, includes) return unless includes.length > 1 journal_ids = aggregated_journals.map(&:id) customizable_journals = if includes.include?(:customizable_journals) Journal::CustomizableJournal .where(journal_id: journal_ids) .all .group_by(&:journal_id) end attachable_journals = if includes.include?(:customizable_journals) Journal::AttachableJournal .where(journal_id: journal_ids) .all .group_by(&:journal_id) end data = if includes.include?(:data) "Journal::#{journable.class}Journal".constantize .where(journal_id: journal_ids) .all .group_by(&:journal_id) end aggregated_journals.each { |journal| if includes.include?(:customizable_journals) journal.set_preloaded_customizable_journals customizable_journals[journal.id] end if includes.include?(:attachable_journals) journal.set_preloaded_attachable_journals attachable_journals[journal.id] end if includes.include?(:data) journal.set_preloaded_data data[journal.id].first end if journable journal.set_preloaded_journable journable end } end end include JournalChanges include JournalFormatter include Redmine::Acts::Journalized::FormatHooks register_journal_formatter :diff, OpenProject::JournalFormatter::Diff register_journal_formatter :attachment, OpenProject::JournalFormatter::Attachment register_journal_formatter :custom_field, OpenProject::JournalFormatter::CustomField alias_method :details, :get_changes delegate :journable_type, :journable_id, :journable, :journable=, :user_id, :user, :notes, :notes?, :activity_type, :created_at, :id, :version, :attributes, :attachable_journals, :attachable_journals=, :customizable_journals, :customizable_journals=, :editable_by?, :notes_id, :notes_version, :data, :data=, to: :journal # Initializes a new AggregatedJournal. Allows to explicitly set a predecessor, if it is already # known. Providing a predecessor is only to improve efficiency, it is not required. # In case the predecessor is not known, it will be lazily retrieved. def initialize(journal, predecessor: false) @journal = journal # explicitly checking false to allow passing nil as "no predecessor" # mind that we check @predecessor with defined? below, so don't assign to it in all cases! unless predecessor == false @predecessor = predecessor end end # returns an instance of this class that is reloaded from the database def reloaded self.class.with_notes_id(notes_id) end def user @user ||= User.find(user_id) end def predecessor unless defined? @predecessor raw_journal = self.class.query_aggregated_journals(journable: journable) .where("#{self.class.version_projection} < ?", version) .except(:order) .order("#{self.class.version_projection} DESC") .first @predecessor = raw_journal ? Journal::AggregatedJournal.new(raw_journal) : nil end @predecessor end def successor unless defined? @successor raw_journal = self.class.query_aggregated_journals(journable: journable) .where("#{self.class.version_projection} > ?", version) .except(:order) .order("#{self.class.version_projection} ASC") .first @successor = raw_journal ? Journal::AggregatedJournal.new(raw_journal) : nil end @successor end def set_preloaded_customizable_journals(loaded_journals) self.customizable_journals = loaded_journals if loaded_journals customizable_journals.proxy_association.loaded! end def set_preloaded_attachable_journals(loaded_journals) self.attachable_journals = loaded_journals if loaded_journals attachable_journals.proxy_association.loaded! end def set_preloaded_data(loaded_data) self.data = loaded_data end def set_preloaded_journable(loaded_journable) self.journable = loaded_journable journal.association(:journable).loaded! end def initial? predecessor.nil? end private attr_reader :journal end