Skip to content

Latest commit

 

History

History
93 lines (74 loc) · 4.57 KB

TODO.md

File metadata and controls

93 lines (74 loc) · 4.57 KB

TODO

Bug

  • [-] Ensure columns referenced by all strategies projected

  • Configuration::Rank#calculation is not applied

  • Application global config is not applied if option is defined in Configuration::Base.defaults. Application defaults should be applied to Configuration::Scopes before PgMultisearch defaults

Incomplete

  • Index::Scopes#with_rank_gt incomplete

Revise

  • Don't reuse Strategies::Tsquery, it should be okay to rebuld. We can however cache the parsed text input against a hash of the input to partially reducing the cost. In general we should try to limit coupling; the queries, filters and ranks should be independent of one another, the Index::Relation binds them together.
  • Filter and Rank tsquery should not be reused, they must remain independent ie. it should be possible to apply weights when filtering but rank without the condition
  • Query shouldn't be coupled to filter as rank may utilise columns other than those referenced by filter.
  • Trigram should be ordered by lexeme position; Use unnest to order lexemes, currently pg_multisearch_words are in what seems to be alphabetical order. Unnest returns their actual position within the original document. Only available within 9.5 and up.
  • Refactor (Count query builder)[pg_multisearch/lib/pg_multisearch/index/relation/count.rb]
    • Store query fragments as curried Procs on the relation
  • Handle strategies
    • document yield params for Configuration::Rank#calculation. Yield all strategies and the ast to the Rank::Base#calc block. Otherwise they should be able to write the arel or sequel that they know.
    • Threshold should be capable of filtering up to 3 strategies. It compares 2 currently.
  • Decouple tsheadline from tsearch filter
  • Find a way to cast bound variable of type array when executing prepared statement. See Filter::Base#by_type
  • Refactor bound variable tracking and application to prepared statements
  • ts_rank({weights}::float4[], ...) and ts_rank_cd should accept weight scale arg
  • Search.call should accept and, if given, overload configuration at runtime
  • Clarify Index::Relation identity; is it possible to cache against runtime options?
  • Ony select pk/fk when performing a JOIN. Since the query uses a CTE this is unnecessary.
  • Ensure aggregate index utilised when searchable_type present in WHERE clause. It is.
  • Document::Base should be read only

Improvement

  • Add ts_headline content field to data::jsonb; runtime concatenation introduces unnecessary complexity and is slower.
  • Use exec_update rather than execute, or call on the PGResult#clear
  • Supported PostgreSQL versions should be able to execute multi insert on rebuild
  • exec ANALYZE on table after bulk update; stats should be refreshed
  • plugin interface and documentation
  • adapter interface, decouple from Rails and ActiveRecor
  • generator flexibility. The user should be able to customise the table schema
    • Add --strategies option or use existing plugins option to build only required columns and indices
    • Add --index option allowing the user to subclass

Compatibility

  • Refactor railtie re/load hooks

    • Ensure compatibility with Rails 4/5

  • Remove Oj, use ActiveSupport::JSON or MultiJSON

  • Verify Kaminari and WillPaginate integration Document Loader

  • Compatibility ActiveRecord < 5

    • Add JSONB OID
    • Add searchable enum OID
      module ActiveRecord::ConnectionAdapters::PostgreSQL
    module OID
      class Jsonb < Type::Json
        def type
          :jsonb
        end
      end

      class Enum < Type::Value # :nodoc:
        def type
          :enum
        end

        private

          def cast_value(value)
            value.to_s
          end
      end
    end

    NATIVE_DATABASE_TYPES[:jsonb] = { name: 'jsonb' }

    OID.register_type('jsonb', OID::Jsonb)
    # OID.register_type('searchable', OID::Enum)
  end