News - 5.x

Release 5.1.2 - 2016-01-29


  • Improved performance for sequential search against constant value such as true and 29.

  • Improved performance for sequential search against binary operation with constant value such as x == 29 and x < 29.

  • [select] Changed score type to Float from Int32 when Command version is 2. The current default command version is 1. Command version 2 is experimental. So this change isn’t affected to normal users. [GitHub#450][Patch by Naoya Murakami]

  • [grn_ts] Supported match operator.

  • [grndb] Added --target option to check command. It reduces check target.

  • [Windows] Updated bundled msgpack to 1.3.0 from 1.0.1.

  • [Windows] Updated bundled MeCab to 0.996 from 0.98.

  • [grn_hash_size()] Added a new API that returns the number of records in the hash table.

  • [GRN_HASH_TINY] Added a new flag to create tiny hash table.

  • [grn_dump_table_create_flags()] Added a new API that converts flags value of a table to flags parameter format of table_create.

  • [grn_dump_column_create_flags()] Added a new API that converts flags value of a column to flags parameter format of column_create.

  • [grn_plugin_get_names()] Added a new API that returns all plugin names in a database.

  • [grn_column_get_all_index_data()] Added a new API that returns all index data for a table or data column.

  • [schema] Added indexes information to indexed tables and data columns.

  • [grn_config_get()] Renamed from grn_conf_get().

  • [grn_config_set()] Renamed from grn_conf_set().

  • [grn_config_delete()] Added a new API that deletes a configuration item.

  • [config_set] Added a new command that sets a configuration item.

  • [config_get] Added a new command that gets a configuration item value.

  • [config_delete] Added a new command that deletes a configuration item.

  • [grn_config_cursor_open()] Added a new API that opens a new cursor that iterates all configuration items.

  • [grn_config_cursor_next()] Added a new API that moves to the next configuration item in the cursor.

  • [grn_config_cursor_get_key()] Added a new API that gets the key of the current configuration item.

  • [grn_config_cursor_get_value()] Added a new API that gets the value of the current configuration item.

  • [Alias] Supported aliasing table and column names.

  • [hash table] Added total key size overflow check.

  • [dump] Supported dumping configurations set by config_set.

  • [patricia trie] Improved inspection. [GitHub#452][GitHub#457][Patch by Naoya Murakami]

  • [grn_get_global_error_message()] Added a new API that gets the current error message in the process.

  • [lock_acquire] Added a new command that acquires a lock of a database, table or column.

  • [lock_release] Added a new command that releases a lock of a database, table or column.

  • [groonga-httpd] Updated bundled nginx version to 1.9.10 from 1.9.7.


  • Fixed mruby related crash bugs.

  • [Windows] Fixed label in installer. [groonga-dev,03825][Reported by Atsushi Shinoda]

  • [doc] Fixed typos. [GitHub#456][GitHub#458][Patch by tSU_RooT]

  • [Memcached binary protocol] Added more description. [GitHub#454][Patch by Hiroyuki Sato]

  • Fixed a bug that Command version specified by command_version parameter in a request isn’t reset.


  • Naoya Murakami

  • Atsushi Shinoda

  • tSU_RooT

  • Hiroyuki Sato

Release 5.1.1 - 2015-12-29


  • [Script syntax] Supported INDEX_COLUMN == VALUE as index search with an equal supported index.

  • Reduced the continuous same messages from inverted index module.

  • Supported detecting truncated tables by another process. [GitHub#447] [groonga-dev,03761] [Reported by Yutaro SHIMAMURA]

  • [grn_db_recover()] Supported detecting a broken object.

  • Improved performance for sequential scan search by Regular expression match with constant pattern such as COLUMN @~ "CONSTANT_PATTERN".


  • Fixed build error on MessagePack support build. [groonga-dev,03708] [Reported by Ryo Sugimoto]

  • [mruby] Fixed a crash bug on error. mruby exception feature is disabled as workaround for now.

  • [thread_limit] Fixed a bug that thread_limit?max=1 may not reduce the number of current running threads to 1.

  • [thread_limit] Fixed a bug that thread_limit?max=1 may not return.

  • [Tuning] Fixed wrong sysctl argument. [GitHub#448] [Reported by Hiroyuki Sato]


  • Ryo Sugimoto

  • Yutaro SHIMAMURA

  • Hiroyuki Sato

Release 5.1.0 - 2015-11-29


  • [patiricia trie] Added an error check for the max total key size. See Limitations about the max total key size.

  • [grndb] Added a check for broken object. The check can detect a case that the object can’t be opened.

  • [grn_obj_reindex()] Added a new API that recreates existing indexes.

  • [reindex] Added a new command that recreates existing indexes.

  • [inverted index] Improved estimation precision for query.

  • [logical_range_filter] Added fallback mode for sequential search. If sequential search checked many records but didn’t find required the number of records, index search is used as fallback.

  • [grn_get_package_label()] Added a new API that returns package label. It returns Groonga.

  • [Groonga HTTP server] Added Server: Groonga/VERSION response header.

  • [groonga-httpd] Improved performance by reusing grn_ctx object.

  • [grn_file_reader] Added a new API that provides fgets() feature. It fixes a crash bug of groonga executable file. If groonga executable file is built with static C runtime by Visual Studio, the crash bug is occurred.

  • [prefix_rk_search] Added a new selector that provides prefix RK search feature.

  • [grn_obj_is_accessor()] Added a new predicate that checks whether the object is an accessor.

  • [grn_obj_is_key_accessor()] Added a new predicate that checks whether the object is an accessor for _key Pseudo column.

  • Supported Pretty print for JSON output.


  • [inverted index] Fixed a possible infinite loop bug when log level is debug.

  • Fixed a bug that @ operator (match operator) may not match record that should be matched in sequential search mode.

  • [patricia trie] Fixed a bug that invalid value may be returned for empty string key. [groonga-dev,03632] [Reported by Naoya Murakami]


  • Naoya Murakami

Release 5.0.9 - 2015-10-29


  • [inverted index] Reduced log levels of logs for developers.

  • Flushed pending changed on creating new database. It guards database from crash.

  • [grn_geo_table_sort()] Added a new API that sorts table by geometry index.

  • [experimental] Added expression rewrite mechanism. You can write custom expression rewriter by mruby. Expression rewriter can be used for optimizing an expression, changing conditions in an expression and so on.

  • [experimental] Added database global configuration mechanism. You can put configurations (key and value pairs) into database. For example, it will be used in TokenFilterStopWord to custom column name from is_stop_word.

  • [grn_conf_set()] Added a new API that sets a configuration.

  • [grn_conf_get()] Added a new API that gets a configuration.

  • [deb] Changed to all from any for Architecture value. [debian-bugs:799167][Reported by Matthias Klose]

  • [Windows][CMake] Supported building bundled MeCab. [groonga-dev,03562][Reported by Sato]

  • [schema] Added a new command that returns schema. Schema is consists with loaded plugins, loaded tokenizers, loaded normalizers, loaded token filters, defined tables and defined columns.

  • [grn_plugin_win32_base_dir()] Deprecated. Use grn_plugin_windows_base_dir() instead.

  • [grn_plugin_windows_base_dir()] Renamed from grn_plugin_win32_base_dir().

  • [grn_obj_is_type()] Add a new API that returns true when the passed object is a type object.

  • [grn_obj_is_tokenizer_proc()] Add a new API that returns true when the passed object is a tokenizer object.

  • [grn_obj_is_normalizer_proc()] Add a new API that returns true when the passed object is a normalizer object.

  • [grn_obj_is_token_filter_proc()] Add a new API that returns true when the passed object is a token filter object.

  • [grn_ctx_get_all_types()] Add a new API that returns all type objects in database.

  • [grn_ctx_get_all_tokenizers()] Add a new API that returns all tokenizer objects in database.

  • [grn_ctx_get_all_normalizers()] Add a new API that returns all normalizer objects in database.

  • [grn_ctx_get_all_token_filters()] Add a new API that returns all token filter objects in database.

  • [grn_ctx_output_uint64()] Add a new API that outputs 64bit unsigned integer value.

  • [grn_ctx_output_null()] Add a new API that outputs NULL.

  • [GRN_OBJ_IS_TRUE()] Add a new API that returns true when the passed object is true value.

  • [experimental] Enabled grn_ts by default.

  • [Ubuntu] Added Ubuntu 15.10 Wily Werewolf support.


  • [patricia trie] Fixed a bug that the number of records may be counted up unexpectedly on adding a new entry. [GitHub#417]

  • [patricia trie] Fixed a bug that a variable may be used uninitialized.

  • [patricia trie] Fixed a bug that grn_pat_cursor_next() may enter an infinite loop. [GitHub#419]

  • [patricia trie] Fixed a bug that deleting an entry may break patricia trie. [GitHub#415][groonga-dev,03515][Reported by Hiroshi Kagami]

  • [patricia trie] Fixed a bug that deleting a nonexistent entry may break patricia trie. [GitHub#420]

  • Fixed a bug that wrong proc type is used for token filter objects.


  • Matthias Klose

  • Hiroshi Kagami

  • Sato

Release 5.0.8 - 2015-09-29


  • [Windows] Supported build with MySQL again.

  • [Script syntax] Changed return value type to Bool from Int32 for predicate operations such as Match operator and Equal operator.

  • [API] Supported owning other grn_obj by GRN_PTR and GRN_PVECTOR bulk. If you specify GRN_OBJ_OWN flag to GRN_PTR and GRN_PVECTOR bulks, they call grn_obj_close() against grn_obj that is held by them when they are closed.

  • [incompatible][Regular expression] Changed to normalize regular expression match target text before matching. It’s for consistency and performance.

    Other operations such as Prefix search operator normalize target text.

    Some simple regular expressions such as \Ahello can be evaluated by index. It’s fast.

    If target text isn’t normalized, you need to use complex regular expressions such as \A[Hh]ello and \A(?i)hello. Complex regular expressions can’t be evaluated by index. If target text is normalized, you can use simple regular expressions. They may be evaluated by index. It’s fast.

  • [doc] Improved documents. [GitHub#393][GitHub#396][GitHub#397][GitHub#399][GitHub#403] [GitHub#405][GitHub#409] [Reported by Hiroyuki Sato][Patch by Hiroyuki Sato]

  • [highlight_html] Improved performance. [groonga-dev,03427] [Reported by Hiroyuki Sato]

  • [snippet_html] Improved performance.

  • [CMake] Stopped to run pkg-config twice. [Patch by Sergei Golubchik]

  • Removed needless check for year in time. B.C. is always invalid time without this change. B.C. is valid time when system (mktime()) supports it with this change.

  • [grn_ctx_is_opened()] Added a new API that checks whether object with the ID is opened or not.

  • [grn_obj_remove()] Reduced the maximum memory usage. If grn_thread_get_limit() returns 1, it closes temporary opened objects after it finished to use them.

  • [doc][table_remove] Updated with many descriptions.

  • [groonga executable file] Ensured to add the last new line to dump result on stand alone mode.

  • [Process log] Added Groonga version into grn_init log.

  • Opened chat room on Gitter.

  • [status] Added start_time as alias of starttime for consistency. Other keys use snake_case style. starttime is deprecated. Use start_time instead.

  • Updated bundled Onigmo.

  • [doc][scorer_tf_at_most] Documented.

  • Supported columns for temporary table. It’s only available C API users for now. select will use this feature in the next release.

  • [grn_vector_pop_element()] Exported.

  • [groonga executable file] Added checks whether acquiring lock is succeeded or not.

  • [groonga-suggest-create-dataset] Changed to use --normalizer instead of KEY_NORMALIZE because KEY_NORMALIZE is deprecated.

  • [grn_obj_cast()] Exported.

  • [experimental][grn_ii_cursor] Exported.

  • [experimental][grn_ii_cursor_open()] Exported.

  • [experimental][grn_ii_cursor_next()] Exported.

  • [experimental][grn_ii_cursor_close()] Exported.

  • [Match operator] Improved index detection. Index its lexicon has a tokenizer is preferred.

  • [groonga-httpd] Updated bundled nginx to 1.9.5 that supports HTTP/2. HTTP/2 module is enabled.


  • [Script syntax] Fixed a bug that &! does nothing when right hand side is true.

  • Fixed performance regression with libtool 2.4.6. [GitHub#406][GitHub#407] [Patch by Hiroyuki Sato]

  • [Equal operator] Fixed a bug that section is ignored.


  • Hiroyuki Sato

  • Sergei Golubchik

Release 5.0.7 - 2015-08-31

This release includes a bug fix of Offline index construction.

If you’re using any multiple column index (index column with WITH_SECTION flag) and Offline index construction, we recommend that you upgrade your Groonga.

This release has an important experimental feature for Windows users. See “sparse file support” entry in the following improvement list for details.


  • [experimental][Windows] Added sparse file support. It’s experimental feature. It’s disabled by default. You can enable it by specifying GRN_IO_USE_SPARSE=yes environment variable.

    It reduces database file size on Windows. Please try the feature and report the result. Groonga developers are interested in the followings:

    • Disk usage

    • Performance (Improved? Degraded? No difference?)

    • Memory usage (Especially virtual memory usage)

  • [experimental][logical_shard_list] Added a command that returns a shard list of the specified logical table.

  • [experimental][Regular expression operator] Supported regular expression match against vector column without index.

  • [logical_range_filter] Supported --cache no option. It’s same as cache option in select.

  • [groonga-httpd] Supported returning the max number of threads feature of thread_limit. You can’t set the max number of threads.

  • [grn_db_unmap()] Added a new API that unmaps all opened tables and columns. It’s a thread unsafe operation. You can’t touch the database while grn_db_unmap() is running.

  • [database_unmap] Added a command that unmaps all opened tables and columns in database.

  • [object_exist] Added a command that checks whether object with the specified name exists or not in database.

  • [column_copy] Added a command that copies all values from source column to destination column.

    You can use this command to change column value type, column type, table type and so on.

  • Stopped to use non-standard __uint32_t type. [GitHub#375][Reported by Natanael Copa]

  • [experimental][Windows] Supported Windows Event log.

  • [mruby] Supported error handling on mruby initialization error.

  • [experimental][thread_limit] Renamed from thread_count.

  • Supported logging used indexes in info level and debug level. It can be used like EXPLAIN in RDBMS. It’s useful to improve slow query.

  • [doc] Replaced deprecated KEY_NORMALIZE flags. [GitHub#378][GitHub#380][GitHub#382] [Patch by Hiroyuki Sato]

  • [doc] Removed needless Sphinx configurations. [GitHub#379] [Patch by Ayumu Osanai]

  • [experimental][incompatible][Regular expression operator] Changed . match behavior. . matches new line. It’s backward incompatible change.

  • [doc][How to build Groonga at the repository] Added a document about building Groonga as Groonga developer. [GitHub#353] [Suggested by Hiro Yoshioka]


  • [mruby] Fixed a time overflow bug.

  • [groonga executable file] Fixed a crash bug when PID file can’t be created. [GitHub#368] [Reported by Hiroyuki Sato]

  • Fixed a bug that Offline index construction may generate broken index. It may be caused for multiple column index. In other words, index column with WITH_SECTION flag may be broken.

    If you’re using Online index construction for index columns with WITH_SECTION flag, this bug isn’t affected.

    You can recover this bug by recreating existing multiple column indexes.

  • [query] Fixed a crash bug when query is used in scorer.

  • [filter] Fixed a bug that Bitwise NOT operator against unsigned int value doesn’t work for comparing to -NUMBER_LITERAL.

    For example, the following expression doesn’t work:

    ~UINT32_COLUMN == -6
  • Fixed a bug that Regular expression operator doesn’t work in multithread.

  • Fixed some memory leaks.

  • Fixed a build error. [GitHub#381] [Patch by Hiroshi Hatake]


  • Hiroyuki Sato

  • Natanael Copa

  • Ayumu Osanai

  • Hiroshi Hatake

  • Hiro Yoshioka

Release 5.0.6 - 2015-07-29


  • [Ubuntu] Dropped Ubuntu 14.10 (Utopic Unicorn) support. It had been End of Life on July 23, 2015.

  • Supported offline index construction for reference vector. For example, load data before column_create an index:

    table_create Entries TABLE_NO_KEY
    column_create Entries numbers COLUMN_VECTOR Int32
    load --table Entries
    [[18, 19, 20]],
    [[100, 200]]
    table_create Numbers TABLE_PAT_KEY Int32
    column_create Numbers entries_numbers COLUMN_INDEX Entries numbers
    select Numbers --output_columns _key
  • Supported 'vector_text_column @ "element"' without index. For example, the select command in the following commands:

    table_create Memos TABLE_NO_KEY
    column_create Memos tags COLUMN_VECTOR Text
    load --table Memos
    {"tags": ["Groonga", "Rroonga", "Mroonga"]}
    select Memos --filter 'tags @ "Rroonga"'
  • Supported 'fixed_size_type_vector_column @ n' without index. For example, the select command in the following commands:

    table_create Memos TABLE_NO_KEY
    column_create Memos numbers COLUMN_VECTOR Int32
    load --table Memos
    {"numbers": [1, 2, 3]}
    select Memos --filter 'numbers @ 2'
  • [CentOS][Server packages][groonga-httpd] Show exit status. [GitHub#357] [Patch by jacob16bit]

  • [Windows][groonga executable file] Supported --pid-path.

  • [Windows] Allowed to delete file that is opened by other process.

  • Accepted selector only proc. Note that the proc can’t be used as function. It means that the proc can’t be used with sequential search.

  • Supported function call with complex argument. An example complex argument is Table["key"].column. For example:

    function(_key, Table["key"].column)
  • [doc][Tutorial] Added more description about database creation fails if DB_PATH points to an existing file. [GitHub#354] [Suggested by Hirotaka Takayama]

  • [doc][Tutorial] Described JSON formatting tools. [GitHub#355] [Suggested by tiwawan]

  • [experimental] Added an API to get/set the number of threads. It’s a experimental API.

  • [experimental][thread_count] Added a command that get/set the number of threads. It’s a experimental command.

  • [experimental][groonga executable file] Supported changing the number of threads by thread_count command. It’s a experimental feature.

  • [experimental][Windows] Added Windows event log support. It’s a experimental feature.

  • [experimental][groonga executable file] Added Windows event log related option --use-windows-event-log. It reports logs as Windows events. It’s a experimental feature.

  • [Windows] Used Groonga’s default encoding for log message.

  • Log used indexes in INFO level. The default level NOTICE. So the logs aren’t showed by default.

  • [API] Added grn_log_level_to_string() and grn_log_level_parse().

  • [groonga executable file] Accepted log level name (e.g. info, debug and so on) for --log-level value.

  • [log_level][log_put] Accepted log level name for --level argument.

  • [plugin] Added grn_command_input_get_arguments().

  • Updated sharding plugins.

  • Added mruby APIs.

    • [mrb] Added Accessor#name.

    • [mrb] Added Column#[].

    • [mrb] Added Column#scalar?, Column#vector? and Column#index?.

    • [mrb] Added Context#command_version and Context#command_version= (accessors).

    • [mrb] Added Context#with_command_version.

    • [mrb] Added Database#each_name.

    • [mrb] Added Groonga::Cache.current.

    • [mrb] Added Record.

    • [mrb] Added Table#each.

    • [mrb] Added TableCursor#key.

    • [mrb] Binded grn_command_input_get_arguments() to CommandInput#arguments.

    • [mrb] Binded grn_table_group() to Table#group.

    • [mrb] Binded grn_table_group_flags() to TableGroupFlags.


    • [mrb] Binded grn_cache.

    • [mrb][estimate_size] Supported (... || ...) && (... || ...) as expression case.

    • [mrb] Supported query log.


  • Fixed a memory leak when an error is occurred in grn_expr_exec(). For example, unsupported operator (e.g. GRN_OP_TERM_EXTRACT) is used (not implemented operator assigned is occurred for the case).

  • [bindings/php] Added a missing check for a memory allocation failure. [Reported by Bill Parker]

  • [CentOS][Server packages][logrotate] Fixed syntax error in script.

  • [CentOS][Server packages][logrotate] Fixed wrong daemon running check.

  • [CentOS][Server packages][logrotate] Stop to set owner/group to log files. Because it’s not consistent. groonga-httpd creates log files with root owner/group. But logrotated log files are created with groonga owner/group. [GitHub#358] [Reported by jacob16bit]

  • [groonga executable file] Fixed reported the maximum number of threads.

  • [groonga-httpd] Remove a needless space in log message:

    |n|  grn_fin (0) ->
    |n| grn_fin (0)
  • Fixed a bug that estimating size by regexp query with anchor (e.g. \\\\A in --filter 'comment @~ "\\\\Abc"') doesn’t work. The feature is used in logical_range_filter.

  • [Request ID] Fixed a memory leak when request_id byte size >= 24.

  • [lock_clear] Fixed a typo in command name in Syntax section. [GitHub#363] [Reported by Christian Kakesa]

  • [sharding] Fixed wrong min include detection for month range type.


  • Bill Parker

  • jacob16bit

  • Hirotaka Takayama

  • tiwawan

  • Christian Kakesa

Release 5.0.5 - 2015-06-29


  • Show correct error information such as NoSuchFileOrDirectory when opening a database.

  • Don’t set the default logger path for library use.

    • It’s backward incompatible change. But it will not effect to many users.

    • Server use (groonga command, Mroonga, PGroonga and so on) users can get log by default. In server use, developers set up log in their software.

    • Most library use (Rroonga, groonga-gobject and so on) users couldn’t get log by default with earlier versions. The default log path is system path such as /var/log/groonga/groonga.log. It’s not writable for normal users.

  • [windows] Show error information when memory isn’t enough on failing CreateFileMapping().

  • [tokenize] Updated example to show new “force_prefix” value. This value is added since 5.0.4.

  • [windows] Show error information when disk has any problem (disk full and so on) on failing FlushViewOfFile().

  • [API] Added grn_obj_flush().

  • [API] Added grn_obj_flush_recursive().

  • [io_flush] Added. It flushes memory mapped data to disk. Usually memory data automatically flush by an OS, but you can explicitly flush with this command.

  • [mruby] Binded grn_obj_remove() to Object#remove.

  • [mruby] Binded grn_table_delete() and grn_table_delete_by_id() to Table#delete.

  • [logical_table_remove] Added.

  • [logical_select] Added. --filter is only supported for now.

  • [cmake] Supported embedded MeCab tokenizer.

  • [logical_count] Supported month and day mixed shards. In the same month, month shard must have earlier records rather than day shards in the same month. For example:

    XXX_201506   <- includes only 2015-06-01 and 2015-06-02 records
    XXX_20150603 <- includes only 2015-06-03 records
    XXX_20150604 <- includes only 2015-06-04 records


  • Fixed wrong macro to include netinet/in.h. [GitHub#348] [Reported by OBATA Akio]

  • [rpm][groonga-httpd] Fixed failing restart. [GitHub#351] [Patch by jacob16bit]


  • OBATA Akio

  • jacob16bit

Release 5.0.4 - 2015-05-29


  • [mruby] Changed to use inspect to show meaningful error message for error value.

  • [mruby] Supported Groonga::Bulk#inspect to inspect bulk content.

  • [mruby] Supported Bulk#value to extract the value of record from bulk content.

  • [mruby] Supported estimating size for reference_column == record_id in logical_range_filter. In above case, it can be searched more effectively.

  • [sub_filter] Supported index column as scope parameter.

  • [Script syntax] Described clearly about numerical value. [GitHub groonga/] [Suggested by Hiroyuki Sato]

  • [select] Supported accessing other table’s record in filter. You can use --filter 'OTHER_TABLE[KEY].COLUMN' for example.

  • [select] Supported operator in table key. You can use --filter 'OTHER_TABLE["SOME STRING" + "ANOTHER STRING"].COLUMN'" for example.

  • [example] Used Ruby 2.0 or later API in script for converting dictionary data.

  • Changed to show error message about invalid type of keys about table.

  • [doc] Fixed link from to or about mailing list preference page. is marked as obsoleted because of branding issue since May 11, 2015.

  • [tokenize] Added force_prefix value to each token information. [Patch by Naoya Murakami]

  • Supported to search by shorter words such as 2 or less characters for TokenTrigram. [Patch by Naoya Murakami]

  • [deb] Added service file for Systemd into groonga-httpd and groonga-server-gqtp packages.

  • [select] Ignored --query when its value consists of only space characters. Space characters include full-width space (U+3000 IDEOGRAPHIC SPACE in Unicode). [Suggested by TomyGX]


  • Fixed a crash bug when empty key is specified for drilldown[label].keys.

  • Fixed a bug that the return value of grn_parse_query_flags is not properly checked. [GitHub#336] [Reported by Hiroaki Nakamura]

  • Fixed a build error on some BSD systems. They doesn’t have -i option for sed.

  • Fixed a build error on Solaris. It is changed to initialize by sizeof(msghdr) in memset() because msg_control, msg_controllen and msg_flags doesn’t exist on Solaris by default.

  • [Tokenizers] Fixed a typo. [GitHub#338] [Reported by Hiroyuki Sato]

  • [Output] Fixed markup. [GitHub groonga/] [Reported by Hiroyuki Sato]

  • Reduced getenv() in each grn_ii_cursor_set_min(). This fixes performance regression on Windows.

  • Fixed a build error on OpenBSD. [groonga-dev,03255] [Reported by fbnteqr]

  • [groonga-httpd] Fixed a bug that same message is logged.

  • Fixed a crash bug which is caused by double free memory.

  • Fixed a memory leak. It’s occurred when --match_columns and --query are used for non indexed text field and text fields have a value that isn’t bulk embeddable. Normally, 32byte over size text isn’t bulk embeddable, so this bug doesn’t affect to the case if only small text less than 32byte are stored.

  • [Tokenizers] [TokenRegexp] Fixed a bug that it can’t be searched correctly when query contains characters which are treated as blank character. For example, the newline - “\n” is typical one.


  • Hiroaki Nakamura

  • Hiroyuki Sato

  • Naoya Murakami

  • fbnteqr

  • TomyGX

Release 5.0.3 - 2015-04-29


  • [Tokenizers][Regular expression] Skip the last one character token.

  • [mruby] Supported regexp pattern for estimating size.

  • [mruby] Supported size estimation for accessor.

  • [logical_range_filter] Removed GRN_LOGICAL_RANGE_FILTER_ENABLED environment variable which is introduced since Groonga 5.0.2. Use GRN_LOGICAL_RANGE_FILTER_THRESHOLD=0 to disable range index search feature.

  • [logical_range_filter] Supported negative limit and offset.

  • [Windows] Used Groonga Admin in package.

  • [logical_range_filter] Changed threshold meaning:

    • threshold <= 0.0: always use range index

    • threshold >= 1.0: never use range index

  • [dump] Supported plugin.

  • [dump] Added the following options:

    • --dump_plugins [yes(default)/no]

    • --dump_schema [yes(default)/no]

    • --dump_records [yes(default)/no]

    • --dump_indexes [yes(default)/no]

  • [API] Added grn_plugin_get_ruby_suffix().

  • [dump] Fixed order to put index columns after reference columns because index column may refer reference columns.

  • [dump] Don’t dump records of lexicon.

  • [dump] Show _id for TABLE_NO_KEY again.

  • [dump] Used offline index construnction.

  • Increased max hash key size from 4KiB (4096Byte) to 64KiB - 1 (65535Byte).

  • Increased max cache key size from 4KiB (4096Byte) to 64KiB - 1 (65535Byte).

  • Improved performance for nested index search.

  • Used index for nonexistent reference column value.

  • [experimental] Added plugin functions/vector. It includes vector_size function.

  • [Windows] Updated Visual Studio version [GitHub groonga/meetup#4] [Reported by Hiroyuki Mizuhara]

  • [cache_limit] Expired old caches when the max N caches is decreased. [Suggested by Gurunavi, Inc.]

  • Show more information such as errno for errors.

  • [windows] Used secure functions on Windows.

  • Added the following APIs to change log rotate threshold in file size.

    • grn_default_logger_set_rotate_threshold_size()

    • grn_default_logger_get_rotate_threshold_size()

    • grn_default_query_logger_set_rotate_threshold_size()

    • grn_default_query_logger_get_rotate_threshold_size()

  • [experimental] Supported log rotation. The feature is disabled by default. You can enable log rotation by the following options:

    • --log-rotate-threshold-size

    • --query-log-rotate-threshold-size

  • [GQTP] Documented about GQTP server.

  • [groonga executable file] Documented groonga executable file partially.

  • Supported Ubuntu 15.04 (Vivid Vervet).

  • Supported Debian 8.0 (Jessie).

  • [groonga-httpd] Updated bundled nginx version to the latest mainline (1.8.0).


  • [windows] Fixed a bug that Offline index construction is failed for large data (at least 1GB or larger) with Groonga built by Microsoft Visual C++. [Reported by Hideki ARAI]

  • [mruby] Made \\ index searchable in regular expression.

  • Fixed a bug that GRN_II_CURSOR_SET_MIN_ENABLE=yes doesn’t return some matched records.

  • [sharding] Fixed a bug that partial range is handled as all range.

  • [logical_range_filter] Fixed a bug that :order => "descending" doesn’t work.

  • [logical_count] Re-supported counting with range index.

  • Fixed a bug causing malfunction of grn_pat_del() and added a test for invalid patricia trie node add case. [groonga-dev,03177] [Reported by yuya sako]


  • Hideki ARAI

  • Hiroyuki Mizuhara

  • Gurunavi, Inc.

  • yuya sako

Release 5.0.2 - 2015-03-31

It’s a bug fix release of 5.0.1.


  • Supported MessagePack 1.0.1. [Reported by Hiroshi Hatake]

  • [logical_range_filter] Disabled range index by default. It’s enabled when you set the enviromnent variable GRN_LOGICAL_RANGE_FILTER_ENABLED to yes.


  • Fixed a regression bug that JSONP doesn’t work. It was introduced in Groonga 4.1.1.

  • [windows] Fixed a bug that crash on x86 for Groonga 5.0.1. [groonga-dev,03131] [Reported by Atsushi Shinoda]

  • Fixed a crash bug that libedit is not properly initialized. The problem is fixed in the environment such as CentOS 7.


  • Atsushi Shinoda

  • Hiroshi Hatake

Release 5.0.1 - 2015-03-29


  • [logical_range_filter] Supported filter and sort.

  • Supported range search by multiple column index.

  • Added API Overview document for users who want to use Groonga as library.

  • [incompatible] Changed internal type of _score to floating point number from 32bit integer number. This is incompatible change for DB API users. This isn’t incompatible change for query API users. It means that users who just use select aren’t affected. Use the following code that works with both older and newer Groonga:

    grn_obj *score;
    double score_value;
    if (score->header.domain == GRN_DB_FLOAT) {
      score_value = GRN_FLOAT_VALUE(score);
    } else {
      score_value = (double)GRN_INT32_VALUE_FLOAT_VALUE(score);
  • [select] Added more strict check for invalid drilldown parameter.

  • Added grn_ctx_get_all_tables(). [Suggested by Masatoshi Teruya]

  • Supported to customize score function. See Scorer for details.

  • [incompatible] Custom score function feature introduced API and ABI incompatibilities in DB API layer. If you’re using grn_search_optarg, please check that your code initializes your grn_search_optarg by 0 like the following:

    grn_search_optarg options;
    memset(&options, 0, sizeof(grn_search_optarg));

    If your code do the above thing, your code is API compatible and ABI incompatible. You just need to rebuild your code without modification.

    If your code doesn’t the above thing, you need to added the above thing to your code.

  • Added the following predicates that check grn_obj type to DB API:

    • grn_obj_is_table()

    • grn_obj_is_proc_proc()

    • grn_obj_is_function_proc()

    • grn_obj_is_selector_proc()

    • grn_obj_is_scorer_proc()

  • [experimental] Supported skipping posting list when searching popular term and rare term at the same time. It will improve performance. Set GRN_II_CURSOR_SET_MIN_ENABLE environment variable to 1 to enable the feature. The feature is disabled by default.

  • [doc] Added in_values document.

  • [doc] Added logical_count document.

  • [mruby] Implemented custom #inspect method. Is is useful for debugging.

  • Added scorer_tf_at_most scorer. It limits not to exceed specified score regardless of term frequency.

  • [mruby] Supported estimating matched records for selecting index search or sequential search.

  • Added the following functions to estimate size by index:

    • grn_expr_estimate_size()

    • grn_ii_estimate_size_for_query()

    • grn_ii_estimate_size_for_lexicon_cursor()

  • Added missing NormalizerAuto availability check. [GitHub#283] [Reported by Tasuku SUENAGA]

  • Dropped Visual Studio 2010 support.

  • [experimental][mecab] Supported chunked tokenization. This feature is a workaround for MeCab’s “too long sentense” error. Specify yes to GRN_MECAB_CHUNKED_TOKENIZE_ENABLED environment variable to enable it. By this configuration, Groonga splits a long text (8192 bytes over text by default) into small chunks and passes each chunk to MeCab. By this process, the above error isn’t occurred. Additionally, you can customize chunk threshold bytes by GRN_MECAB_CHUNK_SIZE_THRESHOLD environment variable. Note that ,, ., !, ?, U+3001 IDEOGRAPHIC COMMA, U+3002 IDEOGRAPHIC FULL STOP, U+FF01 FULLWIDTH EXCLAMATION MARK and U+FF1F FULLWIDTH QUESTION MARK are treated as chunk delimiter characters.

  • Supported --pid-file in server mode of groonga executable file.

  • [groonga-httpd] Supported graceful stop to clean Groonga. It doesn’t terminate the open connections immediately.

  • [experimental] Supported regular expression. See Regular expression to know about how to use regular expression.

  • [experimental] Added plugin_unregister command.

  • [http][load] Added “,” as chunk separator in POST data. It decreases internal buffer size and improves load time when POST data don’t include any new line.

  • [doc] Added Tokenizers document.

  • Improved POSIX.2 compatibility by using . as bash’s “source” command replacement. [GitHub#317] [Patch by Jun Kuriyama]

  • [windows] Changed to the default IO version 1. It reduces disk usage on Windows. [groonga-dev,03118] [Tested by ongaeshi]

  • [httpd] Updated bundled nginx version to the latest mainline (1.7.11).

  • Changed mime-type for TSV output to text/tab-separated-values from text/plain.

  • [TokenFilterStopWord] Supported Offline index construction. [GitHub#296] [Patch by Naoya Murakami]


  • Fixed not to use obsolete --address parameter in the default groonga.conf. --bind-address is used instead. [Groonga-talk] [Reported by Dewangga]

  • [truncate] Fixed a bug that TABLE_NO_KEY table can’t be truncated.

  • [mecab] Reduced needless logs for “empty token” and “ignore empty token”.

  • Fixed a bug that wrong section in index is used. It means that wrong search result is returned. If you satisfy all of the following conditions, this bug is occurred:

    • Multiple indexes are available.

    • The first defined index or the last defined index are multi-column indexes.

    • When both of the first defined index and the last defined index are multi-column indexes, source column orders are different in them.

  • Fixed a bug that passing Groonga command line to groonga executable file from shell command line style usage always returns 0 as exit code. For example, groonga DB_PATH nonexistent_command always returned 0 as exist code.

  • Fixed a bug that plugin path may be broken when two or more plugins registered. [Reported by Naoya Murakami]

  • Fixed a bug that Lexicon.index.source_column_name style in match_columns doesn’t work when source column specified by source_column_name has two or more indexes. [Reported by Naoya Murakami]


  • Masatoshi Teruya

  • Tasuku SUENAGA

  • Dewangga

  • Jun Kuriyama

  • ongaeshi

  • Naoya Murakami

Release 5.0.0 - 2015-02-09

  • Bump version to 5.0.0!


  • [doc] Added Security about Script syntax.

  • [experimental] Added sharding plugin. Execute register sharding to enable this feature in advance, then use logical_count to get the number of records.

  • [cmake] Supported embedded Groonga with Clang. It fixed compilation failure on FreeBSD 10.1. [MDEV-7293] [Reported by Bernard Spil]

  • Supported to customize plugins directory. Set GRN_PLUGINS_DIR environment variable.


  • Fixed build failure when system has an incompatible version of onigmo/oniguruma headers installed. [GitHub#276] [Patch by Akinori MUSHA]

  • Fixed time related build failure on MSVC [GitHub#237]


  • Akinori MUSHA

  • Bernard Spil