Merge branch '21535-multi-wf-delete'
[arvados.git] / services / api / lib / record_filters.rb
index 831e357b4235b6b11217de47ee2c5960a4ef3563..e51223254f7d21a34b5910e2f68a59abee4c22cb 100644 (file)
@@ -19,17 +19,22 @@ module RecordFilters
   # +model_class+    subclass of ActiveRecord being filtered
   #
   # Output:
-  # Hash with two keys:
+  # 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
+  # :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 = []
 
-    ar_table_name = model_class.table_name
+    model_table_name = model_class.table_name
     filters.each do |filter|
       attrs_in, operator, operand = filter
-      if attrs_in == 'any' && operator != '@@'
+      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
@@ -42,99 +47,127 @@ module RecordFilters
         raise ArgumentError.new("Invalid operator '#{operator}' (#{operator.class}) in filter")
       end
 
+      operator = operator.downcase
       cond_out = []
 
-      if operator == '@@'
-        # Full-text search
-        if attrs_in != 'any'
-          raise ArgumentError.new("Full text search on individual columns is not supported")
-        end
-        if operand.is_a? Array
-          raise ArgumentError.new("Full text search not supported for array operands")
-        end
-
+      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 = []
-        # Use to_tsquery since plainto_tsquery does not support prefix
-        # search. And, split operand and join the words with ' & '
-        cond_out << model_class.full_text_tsvector+" @@ to_tsquery(?)"
-        param_out << operand.split.join(' & ')
       end
+
       attrs.each do |attr|
         subproperty = attr.split(".", 2)
 
-        col = model_class.columns.select { |c| c.name == subproperty[0] }.first
+        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
 
-        if subproperty.length == 2
+        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 '#{subproperty[0]}' for subproperty filter")
+            raise ArgumentError.new("Invalid attribute '#{attr}' for subproperty filter")
           end
 
-          if subproperty[1][0] == "<" and subproperty[1][-1] == ">"
-            subproperty[1] = subproperty[1][1..-2]
+          if proppath[0] == "<" and proppath[-1] == ">"
+            proppath = proppath[1..-2]
           end
 
           # jsonb search
-          case operator.downcase
+          case operator
           when '=', '!='
-            not_in = if operator.downcase == "!=" then "NOT " else "" end
-            cond_out << "#{not_in}(#{ar_table_name}.#{subproperty[0]} @> ?::jsonb)"
-            param_out << SafeJSON.dump({subproperty[1] => operand})
+            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 << "#{ar_table_name}.#{subproperty[0]} @> ?::jsonb"
-                param_out << SafeJSON.dump({subproperty[1] => 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 << "#{ar_table_name}.#{subproperty[0]}->? #{operator} ?::jsonb"
-            param_out << subproperty[1]
+            cond_out << "#{attr_table_name}.#{attr}->? #{operator} ?::jsonb"
+            param_out << proppath
             param_out << SafeJSON.dump(operand)
           when 'like', 'ilike'
-            cond_out << "#{ar_table_name}.#{subproperty[0]}->>? #{operator} ?"
-            param_out << subproperty[1]
+            cond_out << "#{attr_table_name}.#{attr}->>? #{operator} ?"
+            param_out << proppath
             param_out << operand
           when 'not in'
             if operand.is_a? Array
-              cond_out << "#{ar_table_name}.#{subproperty[0]}->>? NOT IN (?) OR #{ar_table_name}.#{subproperty[0]}->>? IS NULL"
-              param_out << subproperty[1]
+              cond_out << "#{attr_table_name}.#{attr}->>? NOT IN (?) OR #{attr_table_name}.#{attr}->>? IS NULL"
+              param_out << proppath
               param_out << operand
-              param_out << subproperty[1]
+              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(#{ar_table_name}.#{subproperty[0]}, ?)"
+              cond_out << "jsonb_exists_inline_op(#{attr_table_name}.#{attr}, ?)"
             elsif operand == false
-              cond_out << "(NOT jsonb_exists(#{ar_table_name}.#{subproperty[0]}, ?)) OR #{ar_table_name}.#{subproperty[0]} is NULL"
+              cond_out << "(NOT jsonb_exists_inline_op(#{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 << subproperty[1]
+            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
-        elsif operator.downcase == "exists"
-          if col.type != :jsonb
-            raise ArgumentError.new("Invalid attribute '#{subproperty[0]}' for operator '#{operator}' in filter")
+        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(#{ar_table_name}.#{subproperty[0]}, ?)"
+          cond_out << "jsonb_exists_inline_op(#{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
+          cond_out << "#{attr1} #{operator} #{attr2}"
         else
-          if !model_class.searchable_columns(operator).index subproperty[0]
-            raise ArgumentError.new("Invalid attribute '#{subproperty[0]}' in filter")
+          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
+          attr_type = attr_model_class.attribute_column(attr).type
 
-          case operator.downcase
+          case operator
           when '=', '<', '<=', '>', '>=', '!=', 'like', 'ilike'
-            attr_type = model_class.attribute_column(attr).type
             operator = '<>' if operator == '!='
             if operand.is_a? String
               if attr_type == :boolean
@@ -148,15 +181,15 @@ module RecordFilters
                 when '0', 'f', 'false', 'n', 'no'
                   operand = false
                 else
-                  raise ArgumentError("Invalid operand '#{operand}' for " \
-                                      "boolean attribute '#{attr}'")
+                  raise ArgumentError.new("Invalid operand '#{operand}' for " \
+                                          "boolean attribute '#{attr}'")
                 end
               end
               if operator == '<>'
                 # explicitly allow NULL
-                cond_out << "#{ar_table_name}.#{attr} #{operator} ? OR #{ar_table_name}.#{attr} IS NULL"
+                cond_out << "#{attr_table_name}.#{attr} #{operator} ? OR #{attr_table_name}.#{attr} IS NULL"
               else
-                cond_out << "#{ar_table_name}.#{attr} #{operator} ?"
+                cond_out << "#{attr_table_name}.#{attr} #{operator} ?"
               end
               if (# any operator that operates on value rather than
                 # representation:
@@ -165,32 +198,43 @@ module RecordFilters
               end
               param_out << operand
             elsif operand.nil? and operator == '='
-              cond_out << "#{ar_table_name}.#{attr} is null"
+              cond_out << "#{attr_table_name}.#{attr} is null"
             elsif operand.nil? and operator == '<>'
-              cond_out << "#{ar_table_name}.#{attr} is not null"
+              cond_out << "#{attr_table_name}.#{attr} is not null"
             elsif (attr_type == :boolean) and ['=', '<>'].include?(operator) and
                  [true, false].include?(operand)
-              cond_out << "#{ar_table_name}.#{attr} #{operator} ?"
+              cond_out << "#{attr_table_name}.#{attr} #{operator} ?"
               param_out << operand
             elsif (attr_type == :integer)
-              cond_out << "#{ar_table_name}.#{attr} #{operator} ?"
+              if !operand.is_a?(Integer) || operand.bit_length > 64
+                raise ArgumentError.new("Invalid operand '#{operand}' "\
+                                        "for integer attribute '#{attr}'")
+              end
+              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 << "#{ar_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 #{ar_table_name}.#{attr} IS NULL)"
-              end
-            else
+            if !operand.is_a? Array
               raise ArgumentError.new("Invalid operand type '#{operand.class}' "\
                                       "for '#{operator}' operator in filters")
             end
+            if attr_type == :integer
+              operand.each do |el|
+                if !el.is_a?(Integer) || el.bit_length > 64
+                  raise ArgumentError.new("Invalid element '#{el}' in array "\
+                                          "for integer attribute '#{attr}'")
+                end
+              end
+            end
+            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
           when 'is_a'
             operand = [operand] unless operand.is_a? Array
             cond = []
@@ -198,14 +242,14 @@ module RecordFilters
               cl = ArvadosModel::kind_class op
               if cl
                 if attr == 'uuid'
-                  if model_class.uuid_prefix == cl.uuid_prefix
+                  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(#{ar_table_name}.#{attr}, 7, 5) = ?"
+                  cond << "substring(#{attr_table_name}.#{attr}, 7, 5) = ?"
                   param_out << cl.uuid_prefix
                 end
               else
@@ -213,6 +257,31 @@ module RecordFilters
               end
             end
             cond_out << cond.join(' OR ')
+          when 'contains'
+            if col.andand.type != :jsonb
+              raise ArgumentError.new("Invalid attribute '#{attr}' for '#{operator}' operator")
+            end
+            if operand == []
+              raise ArgumentError.new("Invalid operand '#{operand.inspect}' for '#{operator}' operator")
+            end
+            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_inline_op(a,b) instead of "a ?&
+            # b" because the pg gem thinks "?" is a bind var.
+            #
+            # See note in migration
+            # 20230815160000_jsonb_exists_functions about _inline_op
+            # functions.
+            #
+            # 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_inline_op(#{attr_table_name}.#{attr}, array[#{q}])"
           else
             raise ArgumentError.new("Invalid operator '#{operator}'")
           end
@@ -221,7 +290,7 @@ module RecordFilters
       conds_out << cond_out.join(' OR ') if cond_out.any?
     end
 
-    {:cond_out => conds_out, :param_out => param_out}
+    {:cond_out => conds_out, :param_out => param_out, :joins => joins}
   end
 
 end