- developers - lists.mariadb.org

Re: [Maria-developers] [Commits] 6811945e: MDEV-6756: map a linux pid (child pid) to a connection id shown in the output of SHOW PROCESSLIST
by Sergei Golubchik 14 Sep '15

14 Sep '15

Hi, Sanja! On Aug 12, sanja(a)mariadb.com wrote: > revision-id: 6811945eaf6a6e979d92739399b084bf4e00ae2d (mariadb-10.1.6-12-g6811945e) > parent(s): 86a3613d4e981c341e38291c9eeec5dc9f836fae > committer: Oleksandr Byelkin > timestamp: 2015-08-12 23:09:48 +0200 > message: > > MDEV-6756: map a linux pid (child pid) to a connection id shown in the output of SHOW PROCESSLIST > > Added tid & pid for Linux. > > diff --git a/mysql-test/r/information_schema.result b/mysql-test/r/information_schema.result > index 2a918ad..5b3370b 100644 > --- a/mysql-test/r/information_schema.result > +++ b/mysql-test/r/information_schema.result > @@ -2066,3 +2066,18 @@ Variable_name Value > Opened_tables 3 > drop database mysqltest; > drop database db1; > +SHOW PROCESSLIST; > +Id User Host db Command Time State Info Progress > +# root # NULL Query # # SHOW PROCESSLIST 0.000 > +set @@old_mode=""; > +SHOW PROCESSLIST; Eh, no, please, don't change SHOW PROCESSLIST. Only add new columns to I_S.PROCESSLIST. And without OLD_MODE. > diff --git a/sql/CMakeLists.txt b/sql/CMakeLists.txt > index 033a0e0..d1fbaa3 100644 > --- a/sql/CMakeLists.txt > +++ b/sql/CMakeLists.txt > @@ -143,6 +143,10 @@ SET (SQL_SOURCE > ${MYSYS_LIBWRAP_SOURCE} > ) > > +IF (CMAKE_SYSTEM_NAME MATCHES "Linux") > + ADD_DEFINITIONS(-DHAVE_TID_SYSCALL) > +ENDIF() Don't match OS name please, better use #ifdef HAVE_SYS_SYSCALL_H #include <sys/syscall.h> #endif ... #ifdef __NR_gettid ... #endif > + > IF (CMAKE_SYSTEM_NAME MATCHES "Linux" OR > CMAKE_SYSTEM_NAME MATCHES "Windows" OR > CMAKE_SYSTEM_NAME MATCHES "SunOS" OR > diff --git a/sql/sql_class.cc b/sql/sql_class.cc > index bf161b2..20c03fe 100644 > --- a/sql/sql_class.cc > +++ b/sql/sql_class.cc > @@ -2062,6 +2068,13 @@ bool THD::store_globals() > This allows us to move THD to different threads if needed. > */ > mysys_var->id= thread_id; > +#ifdef HAVE_TID_SYSCALL > + os_pid= getpid(); why would you want to have pid there? isn't it the same for all threads? > + os_thread_id= syscall(__NR_gettid); > +#else > + os_pid= 0; > + os_thread_id= 0; > +#endif > real_id= pthread_self(); // For debugging > mysys_var->stack_ends_here= thread_stack + // for consistency, see libevent_thread_proc > STACK_DIRECTION * (long)my_thread_stack_size; I didn't review the rest of the patch as without old_mode and SHOW PROCESSLIST changes most of it will disappear. Regards, Sergei

2 2

Re: [Maria-developers] [Commits] 5fab5b6: MDEV-8764: Wrong error when encrypted table can't be decrypted.
by Sergei Golubchik 13 Sep '15

13 Sep '15

Hi, Jan! On Sep 08, Jan Lindström wrote: > MDEV-8764: Wrong error when encrypted table can't be decrypted. > > Add a new error message when table is encrypted but decryption > fails. Use this new error message on InnoDB/XtraDB. > > diff --git a/include/my_handler_errors.h b/include/my_handler_errors.h > index a7afcfe..9e452b0 100644 > --- a/include/my_handler_errors.h > +++ b/include/my_handler_errors.h > @@ -94,7 +94,8 @@ static const char *handler_error_messages[]= > "Operation was interrupted by end user (probably kill command?)", > "Disk full", > "Incompatible key or row definition between the MariaDB .frm file and the information in the storage engine. You have to dump and restore the table to fix this", > - "Too many words in a FTS phrase or proximity search" > + "Too many words in a FTS phrase or proximity search", > + "Table encrypted but decryption failed" I'm not sure I like the wording ("table is encrypted" adds no useful information, it's not like there can be another error "table is not encrypted but decryption failed"). But that's up to you. And yes, I agree that this situation deserves a dedicated error message. > }; > > #endif /* MYSYS_MY_HANDLER_ERRORS_INCLUDED */ > diff --git a/sql/handler.cc b/sql/handler.cc > index 1d1dae7..566db8a 100644 > --- a/sql/handler.cc > +++ b/sql/handler.cc > @@ -3518,6 +3519,10 @@ void handler::print_error(int error, myf errflag) > my_error(ER_NO_SUCH_TABLE_IN_ENGINE, errflag, table_share->db.str, > table_share->table_name.str); > DBUG_VOID_RETURN; > + case HA_ERR_DECRYPTION_FAILED: > + my_error(ER_DECRYPTION_FAILED, errflag, table_share->db.str, > + table_share->table_name.str); > + DBUG_VOID_RETURN; Is that needed? I'd rather use the default ER_GET_ERRMSG, which would be Got error 192 'Table encrypted but decryption failed" from InnoDB I think this is perfectly adequate error for internal engine errors. ER_xxx is for upper-level errors. > case HA_ERR_RBR_LOGGING_FAILED: > textno= ER_BINLOG_ROW_LOGGING_FAILED; > break; > diff --git a/sql/share/errmsg-utf8.txt b/sql/share/errmsg-utf8.txt > index 59908dc..ff72a26 100644 > --- a/sql/share/errmsg-utf8.txt > +++ b/sql/share/errmsg-utf8.txt > @@ -7136,3 +7136,5 @@ ER_KILL_QUERY_DENIED_ERROR > eng "You are not owner of query %lu" > ger "Sie sind nicht Eigentümer von Abfrage %lu" > rus "Вы не являетесь владельцем запроса %lu" > +ER_DECRYPTION_FAILED 42S02 > + eng "Table '%-.192s.%-.192s' encrypted but decryption failed." In general, please prefer %`s.%`s over '%s.%s' the former uses correct identifier quoting, the second does not Regards, Sergei

1 0

Re: [Maria-developers] [Commits] 7fd5554: MDEV-5171: Add support for --innodb-optimize-keys to mysqldump.
by Sergei Golubchik 12 Sep '15

12 Sep '15

Hi, Jan! First, we've discussed adding support for ALTER TABLE ... DISABLE KEYS to InnoDB. Did you try that? That would've been a better approach, generally I'd like to avoid adding plugin-specific options to a mysqldump utility (although, of course, InnoDB is not just any plugin, so we can always make an exception for it). On Jul 27, Jan Lindström wrote: > revision-id: 7fd55542893b6d9e75406e39b7be672d8a8f473c > parent(s): a6ab8ef9d7d465a3f0bc1cff6034ca97575629dd > committer: Jan Lindström > branch nick: 10.0-git > timestamp: 2015-07-27 13:02:57 +0300 > message: > > MDEV-5171: Add support for --innodb-optimize-keys to mysqldump. > > Merged revision 93 from bzr+ssh://bazaar.launchpad.net/+branch/percona-server/5.6/ > authored by Alexey Kopytov. > > However, internal parser is rewriten to support unquoted identifiers and > multi-column keys. > > This patch expands the applicability of InnoDB fast index creation to > mysqldump, ALTER TABLE and OPTIMIZE TABLE as follows: > > 1. mysqldump has now a new option, --innodb-optimize-keys, which changes > the way InnoDB tables are dumped so that secondary and foreign keys are > created after loading the data thus taking advantage of fast index > creation. > > This part of the patch is an implementation of the feature request > reported as MySQL bug #49120. > > More specifically: > > - KEY, UNIQUE KEY and CONSTRAINT specifications are omitted from CREATE > TABLE corresponding to InnoDB tables. > > - an additional ALTER TABLE is issued after dumping the data to create > the previously omitted keys. > > Delaying foreign key creation does not introduce any additional risks as > mysqldump always prepends its output with SET FOREIGN_KEY_CHECKS=0 anyway. > > 2. When ALTER TABLE requires a table copy, secondary keys are now dropped > and recreated later after copying the data. The following restrictions > apply: > > - only non-unique keys can be involved in this optimization > > - if the table contains foreign keys, or a foreign key is being added as > a part of the current ALTER TABLE statement, the optimization is > disabled for all keys. > > This part of the patch is an implementation of the feature request > reported as MySQL bug #57583. > > 3. As OPTIMIZE TABLE is mapped to ALTER TABLE ... ENGINE=InnoDB for > InnoDB tables, it now also benefits from fast index creation with the > same restrictions as for ALTER TABLE. > This is a good comment, but a bit confusing. It should explicitly say somehow that item 1 is what this patch implements and items 2 and 3 are existing InnoDB features. Currently the comment creates an impression that all 1-3 are implemented in the patch. > > diff --git a/client/mysqldump.c b/client/mysqldump.c > index e4683ab..b8874e7 100644 > --- a/client/mysqldump.c > +++ b/client/mysqldump.c > @@ -84,12 +86,30 @@ > #define IGNORE_DATA 0x01 /* don't dump data for this table */ > #define IGNORE_INSERT_DELAYED 0x02 /* table doesn't support INSERT DELAYED */ > > +typedef enum { > + KEY_TYPE_NONE, > + KEY_TYPE_PRIMARY, > + KEY_TYPE_UNIQUE, > + KEY_TYPE_NON_UNIQUE, > + KEY_TYPE_FOREIGN > +} key_type_t; > + > /* Chars needed to store LONGLONG, excluding trailing '\0'. */ > #define LONGLONG_LEN 20 > > /* Max length GTID position that we will output. */ > #define MAX_GTID_LENGTH 1024 > > +/* Max length of identifier (columnname, keyname, tablename, etc). */ > +#define MAX_ID_LENGTH 65 Use SAFE_NAME_LEN instead, it's defined in mysql_com.h that mysqldump.c already includes > + > +/* Data structure for holding position information > +for simple keyword parser. */ > +typedef struct { > + char* m_start; > + char* m_pos; > +} simple_parser_t; > + > static void add_load_option(DYNAMIC_STRING *str, const char *option, > const char *option_value); > static ulong find_set(TYPELIB *lib, const char *x, uint length, > @@ -2843,6 +3296,9 @@ static uint get_table_structure(char *table, char *db, char *table_type, > > row= mysql_fetch_row(result); > > + if (opt_innodb_optimize_keys && !opt_comments && !strcmp(table_type, "InnoDB")) > + skip_secondary_keys(row[1], table); Okay, so opt_innodb_optimize_keys needs !opt_comments. Please, add a check, like if (opt_innodb_optimize_keys && opt_comments) { fprintf(stderr, "%s: You can't use --innodb-optimize-keys and " "--comments at the same time.\n", my_progname_short); return(EX_USAGE); } a good place for a check like this is get_options() function. > + > is_log_table= general_log_or_slow_log_tables(db, table); > if (is_log_table) > row[1]+= 13; /* strlen("CREATE TABLE ")= 13 */ The parser looked fine on the first glance, but I didn't review every single line of it. Regards, Sergei

1 0

Re: [Maria-developers] [Commits] 7206e6f: MDEV-8577: With enforce-storage-engine mysql_upgrade corrupts the schema:
by Sergei Golubchik 11 Sep '15

11 Sep '15

Hi, Jan! On Aug 08, Jan Lindström wrote: > MDEV-8577: With enforce-storage-engine mysql_upgrade corrupts the schema: > ALTER TABLE should either bypass enforce-storage-engine, or mysql_upgrade > should refuse to run > > Allow user to alter contents of existing table without enforcing > storage engine. However, enforce storage engine on ALTER TABLE x ENGINE=y; Good idea, I like this behavior. > diff --git a/sql/sql_table.cc b/sql/sql_table.cc > index 69c87dd..60c4407 100644 > --- a/sql/sql_table.cc > +++ b/sql/sql_table.cc > @@ -9801,8 +9801,12 @@ static bool check_engine(THD *thd, const char *db_name, > *new_engine= ha_checktype(thd, req_engine, no_substitution); > DBUG_ASSERT(*new_engine); > > - /* Enforced storage engine should not be used on bootstrap */ > - if (!in_bootstrap) > + /* Enforced storage engine should not be used on bootstrap > + and in ALTER TABLE that does not use explicit ENGINE = x to > + avoid unwanted unrelated changes.*/ > + if (!in_bootstrap && Note that !in_bootstrap belongs to another bug fix, please don't accidentally push it together with this one. > + !(thd->lex->sql_command == SQLCOM_ALTER_TABLE && > + !(create_info->used_fields & HA_CREATE_USED_ENGINE))) > enf_engine= thd->variables.enforced_table_plugin ? > plugin_hton(thd->variables.enforced_table_plugin) : NULL; Looks good, ok to push! Regards, Sergei

1 0

[Maria-developers] MDEV-8372 Use helper methods introduced in MDEV-7824 all around the code
by Alexander Barkov 11 Sep '15

11 Sep '15

Hi Sergei, Please review my patch for mdev-8372. I found only a few places to reuse the new methods. Also, had to add some more helper methods to make the code clearer. Thanks.

2 3

Re: [Maria-developers] [Commits] 32b1497: MDEV-7990: ERROR 1526 when procedure executed for second time ALTER TABLE partition ... pMAX values less than MAXVALUE
by Sergei Golubchik 11 Sep '15

11 Sep '15

Hi, Sanja! On Aug 06, sanja(a)mariadb.com wrote: > revision-id: 32b14970a1b88571aa96f604e5c59ab6bf1b298c > parent(s): aa5095627e2619bdad7916d33d1016802a84a9e1 > committer: Oleksandr Byelkin > branch nick: server > timestamp: 2015-08-06 14:00:48 +0200 > message: > > MDEV-7990: ERROR 1526 when procedure executed for second time ALTER > TABLE partition ... pMAX values less than MAXVALUE > > Added cleaning partition data which are saved in LEX structure to make > possible correct re-execution of a SP or PS using the code. This seems to be wrong now. It's an old comment, but now your fix works differently, doesn't it? The patch is ok to push, thanks! Regards, Sergei

1 0

Re: [Maria-developers] [Commits] 0a1b60a: MDEV-8450: PATCH] Wrong macro expansion in Query_cache::send_result_to_client()
by Sergey Vojtovich 06 Sep '15

06 Sep '15

Hi Sanja, ok to push, a few minor notes JFYI inline. I found only one real bug, but generally I agree that we should fix these fragile macros. On Tue, Sep 01, 2015 at 11:47:09AM +0200, sanja(a)mariadb.com wrote: > revision-id: 0a1b60a8aee4bce3235fd5a1d8cb2231b51ff5cc (mariadb-10.0.21-7-g0a1b60a) > parent(s): 4b41e3c7f33714186c97a6cc2e6d3bb93b050c61 > committer: Oleksandr Byelkin > timestamp: 2015-09-01 11:47:06 +0200 > message: > > MDEV-8450: PATCH] Wrong macro expansion in Query_cache::send_result_to_client() > > Expression in macro protected by () > > diff --git a/include/maria.h b/include/maria.h > index 16a9bea..cdaeade 100644 > --- a/include/maria.h > +++ b/include/maria.h > @@ -43,7 +43,7 @@ extern "C" { > #define MARIA_NAME_IEXT ".MAI" > #define MARIA_NAME_DEXT ".MAD" > /* Max extra space to use when sorting keys */ > -#define MARIA_MAX_TEMP_LENGTH 2*1024L*1024L*1024L > +#define MARIA_MAX_TEMP_LENGTH (2*1024L*1024L*1024L) > /* Possible values for maria_block_size (must be power of 2) */ > #define MARIA_KEY_BLOCK_LENGTH 8192 /* default key block length */ > #define MARIA_MIN_KEY_BLOCK_LENGTH 1024 /* Min key block length */ Was this used at all? May be remove it? > diff --git a/sql/log_slow.h b/sql/log_slow.h > index 2ae07da..3ae2060 100644 > --- a/sql/log_slow.h > +++ b/sql/log_slow.h > @@ -16,23 +16,23 @@ > /* Defining what to log to slow log */ > > #define LOG_SLOW_VERBOSITY_INIT 0 > -#define LOG_SLOW_VERBOSITY_INNODB 1 << 0 > -#define LOG_SLOW_VERBOSITY_QUERY_PLAN 1 << 1 > -#define LOG_SLOW_VERBOSITY_EXPLAIN 1 << 2 > +#define LOG_SLOW_VERBOSITY_INNODB (1 << 0) Was LOG_SLOW_VERBOSITY_INNODB used at all? May be remove it? > +#define LOG_SLOW_VERBOSITY_QUERY_PLAN (1 << 1) > +#define LOG_SLOW_VERBOSITY_EXPLAIN (1 << 2) > > #define QPLAN_INIT QPLAN_QC_NO > > -#define QPLAN_ADMIN 1 << 0 > -#define QPLAN_FILESORT 1 << 1 > -#define QPLAN_FILESORT_DISK 1 << 2 > -#define QPLAN_FULL_JOIN 1 << 3 > -#define QPLAN_FULL_SCAN 1 << 4 > -#define QPLAN_QC 1 << 5 > -#define QPLAN_QC_NO 1 << 6 > -#define QPLAN_TMP_DISK 1 << 7 > -#define QPLAN_TMP_TABLE 1 << 8 > -#define QPLAN_FILESORT_PRIORITY_QUEUE 1 << 9 > +#define QPLAN_ADMIN (1 << 0) > +#define QPLAN_FILESORT (1 << 1) > +#define QPLAN_FILESORT_DISK (1 << 2) > +#define QPLAN_FULL_JOIN (1 << 3) > +#define QPLAN_FULL_SCAN (1 << 4) > +#define QPLAN_QC (1 << 5) > +#define QPLAN_QC_NO (1 << 6) > +#define QPLAN_TMP_DISK (1 << 7) > +#define QPLAN_TMP_TABLE (1 << 8) > +#define QPLAN_FILESORT_PRIORITY_QUEUE (1 << 9) > > /* ... */ > -#define QPLAN_MAX ((ulong) 1) << 31 /* reserved as placeholder */ > +#define QPLAN_MAX (((ulong) 1) << 31) /* reserved as placeholder */ > log_slow.h was fixed in rev.203f4d419 in my 10.0. Did you use outdated tree? I may be wrong, but there seem to be no real bug with QPLAN_QC_NO currently. Just because nobody sets it. So in fact it clears QPLAN_ADMIN, which FWICS is never set at this point. But it is a good to have it fixed nevertheless. Btw, QPLAN_QC_NO doesn't seem to be used in fact. May be remove it? > diff --git a/storage/archive/archive_test.c b/storage/archive/archive_test.c > index d01c1e0..bb052b8 100644 > --- a/storage/archive/archive_test.c > +++ b/storage/archive/archive_test.c > @@ -33,7 +33,7 @@ > > #define ARCHIVE_ROW_HEADER_SIZE 4 > > -#define BUFFER_LEN 1024 + ARCHIVE_ROW_HEADER_SIZE > +#define BUFFER_LEN (1024 + ARCHIVE_ROW_HEADER_SIZE) > > char test_string[BUFFER_LEN]; Alright, there was a bug in this line: assert(write_length != count * BUFFER_LEN); > diff --git a/strings/ctype.c b/strings/ctype.c > index 048fbe3..d8a1dd7 100644 > --- a/strings/ctype.c > +++ b/strings/ctype.c > @@ -264,7 +264,7 @@ static const struct my_cs_file_section_st > } > > #define MY_CS_CSDESCR_SIZE 64 > -#define MY_CS_TAILORING_SIZE 32*1024 > +#define MY_CS_TAILORING_SIZE (32*1024) > #define MY_CS_UCA_VERSION_SIZE 64 > #define MY_CS_CONTEXT_SIZE 64 Was this used at all? May be remove it? Regards, Sergey

2 1

Re: [Maria-developers] [Commits] 62fe87c: MDEV-8663: IF Statement returns multiple values erroneously (or Assertion `!null_value' failed in Item::send(Protocol*, String*))
by Alexander Barkov 04 Sep '15

04 Sep '15

Hi Sanja, From my understanding, the patch fixes consequences of the problem, but not the problem itself. It very likely will still have crashes with slightly modified SQL scripts. The source of the problem is that Item_func_hybrid_result_type::val_decimal() does not return NULL when int_op() or real_op() set null_value to "true". By the way, compare Item_func_hybrid_result_type::val_decimal() with a very similar method Item_func_hybrid_result_type::val_str(). The latter already handle the INT_RESULT and REAL_RESULT cases well and return NULL properly in case of null_value. get_date(), val_int(), val_real() also look fine for me. So the are problems only in val_decimal(). I propose a small patch that fixes the source of the problem. This is the small patch: > @@ -1038,17 +1026,21 @@ my_decimal *Item_func_hybrid_result_type::val_decimal(my_decimal *decimal_value) > case INT_RESULT: > { > longlong result= int_op(); > + if (null_value) > + return NULL; > int2my_decimal(E_DEC_FATAL_ERROR, result, unsigned_flag, decimal_value); > break; > } > case REAL_RESULT: > { > double result= (double)real_op(); > + if (null_value) > + return NULL; > double2my_decimal(E_DEC_FATAL_ERROR, result, decimal_value); > break; > } I also propose an extended patch version, to cover the code with more DBUG_ASSERTs, so in case of the null_value related bugs it crashes much earlier than in Item::send(). It will be easier to debug problems this way. To avoid much code duplication, I added three private helper methods. The extended version of the patch is attached. The scripts from the bug report start to return expected results, with no crashes in DEBUG build, and "mtr --suite=main" works fine. I have not run the entire mtr though, to save time. Greetings. On 09/03/2015 08:00 PM, sanja(a)mariadb.com wrote: > revision-id: 62fe87c5a239c7304b245ce5954c87aecb6382ad (mariadb-5.5.45-1-g62fe87c) > parent(s): fa51f70dc68fe2f3afe943e2c81fcbdb34f16cad > committer: Oleksandr Byelkin > timestamp: 2015-09-03 18:00:43 +0200 > message: > > MDEV-8663: IF Statement returns multiple values erroneously (or Assertion `!null_value' failed in Item::send(Protocol*, String*)) > > keeping contract: NULL value mean NULL pointer in val_str and val_deciman. > > --- > mysql-test/r/func_if.result | 17 +++++++++++++++++ > mysql-test/t/func_if.test | 14 ++++++++++++++ > sql/item_cmpfunc.cc | 6 ++++-- > sql/item_func.cc | 5 +++++ > 4 files changed, 40 insertions(+), 2 deletions(-) > > diff --git a/mysql-test/r/func_if.result b/mysql-test/r/func_if.result > index c7f548a..61f63cc 100644 > --- a/mysql-test/r/func_if.result > +++ b/mysql-test/r/func_if.result > @@ -234,3 +234,20 @@ SELECT if(1, NULL, (SELECT min('hello'))); > if(1, NULL, (SELECT min('hello'))) > NULL > End of 5.2 tests > +# > +# MDEV-8663: IF Statement returns multiple values erroneously > +# (or Assertion `!null_value' failed in Item::send(Protocol*, String*) > +# > +CREATE TABLE `t1` ( > +`datas` VARCHAR(25) NOT NULL > +) ENGINE=InnoDB DEFAULT CHARSET=utf8; > +Warnings: > +Warning 1286 Unknown storage engine 'InnoDB' > +Warning 1266 Using storage engine MyISAM for table 't1' > +INSERT INTO `t1` VALUES ('1,2'), ('2,3'), ('3,4'); > +SELECT IF(FIND_IN_SET('1', `datas`), 1.5, IF(FIND_IN_SET('2', `datas`), 2, NULL)) AS `First`, '1' AS `Second`, '2' AS `Third` FROM `t1`; > +First Second Third > +1.5 1 2 > +2.0 1 2 > +NULL 1 2 > +drop table t1; > diff --git a/mysql-test/t/func_if.test b/mysql-test/t/func_if.test > index 2b89a61..8fdba77 100644 > --- a/mysql-test/t/func_if.test > +++ b/mysql-test/t/func_if.test > @@ -206,6 +206,20 @@ SELECT if(1, NULL, (SELECT min('hello'))); > > --echo End of 5.2 tests > > +--echo # > +--echo # MDEV-8663: IF Statement returns multiple values erroneously > +--echo # (or Assertion `!null_value' failed in Item::send(Protocol*, String*) > +--echo # > +CREATE TABLE `t1` ( > +`datas` VARCHAR(25) NOT NULL > +) ENGINE=InnoDB DEFAULT CHARSET=utf8; > + > +INSERT INTO `t1` VALUES ('1,2'), ('2,3'), ('3,4'); > + > +SELECT IF(FIND_IN_SET('1', `datas`), 1.5, IF(FIND_IN_SET('2', `datas`), 2, NULL)) AS `First`, '1' AS `Second`, '2' AS `Third` FROM `t1`; > + > +drop table t1; > + > --disable_query_log > # Restore timezone to default > set time_zone= @@global.time_zone; > diff --git a/sql/item_cmpfunc.cc b/sql/item_cmpfunc.cc > index 998cb1c..4f07318 100644 > --- a/sql/item_cmpfunc.cc > +++ b/sql/item_cmpfunc.cc > @@ -2692,7 +2692,8 @@ Item_func_if::str_op(String *str) > String *res=arg->val_str(str); > if (res) > res->set_charset(collation.collation); > - null_value=arg->null_value; > + if (null_value=arg->null_value) > + res= NULL; > return res; > } > > @@ -2703,7 +2704,8 @@ Item_func_if::decimal_op(my_decimal *decimal_value) > DBUG_ASSERT(fixed == 1); > Item *arg= args[0]->val_bool() ? args[1] : args[2]; > my_decimal *value= arg->val_decimal(decimal_value); > - null_value= arg->null_value; > + if (null_value= arg->null_value) > + value= NULL; > return value; > } > > diff --git a/sql/item_func.cc b/sql/item_func.cc > index 2f51308..cbb8aa0 100644 > --- a/sql/item_func.cc > +++ b/sql/item_func.cc > @@ -924,6 +924,7 @@ String *Item_func_hybrid_result_type::val_str(String *str) > ltime.time_type= mysql_type_to_time_type(field_type()); > str->length(my_TIME_to_str(&ltime, const_cast<char*>(str->ptr()), decimals)); > str->set_charset(&my_charset_bin); > + DBUG_ASSERT(!null_value); > return str; > } > return str_op(&str_value); > @@ -932,6 +933,7 @@ String *Item_func_hybrid_result_type::val_str(String *str) > case IMPOSSIBLE_RESULT: > DBUG_ASSERT(0); > } > + DBUG_ASSERT(!null_value || (str == NULL)); > return str; > } > > @@ -1071,7 +1073,10 @@ my_decimal *Item_func_hybrid_result_type::val_decimal(my_decimal *decimal_value) > } > String *res; > if (!(res= str_op(&str_value))) > + { > + null_value= 1; > return NULL; > + } > > str2my_decimal(E_DEC_FATAL_ERROR, (char*) res->ptr(), > res->length(), res->charset(), decimal_value); > _______________________________________________ > commits mailing list > commits(a)mariadb.org > https://lists.askmonty.org/cgi-bin/mailman/listinfo/commits >

1 0

[Maria-developers] [GSoC] MDEV-4674 Port InnoDB memcached interface to MariaDB - Merging the code
by Piotr Jurkiewicz 02 Sep '15

02 Sep '15

Hi again, there are some issues which I think needs to be resolved before merging the code. 1. Bundle libevent or not? Memcached daemon is built with the help of libevent library. MySQL decided to bundle this library inside its sources, in 'libevent' top level directory. This allows to compile and run the plugin on machines without installed system-wide libevent library. The bundled libevent is used by default during the compilation of plugin. Optionally, it can be specified in CMake variable to use system-wide libevent instead. It needs to be decided whether MariaDB should bundle libevent as well inside its sources. Possible outcomes: a) Leave bundled libevent as is, in the 'libevent' top level directory. b) Move bundled libevent into the 'daemon_memcached' plugin directory, in order to not clutter the top level directory. c) Remove bundled libevent. After that, system libevent headers will become compile-time dependency and system libevent shared library runtime dependency of the plugin. 2. How to transfer InnoDB API callback array pointer to the plugin? InnoDB Memcached interface engine needs to obtain the pointer to InnoDB API callback array in some way. MySQL originally implemented that in 'sql/sql_plugin.cc:plugin_initialize()' in the following way: https://github.com/piotrjurkiewicz/mariadb-server/commit/4eb8fd9c211f396694… As you can see, this is not a clean solution. Possible outcomes: a) Leave the MySQL's solution as is. b) Figure out a better way to transfer the pointer. 3. Where should I post the documentation? 4. How should I prepare the pull request? Should I rebase my branch on the top of current branch before that? I looking forward for your suggestions and decisions. Piotr

2 1

Re: [Maria-developers] MDEV-8389 patch review (part#1)
by Sergey Petrunia 28 Aug '15

28 Aug '15

Hi Diwas, On Thu, Aug 27, 2015 at 09:23:55PM +0530, Diwas Joshi wrote: > hello sergei, I am attaching two patches with this mail. > MDEV-8389_first is the original patch with mysql_create_table() method used > to create tables. I have fixed the crash that was occurring with queries > like select name from f21('aaa') where name like '%foo%'; So, this patch is > giving the results. > MDEV-8389_second contains use of create_tmp_table() so far the table is > being created, but the queries can't see it so it is still giving errors > like table not present. I guess we'll have to fix the Name Resolution for > queries to be able to use the created table. > > Also both patches have fix for MDEV-8650. I've looked at the patch. Unfortunately, it gives little confidence that it will work in all cases. There are a lot of weird things. To name a few: > + tmp_table_param->schema_table = 1; why? ... > + switch (field->sql_type) > + { > + case MYSQL_TYPE_VARCHAR: > + case MYSQL_TYPE_VAR_STRING: > + case MYSQL_TYPE_STRING: > + case MYSQL_TYPE_SET: > + field->pack_flag= 0; > + break; ... This big switch statement is copied from Create_field::init_for_tmp_table. Why is it copied (and not factored out?) If there is a need to copy the code from that function into here, why do you not also copy this part: pack_flag|= (maybe_null ? FIELDFLAG_MAYBE_NULL : 0) | (is_unsigned ? 0 : FIELDFLAG_DECIMAL); > + TABLE *vtable= create_virtual_tmp_table(thd, sp->m_cols_list); ... > + if (! (table = create_tmp_table(thd, tmp_table_param, *types, > + (ORDER*) 0, 1, 1, > + TMP_TABLE_ALL_COLUMNS, HA_POS_ERROR, > + sp->m_table_alias.str))) why create_virtual_tmp_table and then create_tmp_table? 'vtable' is created and then not freed. I look at the comment near create_virtual_tmp_table() definition and see: The table is created in THD mem_root, so are the table's fields. Consequently, if you don't BLOB fields, you don't need to free it. But what if the table does have blob fields? > + table->file->extra(HA_EXTRA_IGNORE_DUP_KEY); What's this? I grep for HA_EXTRA_IGNORE_DUP_KEY and find this explanation: HA_EXTRA_IGNORE_DUP_KEY: HA_EXTRA_NO_IGNORE_DUP_KEY: Informs the handler to we will not stop the transaction if we get an duplicate key errors during insert/upate. Do we expect to get duplicate key errors when inserting into the table? > +bool execute_table_function(THD *thd, sp_head *sp, List<Item> *value_list) > +{ > + /* > + We never write CALL statements into binlog: > + - If the mode is non-prelocked, each statement will be logged > + separately. > + - If the mode is prelocked, the invoking statement will care > + about writing into binlog. > + So just execute the statement. How is CALL statement relevant to executing a table function? I think at this point it is clear that this patch needs more work before it can be pushed. BR Sergei -- Sergei Petrunia, Software Developer MariaDB Corporation | Skype: sergefp | Blog: http://s.petrunia.net/blog

1 0