Merge branch '19972-go-client-retry'
[arvados.git] / services / api / lib / record_filters.rb
index fab16bbc1d3585f749ab2f69f4c94fd46dd0f7d4..65c25810acf2e3ef98422a1de3a5c8503e75edfe 100644 (file)
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+# Mixin module providing a method to convert filters into a list of SQL
+# fragments suitable to be fed to ActiveRecord #where.
+#
 # Expects:
-#   @where
-#   @filters
-#   +model_class+
+#   model_class
 # Operates on:
 #   @objects
+
+require 'safe_json'
+
 module RecordFilters
 
-  def apply_where_limit_order_params
-    if @filters.is_a? Array and @filters.any?
+  # Input:
+  # +filters+        array of conditions, each being [column, operator, operand]
+  # +model_class+    subclass of ActiveRecord being filtered
+  #
+  # Output:
+  # Hash with the following keys:
+  # :cond_out  array of SQL fragments for each filter expression
+  # :param_out array of values for parameter substitution in cond_out
+  # :joins     array of joins: either [] or ["JOIN containers ON ..."]
+  def record_filters filters, model_class
+    conds_out = []
+    param_out = []
+    joins = []
+
+    model_table_name = model_class.table_name
+    filters.each do |filter|
+      attrs_in, operator, operand = filter
+      if operator == '@@'
+        raise ArgumentError.new("Full text search operator is no longer supported")
+      end
+      if attrs_in == 'any'
+        attrs = model_class.searchable_columns(operator)
+      elsif attrs_in.is_a? Array
+        attrs = attrs_in
+      else
+        attrs = [attrs_in]
+      end
+      if !filter.is_a? Array
+        raise ArgumentError.new("Invalid element in filters array: #{filter.inspect} is not an array")
+      elsif !operator.is_a? String
+        raise ArgumentError.new("Invalid operator '#{operator}' (#{operator.class}) in filter")
+      end
+
+      operator = operator.downcase
       cond_out = []
-      param_out = []
-      @filters.each do |attr, operator, operand|
-        if !model_class.searchable_columns(operator).index attr.to_s
-          raise ArgumentError.new("Invalid attribute '#{attr}' in condition")
+
+      if attrs_in == 'any' && (operator == 'ilike' || operator == 'like') && (operand.is_a? String) && operand.match('^[%].*[%]$')
+        # Trigram index search
+        cond_out << model_class.full_text_trgm + " #{operator} ?"
+        param_out << operand
+        # Skip the generic per-column operator loop below
+        attrs = []
+      end
+
+      attrs.each do |attr|
+        subproperty = attr.split(".", 2)
+
+        if subproperty.length == 2 && subproperty[0] == 'container' && model_table_name == "container_requests"
+          # attr is "tablename.colname" -- e.g., ["container.state", "=", "Complete"]
+          joins = ["JOIN containers ON container_requests.container_uuid = containers.uuid"]
+          attr_model_class = Container
+          attr_table_name = "containers"
+          subproperty = subproperty[1].split(".", 2)
+        else
+          attr_model_class = model_class
+          attr_table_name = model_table_name
         end
-        case operator.downcase
-        when '=', '<', '<=', '>', '>=', 'like'
-          if operand.is_a? String
-            cond_out << "#{table_name}.#{attr} #{operator} ?"
-            if (# any operator that operates on value rather than
-                # representation:
-                operator.match(/[<=>]/) and
-                model_class.attribute_column(attr).type == :datetime)
-              operand = Time.parse operand
+
+        attr = subproperty[0]
+        proppath = subproperty[1]
+        col = attr_model_class.columns.select { |c| c.name == attr }.first
+
+        if proppath
+          if col.nil? or col.type != :jsonb
+            raise ArgumentError.new("Invalid attribute '#{attr}' for subproperty filter")
+          end
+
+          if proppath[0] == "<" and proppath[-1] == ">"
+            proppath = proppath[1..-2]
+          end
+
+          # jsonb search
+          case operator
+          when '=', '!='
+            not_in = if operator == "!=" then "NOT " else "" end
+            cond_out << "#{not_in}(#{attr_table_name}.#{attr} @> ?::jsonb)"
+            param_out << SafeJSON.dump({proppath => operand})
+          when 'in'
+            if operand.is_a? Array
+              operand.each do |opr|
+                cond_out << "#{attr_table_name}.#{attr} @> ?::jsonb"
+                param_out << SafeJSON.dump({proppath => opr})
+              end
+            else
+              raise ArgumentError.new("Invalid operand type '#{operand.class}' "\
+                                      "for '#{operator}' operator in filters")
             end
+          when '<', '<=', '>', '>='
+            cond_out << "#{attr_table_name}.#{attr}->? #{operator} ?::jsonb"
+            param_out << proppath
+            param_out << SafeJSON.dump(operand)
+          when 'like', 'ilike'
+            cond_out << "#{attr_table_name}.#{attr}->>? #{operator} ?"
+            param_out << proppath
             param_out << operand
+          when 'not in'
+            if operand.is_a? Array
+              cond_out << "#{attr_table_name}.#{attr}->>? NOT IN (?) OR #{attr_table_name}.#{attr}->>? IS NULL"
+              param_out << proppath
+              param_out << operand
+              param_out << proppath
+            else
+              raise ArgumentError.new("Invalid operand type '#{operand.class}' "\
+                                      "for '#{operator}' operator in filters")
+            end
+          when 'exists'
+            if operand == true
+              cond_out << "jsonb_exists(#{attr_table_name}.#{attr}, ?)"
+            elsif operand == false
+              cond_out << "(NOT jsonb_exists(#{attr_table_name}.#{attr}, ?)) OR #{attr_table_name}.#{attr} is NULL"
+            else
+              raise ArgumentError.new("Invalid operand '#{operand}' for '#{operator}' must be true or false")
+            end
+            param_out << proppath
+          when 'contains'
+            cond_out << "#{attr_table_name}.#{attr} @> ?::jsonb OR #{attr_table_name}.#{attr} @> ?::jsonb"
+            param_out << SafeJSON.dump({proppath => operand})
+            param_out << SafeJSON.dump({proppath => [operand]})
+          else
+            raise ArgumentError.new("Invalid operator for subproperty search '#{operator}'")
           end
-        when 'in'
-          if operand.is_a? Array
-            cond_out << "#{table_name}.#{attr} IN (?)"
-            param_out << operand
+        elsif operator == "exists"
+          if col.nil? or col.type != :jsonb
+            raise ArgumentError.new("Invalid attribute '#{attr}' for operator '#{operator}' in filter")
+          end
+
+          cond_out << "jsonb_exists(#{attr_table_name}.#{attr}, ?)"
+          param_out << operand
+        elsif expr = /^ *\( *(\w+) *(<=?|>=?|=) *(\w+) *\) *$/.match(attr)
+          if operator != '=' || ![true,"true"].index(operand)
+            raise ArgumentError.new("Invalid expression filter '#{attr}': subsequent elements must be [\"=\", true]")
+          end
+          operator = expr[2]
+          attr1, attr2 = expr[1], expr[3]
+          allowed = attr_model_class.searchable_columns(operator)
+          [attr1, attr2].each do |tok|
+            if !allowed.index(tok)
+              raise ArgumentError.new("Invalid attribute in expression: '#{tok}'")
+            end
+            col = attr_model_class.columns.select { |c| c.name == tok }.first
+            if col.type != :integer
+              raise ArgumentError.new("Non-numeric attribute in expression: '#{tok}'")
+            end
           end
-        when 'is_a'
-          operand = [operand] unless operand.is_a? Array
-          cond = []
-          operand.each do |op|
+          cond_out << "#{attr1} #{operator} #{attr2}"
+        else
+          if !attr_model_class.searchable_columns(operator).index(attr) &&
+             !(col.andand.type == :jsonb && ['contains', '=', '<>', '!='].index(operator))
+            raise ArgumentError.new("Invalid attribute '#{attr}' in filter")
+          end
+
+          case operator
+          when '=', '<', '<=', '>', '>=', '!=', 'like', 'ilike'
+            attr_type = attr_model_class.attribute_column(attr).type
+            operator = '<>' if operator == '!='
+            if operand.is_a? String
+              if attr_type == :boolean
+                if not ['=', '<>'].include?(operator)
+                  raise ArgumentError.new("Invalid operator '#{operator}' for " \
+                                          "boolean attribute '#{attr}'")
+                end
+                case operand.downcase
+                when '1', 't', 'true', 'y', 'yes'
+                  operand = true
+                when '0', 'f', 'false', 'n', 'no'
+                  operand = false
+                else
+                  raise ArgumentError("Invalid operand '#{operand}' for " \
+                                      "boolean attribute '#{attr}'")
+                end
+              end
+              if operator == '<>'
+                # explicitly allow NULL
+                cond_out << "#{attr_table_name}.#{attr} #{operator} ? OR #{attr_table_name}.#{attr} IS NULL"
+              else
+                cond_out << "#{attr_table_name}.#{attr} #{operator} ?"
+              end
+              if (# any operator that operates on value rather than
+                # representation:
+                operator.match(/[<=>]/) and (attr_type == :datetime))
+                operand = Time.parse operand
+              end
+              param_out << operand
+            elsif operand.nil? and operator == '='
+              cond_out << "#{attr_table_name}.#{attr} is null"
+            elsif operand.nil? and operator == '<>'
+              cond_out << "#{attr_table_name}.#{attr} is not null"
+            elsif (attr_type == :boolean) and ['=', '<>'].include?(operator) and
+                 [true, false].include?(operand)
+              cond_out << "#{attr_table_name}.#{attr} #{operator} ?"
+              param_out << operand
+            elsif (attr_type == :integer)
+              cond_out << "#{attr_table_name}.#{attr} #{operator} ?"
+              param_out << operand
+            else
+              raise ArgumentError.new("Invalid operand type '#{operand.class}' "\
+                                      "for '#{operator}' operator in filters")
+            end
+          when 'in', 'not in'
+            if operand.is_a? Array
+              cond_out << "#{attr_table_name}.#{attr} #{operator} (?)"
+              param_out << operand
+              if operator == 'not in' and not operand.include?(nil)
+                # explicitly allow NULL
+                cond_out[-1] = "(#{cond_out[-1]} OR #{attr_table_name}.#{attr} IS NULL)"
+              end
+            else
+              raise ArgumentError.new("Invalid operand type '#{operand.class}' "\
+                                      "for '#{operator}' operator in filters")
+            end
+          when 'is_a'
+            operand = [operand] unless operand.is_a? Array
+            cond = []
+            operand.each do |op|
               cl = ArvadosModel::kind_class op
               if cl
-                cond << "#{table_name}.#{attr} like ?"
-                param_out << cl.uuid_like_pattern
+                if attr == 'uuid'
+                  if attr_model_class.uuid_prefix == cl.uuid_prefix
+                    cond << "1=1"
+                  else
+                    cond << "1=0"
+                  end
+                else
+                  # Use a substring query to support remote uuids
+                  cond << "substring(#{attr_table_name}.#{attr}, 7, 5) = ?"
+                  param_out << cl.uuid_prefix
+                end
               else
                 cond << "1=0"
               end
-          end
-          cond_out << cond.join(' OR ')
-        end
-      end
-      if cond_out.any?
-        @objects = @objects.where(cond_out.join(' AND '), *param_out)
-      end
-    end
-    if @where.is_a? Hash and @where.any?
-      conditions = ['1=1']
-      @where.each do |attr,value|
-        if attr.to_s == 'any'
-          if value.is_a?(Array) and
-              value.length == 2 and
-              value[0] == 'contains' then
-            ilikes = []
-            model_class.searchable_columns('ilike').each do |column|
-              ilikes << "#{table_name}.#{column} ilike ?"
-              conditions << "%#{value[1]}%"
             end
-            if ilikes.any?
-              conditions[0] << ' and (' + ilikes.join(' or ') + ')'
+            cond_out << cond.join(' OR ')
+          when 'contains'
+            if col.andand.type != :jsonb
+              raise ArgumentError.new("Invalid attribute '#{attr}' for '#{operator}' operator")
             end
-          end
-        elsif attr.to_s.match(/^[a-z][_a-z0-9]+$/) and
-            model_class.columns.collect(&:name).index(attr.to_s)
-          if value.nil?
-            conditions[0] << " and #{table_name}.#{attr} is ?"
-            conditions << nil
-          elsif value.is_a? Array
-            if value[0] == 'contains' and value.length == 2
-              conditions[0] << " and #{table_name}.#{attr} like ?"
-              conditions << "%#{value[1]}%"
-            else
-              conditions[0] << " and #{table_name}.#{attr} in (?)"
-              conditions << value
+            if operand == []
+              raise ArgumentError.new("Invalid operand '#{operand.inspect}' for '#{operator}' operator")
             end
-          elsif value.is_a? String or value.is_a? Fixnum or value == true or value == false
-            conditions[0] << " and #{table_name}.#{attr}=?"
-            conditions << value
-          elsif value.is_a? Hash
-            # Not quite the same thing as "equal?" but better than nothing?
-            value.each do |k,v|
-              if v.is_a? String
-                conditions[0] << " and #{table_name}.#{attr} ilike ?"
-                conditions << "%#{k}%#{v}%"
+            operand = [operand] unless operand.is_a? Array
+            operand.each do |op|
+              if !op.is_a?(String)
+                raise ArgumentError.new("Invalid element #{operand.inspect} in operand for #{operator.inspect} operator (operand must be a string or array of strings)")
               end
             end
+            # We use jsonb_exists_all(a,b) instead of "a ?& b" because
+            # the pg gem thinks "?" is a bind var. And we use string
+            # interpolation instead of param_out because the pg gem
+            # flattens param_out / doesn't support passing arrays as
+            # bind vars.
+            q = operand.map { |s| ActiveRecord::Base.connection.quote(s) }.join(',')
+            cond_out << "jsonb_exists_all(#{attr_table_name}.#{attr}, array[#{q}])"
+          else
+            raise ArgumentError.new("Invalid operator '#{operator}'")
           end
         end
       end
-      if conditions.length > 1
-        conditions[0].sub!(/^1=1 and /, '')
-        @objects = @objects.
-          where(*conditions)
-      end
+      conds_out << cond_out.join(' OR ') if cond_out.any?
     end
 
-    if params[:limit]
-      begin
-        @limit = params[:limit].to_i
-      rescue
-        raise ArgumentError.new("Invalid value for limit parameter")
-      end
-    else
-      @limit = 100
-    end
-    @objects = @objects.limit(@limit)
-
-    orders = []
-
-    if params[:offset]
-      begin
-        @objects = @objects.offset(params[:offset].to_i)
-        @offset = params[:offset].to_i
-      rescue
-        raise ArgumentError.new("Invalid value for limit parameter")
-      end
-    else
-      @offset = 0
-    end
-
-    orders = []
-    if params[:order]
-      params[:order].split(',').each do |order|
-        attr, direction = order.strip.split " "
-        direction ||= 'asc'
-        if attr.match /^[a-z][_a-z0-9]+$/ and
-            model_class.columns.collect(&:name).index(attr) and
-            ['asc','desc'].index direction.downcase
-          orders << "#{table_name}.#{attr} #{direction.downcase}"
-        end
-      end
-    end
-    if orders.empty?
-      orders << "#{table_name}.modified_at desc"
-    end
-    @objects = @objects.order(orders.join ", ")
+    {:cond_out => conds_out, :param_out => param_out, :joins => joins}
   end
 
 end