- developers - lists.mariadb.org

Re: [Maria-developers] c3d2998a038: MDEV-16130 wrong error message adding AS ROW START to versioned table
by Sergei Golubchik 21 Sep '19

21 Sep '19

Hi, Aleksey! On Aug 30, Aleksey Midenkov wrote: > revision-id: c3d2998a038 (versioning-1.0.3-71-gc3d2998a038) > parent(s): 611488e3d90 > author: Aleksey Midenkov <midenok(a)gmail.com> > committer: Aleksey Midenkov <midenok(a)gmail.com> > timestamp: 2018-05-23 22:45:08 +0300 > message: > > MDEV-16130 wrong error message adding AS ROW START to versioned table > > Closes tempesta-tech/mariadb#494 > > --- > mysql-test/suite/versioning/r/alter.result | 12 ++++++++---- > mysql-test/suite/versioning/t/alter.test | 7 ++++++- > sql/handler.cc | 3 ++- > sql/share/errmsg-utf8.txt | 2 +- > sql/sql_table.cc | 6 ------ > 5 files changed, 17 insertions(+), 13 deletions(-) > > diff --git a/mysql-test/suite/versioning/r/alter.result b/mysql-test/suite/versioning/r/alter.result > index fafcf3c30b0..666420dc2e5 100644 > --- a/mysql-test/suite/versioning/r/alter.result > +++ b/mysql-test/suite/versioning/r/alter.result > @@ -76,13 +76,17 @@ t CREATE TABLE `t` ( > `a` int(11) DEFAULT NULL > ) ENGINE=MyISAM DEFAULT CHARSET=latin1 > alter table t add column trx_start timestamp(6) as row start; > -ERROR HY000: Table `t` is not system-versioned > +ERROR HY000: Can not add system property AS ROW START/END for field `trx_start` This is very strange wording. What is a "system property"? The standard has no such concept. Neither does MariaDB, it's not used anywhere in the manual, as far as I know. Here you can use, for example, ER_VERS_DUPLICATE_ROW_START_END: Duplicate ROW START column `trx_start` > alter table t add system versioning; > show create table t; > Table Create Table > t CREATE TABLE `t` ( > `a` int(11) DEFAULT NULL > ) ENGINE=MyISAM DEFAULT CHARSET=latin1 WITH SYSTEM VERSIONING > +alter table t add column trx_start timestamp(6) as row start; > +ERROR HY000: Can not add system property AS ROW START/END for field `trx_start` > +alter table t modify a int as row start; > +ERROR HY000: Can not add system property AS ROW START/END for field `a` > alter table t add column b int; > show create table t; > Table Create Table > @@ -527,6 +531,6 @@ use test; > # MDEV-15956 Strange ER_UNSUPPORTED_ACTION_ON_GENERATED_COLUMN upon ALTER on versioning column > create or replace table t1 (i int, j int as (i), s timestamp(6) as row start, e timestamp(6) as row end, period for system_time(s,e)) with system versioning; > alter table t1 modify s timestamp(6) as row start; > -ERROR HY000: Can not change system versioning field `s` > +ERROR HY000: Can not add system property AS ROW START/END for field `s` This doesn't look right either. The statement does not add anything, the field `s` is already AS ROW START. Why is it an error at all? > drop database test; > create database test; Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 1

Re: [Maria-developers] [Commits] cde9170709c: MDEV-18648: slave_parallel_mode= optimistic default in 10.5
by Kristian Nielsen 20 Sep '19

20 Sep '19

sujatha <sujatha.sivakumar(a)mariadb.com> writes: Hi Sujutha, > @sql/sql_class.h > Moved 'wait_for_prior_commit(THD *thd)' method inside sql_class.cc > > @sql/sql_class.cc > Added code to check for 'stop_on_error_sub_id' for event groups which get skipped > and don't have any preceding group to wait for. This looks like the wrong fix. The wait_for_commit mechanism is a low-level API, it should not deal with things like stop_on_error_sub_id. > 1,2,3 are scheduled in parallel. > Since the above is true 'skip_event_group=true' is set. Simply call > 'wait_for_prior_commit' to wakeup all waiters. Group '2' didn't had any > waitee and its execution is skipped. Hence its wakeup_error=0.It sends a > positive wakeup signal to '3'. Which commits. This results in a missed > transaction. i.e 33 is missed. I think this the the real problem. 2 is stopping because of error, it must not call wakeup_subsequent_commits() without propagating the error, that breaks the whole semantics of wait_for_commit. Maybe it's enough just to also set rgi->worker_error here, when skip_event_group is set due to error in an earlier group? if (unlikely(entry->stop_on_error_sub_id <= rgi->wait_commit_sub_id)) skip_event_group= true; Then wakeup_subsequent_commits() will correctly inform the following transaction about the error so it doesn't try to commit on its own. There is already similar code in the retry_event_group() function: if (entry->stop_on_error_sub_id == (uint64) ULONGLONG_MAX || rgi->gtid_sub_id < entry->stop_on_error_sub_id) ... else { err= rgi->worker_error= 1; Then all of the changes to sql_class.h and sql_class.cc can be omittet. Also, this fix doesn't seem to belong with the other MDEV-18648 changes, it is unrelated to what the default parallel replication mode is. So please do it in a separate commit. Hope this helps, - Kristian. > diff --git a/sql/sql_class.cc b/sql/sql_class.cc > index 4eab241232b..5ba9c5fe456 100644 > --- a/sql/sql_class.cc > +++ b/sql/sql_class.cc > @@ -7365,6 +7365,33 @@ wait_for_commit::register_wait_for_prior_commit(wait_for_commit *waitee) > } > > > +int wait_for_commit::wait_for_prior_commit(THD *thd) > +{ > + /* > + Quick inline check, to avoid function call and locking in the common case > + where no wakeup is registered, or a registered wait was already signalled. > + */ > + if (waitee) > + return wait_for_prior_commit2(thd); > + else > + { > + if (wakeup_error) > + my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); > + else > + { > + rpl_group_info* rgi= thd->rgi_slave; > + if (rgi && rgi->is_parallel_exec && > + rgi->parallel_entry->stop_on_error_sub_id < (uint64)ULONGLONG_MAX && > + rgi->gtid_sub_id >= rgi->parallel_entry->stop_on_error_sub_id) > + { > + my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); > + wakeup_error= ER_PRIOR_COMMIT_FAILED; > + } > + } > + return wakeup_error; > + } > +} > + > /* > Wait for commit of another transaction to complete, as already registered > with register_wait_for_prior_commit(). If the commit already completed, > @@ -7387,6 +7414,17 @@ wait_for_commit::wait_for_prior_commit2(THD *thd) > { > if (wakeup_error) > my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); > + else > + { > + rpl_group_info* rgi= thd->rgi_slave; > + if (rgi && rgi->is_parallel_exec && > + rgi->parallel_entry->stop_on_error_sub_id < (uint64)ULONGLONG_MAX && > + rgi->gtid_sub_id >= rgi->parallel_entry->stop_on_error_sub_id) > + { > + my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); > + wakeup_error= ER_PRIOR_COMMIT_FAILED; > + } > + } > goto end; > } > /* > diff --git a/sql/sql_class.h b/sql/sql_class.h > index 72cb8148895..0a0a1aa9fa1 100644 > --- a/sql/sql_class.h > +++ b/sql/sql_class.h > @@ -2062,21 +2062,6 @@ struct wait_for_commit > bool commit_started; > > void register_wait_for_prior_commit(wait_for_commit *waitee); > - int wait_for_prior_commit(THD *thd) > - { > - /* > - Quick inline check, to avoid function call and locking in the common case > - where no wakeup is registered, or a registered wait was already signalled. > - */ > - if (waitee) > - return wait_for_prior_commit2(thd); > - else > - { > - if (wakeup_error) > - my_error(ER_PRIOR_COMMIT_FAILED, MYF(0)); > - return wakeup_error; > - } > - } > void wakeup_subsequent_commits(int wakeup_error_arg) > { > /* > @@ -2123,6 +2108,7 @@ struct wait_for_commit > > void wakeup(int wakeup_error); > > + int wait_for_prior_commit(THD *thd); > int wait_for_prior_commit2(THD *thd); > void wakeup_subsequent_commits2(int wakeup_error); > void unregister_wait_for_prior_commit2();

1 0

[Maria-developers] mdev-19956 - error messages
by Alexander Barkov 19 Sep '19

19 Sep '19

Hi Igor, We discussed the problem with error messages with Sergei and both agreed that a new error message like: Subquery is not allowed in '%s' will be more informative, and it's easy to do. Why not replace: bool expr_allows_subselect; to const char *clause_that_disallows_subselect; ? So we can change the grammar to something like this: | opt_generated_always AS { Lex->clause_that_disallows_subselect= "GENERATED ALWAYS AS"; } virtual_column_func { Lex->clause_that_disallows_subselect= NULL; } The same for other places where we disallow subselects. This can be done in 10.4 before your patch, as a separate commit.

1 0

Re: [Maria-developers] Please review mdev-19956
by Alexander Barkov 19 Sep '19

19 Sep '19

Hi Igor, I have the following suggestions so far. Can you please fix them and replace the commit in bb-10.4-igor? 1. In statements where subquery is not allowed, the error message become confusing: EXECUTE stmt USING (SELECT 1); < ERROR 1064 (42000): ... syntax to use near 'SELECT 1)' at line 1 > ERROR 1064 (42000): ... syntax to use near ')' at line 1 Please change SELECT and WITH to have the '<kwd>' type, like this: %token <kwd> SELECT_SYM /* SQL-2003-R */ %token <kwd> WITH /* SQL-2003-R */ Then in the grammar like this: query_specification_start: SELECT_SYM { SELECT_LEX *sel; LEX *lex= Lex; if (!(sel= lex->alloc_select(TRUE)) || lex->push_select(sel)) MYSQL_YYABORT; sel->init_select(); sel->braces= FALSE; you will be able to use $1.pos() to get the position of the keyword, and later use this position to generate the error properly. 2. The change in sql_cte.cc is not relevant to this task. Please make a separate MDEV for this and push it to 10.4, then rebase MDEV-19956. 3. There are a few mistakes in comments: + <query expression body> :: + <query term> + | <query expression body> UNION [ ALL | DISTINCT ] <query term> + | <query expression body> EXCEPT [ DISTINCT ] + Consider the production rule of the SQL Standard + subquery: + '(' query_expression_no_with_clause ')' + The latter can be re-written into + subquery: + '(' query_expression_body_ext_parens ')' + | '(' with_clause query_expression_no_with_clause ')' Please fix them as discussed on slack. Thanks.

1 0

[Maria-developers] Implementation of Threadpool
by GUESNET, ETIENNE (ext) 18 Sep '19

18 Sep '19

Hi, I am implementing a threadpool system to AIX. The AIX equivalent of epoll / kqueue on AIX is pollset (and IOCP, but partial implementation only). However, pollset has only a level-trigger mode and MariaDB needs edge-trigger (see comments of sql/threadpoll_generic.h file). Adding a pollset support in MariaDB would be difficult, and probably not so efficient, as we need to simulate the edge-trigger behavior. Obviously, AIX has poll and select support. MariaDB has not. Is there a reason to don't implement threadpoll through poll or select? No interest? Performance issues? MariaDB currently works on AIX without threadpool; in term of efficiency, do you know what can be obtained using threadpool with poll/select or a more modern solution? As far I know, SunOS/Solaris/Illumos threadpoll system (called "port") is also level-trigger only, but I do not find specific functions to manage this. Thanks! Etienne Guesnet.

4 6

[Maria-developers] Congratulations on a successful GSoC!
by Vicențiu Ciorbaru 05 Sep '19

05 Sep '19

Hi! By now, all of you have probably received the results of this year's Google Summer of Code. I want to congratulate you all for the work that you have done this year, students and mentors alike. We have had 4 projects this year: 1. EXCEPT ALL and INTERSECT ALL operations - Ruihang Xia https://github.com/MariaDB/server/pull/1378 2. INSERT ... RETURNING - Rucha Deodar https://medium.com/@ruchad1998/google-summer-of-code-2019-add-returning-to-… 3. UPDATE ... RETURNING - Miroslav Koberskii https://gist.github.com/Mup0c/43c781e2135e55bf5126e0f8c60e2a24 4. Support for indexes on expressions - Alexey Mogilyovkin https://gist.github.com/ZeroICQ/f0c35a20cae7065368dae31d7ad9b001 Implementing these projects certainly took a lot of effort. The first one is already merged within MariaDB and the rest will probably follow, after a bit more polishing. The MariaDB community would be grateful if you stuck around to watch over your projects, extend them and improve them over time. I am sure there was a lot to learn from this experience. Now that you have gotten a feel to what it means to contribute significant features to MariaDB you are in a great position to help others! In this sense, I encourage you to keep being active in the community and share what you have learned. :) Thank you for a great summer! Vicențiu Ciorbaru GSoC MariaDB Foundation Admin

1 0

[Maria-developers] Review for MDEV-19708 RBR replication loses data silently ignoring important column attributes
by Alexander Barkov 02 Sep '19

02 Sep '19

Hi Sachin, I've reviewed the code in bb-10.5-19708. It looks good. I suggest some cleanups. - Remove Field::binlog_type_info_slave, it's not used. - Remove Field::binlog_type_info_master. Please don't cache Binlog_type_info inside the Field itself. Let the caller code cache it in an array of Binlog_type_info pointers. - Change this method in Field: virtual Binlog_type_info *binlog_type_info(); to virtual Binlog_type_info *binlog_type_info(MEM_ROOT *) const; Let's pass mem_root as a parameter instead of using table->mem_root. Please add the 'const' qualifier! - Restore the 'const' qualifier to Field::enum_conv_type rpl_conv_type_from - Instead of doing this: Binlog_type_info * Field_new_decimal::binlog_type_info() { Field_num::binlog_type_info(); this->binlog_type_info_master->m_metadata= precision + (decimals() << 8); this->binlog_type_info_master->m_metadata_size= 2; return this->binlog_type_info_master; } Please add a number of Binlog_type_info constructor helpers, in addition to the current one. For example, for numeric types this would be good: Binlog_type_info(uchar type_code, uint16 metadata, uint8 metadata_size, binlog_signess_t signess) :m_type_code(type_code), m_metadata(metadata), m_metadata_size(metadata_size), m_signess(signess), m_cs(NULL), m_enum_typelib(NULL), m_set_typelib(NULL), m_geom_type(GEOM_GEOMETRY) { }; so you can do something simple like this: Binlog_type_info Field_new_decimal::binlog_type_info(MEM_ROOT *root) const { return new (root) Binlog_type_info(type(), precision + (decimals() << 8), 2, signess_arg); } - It seems you always set signess to SIGNESS_NOT_RELEVANT. Where is signess set to SIGNED or UNSIGNED? - This is something we've never used before: +#include <vector> +#include <string> +#include <functional> +#include <memory> We need to discuss this with Serg. Is it possible to avoid this? - There is a new test file mysql-test/suite/rpl/t/rpl_mdev_19708.test.diff but I could not find a corresponding .result.diff. Where is it? - Please use error names instead of numbers in tests: Instead of: --error 1231 SET GLOBAL binlog_row_metadata = -1; it should be: --error ER_WRONG_VALUE_FOR_VAR SET GLOBAL binlog_row_metadata = -1; You can find error names in include/mysqld_ername.h Thanks.

2 1

Re: [Maria-developers] 899c0b3ec6e: Part2: MDEV-18156 Assertion `0' failed or `btr_validate_index(index, 0, false)' in row_upd_sec_index_entry or error code 126: Index is corrupted upon DELETE with PAD_CHAR_TO_FULL_LENGTH
by Sergei Golubchik 30 Aug '19

30 Aug '19

Hi, Alexander! On Aug 30, Alexander Barkov wrote: > revision-id: 899c0b3ec6e (mariadb-10.2.26-51-g899c0b3ec6e) > parent(s): e4415549e53 > author: Alexander Barkov <bar(a)mariadb.com> > committer: Alexander Barkov <bar(a)mariadb.com> > timestamp: 2019-08-29 12:35:19 +0400 > message: > > Part2: MDEV-18156 Assertion `0' failed or `btr_validate_index(index, 0, false)' in row_upd_sec_index_entry or error code 126: Index is corrupted upon DELETE with PAD_CHAR_TO_FULL_LENGTH > > This patch allows the server to open old tables that have > "bad" generated columns (i.e. indexed virtual generated columns, > persistent generated columns) that depend on sql_mode, > for general things like SELECT, INSERT, DROP, etc. > Warning are issued in such cases. > > Only these commands are now disallowed and return an error: > - CREATE TABLE introducing a "bad" generated column > - ALTER TABLE introducing a "bad" generated column > - CREATE INDEX introdicing a "bad" generated column > (i.e. adding an index on a virtual generated column > that depends on sql_mode). > > Note, these commands are allowed: > - ALTER TABLE removing a "bad" generate column > - ALTER TABLE removing an index from a "bad" virtual generated column > - DROP INDEX removing an index from a "bad" virtual generated column > but only if the table does not have any "bad" columns as a result. > > diff --git a/sql/field.cc b/sql/field.cc > index e2b745743d2..b720db19ebf 100644 > --- a/sql/field.cc > +++ b/sql/field.cc > @@ -1428,16 +1428,24 @@ void Field::load_data_set_value(const char *pos, uint length, > } > > > -void Field::error_generated_column_function_is_not_allowed(THD *thd) const > +void Field::error_generated_column_function_is_not_allowed(THD *thd, > + bool error) const > { > StringBuffer<64> tmp; > vcol_info->expr->print(&tmp, (enum_query_type) > (QT_TO_SYSTEM_CHARSET | > QT_ITEM_IDENT_SKIP_DB_NAMES | > QT_ITEM_IDENT_SKIP_TABLE_NAMES)); > - my_error(ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED, MYF(0), > - tmp.c_ptr(), vcol_info->get_vcol_type_name(), > - const_cast<const char*>(field_name)); > + if (error) > + my_error(ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED, MYF(0), > + tmp.c_ptr(), vcol_info->get_vcol_type_name(), > + const_cast<const char*>(field_name)); > + else > + push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, > + ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED, > + ER_THD(thd, ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED), > + tmp.c_ptr(), vcol_info->get_vcol_type_name(), > + const_cast<const char*>(field_name)); an easier way of doing it would be my_error(ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED, MYF(error ? 0 : ME_WARNING), tmp.c_ptr(), vcol_info->get_vcol_type_name(), const_cast<const char*>(field_name)); > } > > > diff --git a/sql/table.h b/sql/table.h > index 7786679982f..23a61345242 100644 > --- a/sql/table.h > +++ b/sql/table.h > @@ -325,6 +325,20 @@ enum tmp_table_type > }; > enum release_type { RELEASE_NORMAL, RELEASE_WAIT_FOR_DROP }; > > + > +enum vcol_init_mode > +{ > + VCOL_INIT_DEPENDENCY_FAILURE_IS_WARNING= 1, > + VCOL_INIT_DEPENDENCY_FAILURE_IS_ERROR= 2 > + /* > + There will be a new flags soon, better say "There may be new flags here" because may be there won't be any new flags soon after all, if this your fix will be sufficient :) > + e.g. to automatically remove sql_mode dependency: > + GENERATED ALWAYS AS (char_col) -> > + GENERATED ALWAYS AS (RTRIM(char_col)) > + */ > +}; > + > + > enum enum_vcol_update_mode > { > VCOL_UPDATE_FOR_READ= 0, Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

1 0

Re: [Maria-developers] 5ba0b11c5d1: MDEV-20257 Fix crash in Grant_table_base::init_read_record
by Sergei Golubchik 29 Aug '19

29 Aug '19

Hi, Robert! On Aug 28, Robert Bindar wrote: > revision-id: 5ba0b11c5d1 (mariadb-10.4.7-35-g5ba0b11c5d1) > parent(s): 4a490d1a993 > author: Robert Bindar <robert(a)mariadb.org> > committer: Robert Bindar <robert(a)mariadb.org> > timestamp: 2019-08-27 11:38:52 +0300 > message: > > MDEV-20257 Fix crash in Grant_table_base::init_read_record > > The bug shows up when a 10.4+ server starts on a <10.4 datadir > with a crashed mysql.user table. > Grant_tables::open_and_lock tries to open the list of requested > tables (host, db, global_priv,..) and because global_priv doesn't > exist (pre-10.4 datadir), it falls back to opening mysql.user. > The first open_tables call for [host, db, global_priv,..] works > just fine. The second open_tables call (trying to open mysql.user > alone) sees the crashed user table and performs 3 steps: > - closes all the open tables > - attempts a table fix for mysql.user (succeeds) > - opens the tables initially passed as arguments > In an ideal world, the first step should only close the tables > passed as arguments (i.e. mysql.user), but for some reasons > close_tables_for_reopen makes a close_thread_tables call which > closes everything in THD::open_tables (nevertheless, this is > probably the intended behavior). > This side effect causes all the tables opened in the first > open_tables call to now be closed and Grant_table_base::init_read_record > tries to read from a released table. > > diff --git a/mysql-test/main/mdev20257.test b/mysql-test/main/mdev20257.test > new file mode 100644 > index 00000000000..8506292d57c > --- /dev/null > +++ b/mysql-test/main/mdev20257.test > @@ -0,0 +1,51 @@ > +--source include/not_embedded.inc > +--echo # > +--echo # MDEV 20257 Server crashes in Grant_table_base::init_read_record > +--echo # upon crash-upgrade > +--echo # > + > +let $MYSQLD_DATADIR= `select @@datadir`; > + > +rename table mysql.global_priv to mysql.global_priv_bak; > +rename table mysql.user to mysql.user_bak; > +rename table mysql.db to mysql.db_bak; > +rename table mysql.proxies_priv to mysql.proxies_priv_bak; > +rename table mysql.roles_mapping to mysql.roles_mapping_bak; > + > +--source include/shutdown_mysqld.inc > + > +# Bring in a crashed user table > +# Ideally we should've copied only the crashed mysql.user, but this > +# would make mysqld crash in some other place before hitting the > +# crash spot described in MDEV-20257 which is what we're trying to > +# test against. I don't understand that. Where does it crash then? What difference does it make that you replace all tables - are they all corrupted? > +--copy_file std_data/mdev20257/user.frm $MYSQLD_DATADIR/mysql/user.frm > +--copy_file std_data/mdev20257/user.MYD $MYSQLD_DATADIR/mysql/user.MYD > +--copy_file std_data/mdev20257/user.MYI $MYSQLD_DATADIR/mysql/user.MYI > + > +--copy_file std_data/mdev20257/host.frm $MYSQLD_DATADIR/mysql/host.frm > +--copy_file std_data/mdev20257/host.MYD $MYSQLD_DATADIR/mysql/host.MYD > +--copy_file std_data/mdev20257/host.MYI $MYSQLD_DATADIR/mysql/host.MYI > + > +--copy_file std_data/mdev20257/db.frm $MYSQLD_DATADIR/mysql/db.frm > +--copy_file std_data/mdev20257/db.MYD $MYSQLD_DATADIR/mysql/db.MYD > +--copy_file std_data/mdev20257/db.MYI $MYSQLD_DATADIR/mysql/db.MYI > +--copy_file std_data/mdev20257/db.opt $MYSQLD_DATADIR/mysql/db.opt > + > +--source include/start_mysqld.inc > + > +call mtr.add_suppression("Table \'\..mysql\.user\' is marked as crashed and should be repaired"); > +call mtr.add_suppression("Checking table: \'\..mysql\.user\'"); > +call mtr.add_suppression("mysql.user: 1 client is using or hasn't closed the table properly"); > +call mtr.add_suppression("Missing system table mysql..*; please run mysql_upgrade to create it"); > + > +# Cleanup > +drop table mysql.user; > +drop table mysql.db; > +drop table mysql.host; > +rename table mysql.global_priv_bak to mysql.global_priv; > +rename table mysql.user_bak to mysql.user; > +rename table mysql.db_bak to mysql.db; > +rename table mysql.proxies_priv_bak to mysql.proxies_priv; > +rename table mysql.roles_mapping_bak to mysql.roles_mapping; > + > diff --git a/mysql-test/std_data/mdev20257/db.opt b/mysql-test/std_data/mdev20257/db.opt > --- /dev/null > +++ b/mysql-test/std_data/mdev20257/db.opt > @@ -0,0 +1,2 @@ > +default-character-set=latin1 > +default-collation=latin1_swedish_ci why is that? > diff --git a/sql/sql_acl.cc b/sql/sql_acl.cc > index 847d2bd777b..e028d084703 100644 > --- a/sql/sql_acl.cc > +++ b/sql/sql_acl.cc > - DBUG_RETURN(res); > + first = build_open_list(tables, which_tables, lock_type, true); > + > + uint counter; > + if (int rv= really_open(thd, first, &counter)) > + DBUG_RETURN(rv); > + > + TABLE *user_table= tables[USER_TABLE].table; > + if ((which_tables & Table_user) && !user_table) > + { > + close_thread_tables(thd); > + first = build_open_list(tables, which_tables, lock_type, false); Hmm... Here you always close/reopen all tables, just to account for the case when the user table is found corrupted. I'd argue that a corrupted user table happens rarely (compared to the non-corrupted), so it makes sense to optimize for the normal use case. That is, only open the user table as before. After it's opened you check if other tables are still opened - if they aren't you can close and reopen everything. This way you only close/reopen all tables if the user table was corrupted. Makes sense? > + if (int rv= really_open(thd, first, &counter)) > + DBUG_RETURN(rv); > + > + p_user_table= &m_user_table_tabular; > + user_table= tables[USER_TABLE + 1].table; > + } Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 1

Re: [Maria-developers] 5f1cd8fe7a5: MDEV-18156 Assertion `0' failed or `btr_validate_index(index, 0, false)' in row_upd_sec_index_entry or error code 126: Index is corrupted upon DELETE with PAD_CHAR_TO_FULL_LENGTH
by Sergei Golubchik 28 Aug '19

28 Aug '19

Hi, Alexander! On Aug 28, Alexander Barkov wrote: > revision-id: 5f1cd8fe7a5 (mariadb-10.2.26-50-g5f1cd8fe7a5) > parent(s): 9de2e60d749 > author: Alexander Barkov <bar(a)mariadb.com> > committer: Alexander Barkov <bar(a)mariadb.com> > timestamp: 2019-08-26 15:28:32 +0400 > message: > > MDEV-18156 Assertion `0' failed or `btr_validate_index(index, 0, false)' in row_upd_sec_index_entry or error code 126: Index is corrupted upon DELETE with PAD_CHAR_TO_FULL_LENGTH > This is not a trivial one-liner fix. Before pushing, please add a comment explaning the problem and your fix. > > diff --git a/mysql-test/suite/vcol/r/vcol_sql_mode.result b/mysql-test/suite/vcol/r/vcol_sql_mode.result > new file mode 100644 > index 00000000000..6af46dffb80 > --- /dev/null > +++ b/mysql-test/suite/vcol/r/vcol_sql_mode.result > @@ -0,0 +1,234 @@ > +# > +# Start of 10.2 tests > +# > +# > +# MDEV-18156 Assertion `0' failed or `btr_validate_index(index, 0, false)' in row_upd_sec_index_entry or error code 126: Index is corrupted upon DELETE with PAD_CHAR_TO_FULL_LENGTH > +# > +# > +# PAD_CHAR_TO_FULL_LENGTH + various virtual column data types > +# > +CREATE TABLE t1 (a CHAR(5), v CHAR(5) AS (a) VIRTUAL, KEY(v)); > +DROP TABLE t1; please add SHOW CREATE after every CREATE. To show what automatic expression rewriting does (or does not) happen here. > +CREATE TABLE t1 (a CHAR(5), v INT AS (a) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v TIME AS (a) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (c CHAR(8), v BINARY(8) AS (c), KEY(v)); > +ERROR HY000: Function or expression '`c`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; Better do --enable_warnings at the beginning of the test. It'll serve two purposes - you won't need to do SHOW WARNINGS manually and it'll make clear when there are no warnings after a CREATE. > +Level Code Message > +Error 1901 Function or expression '`c`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH Let's make it a proper sentence: "Expression `c` implicitly depends on the @@sql_mode value PAD_CHAR_TO_FULL_LENGTH" > +CREATE TABLE t1 (a CHAR(5), v BIT(64) AS (a) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (a) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v TEXT AS (a) VIRTUAL, KEY(v(100))); > +ERROR HY000: Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a`' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# PAD_CHAR_TO_FULL_LENGTH + TRIM resolving dependency > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RTRIM(a)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v TEXT AS (RTRIM(a)) VIRTUAL, KEY(v(100))); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(TRAILING ' ' FROM a)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v TEXT AS (TRIM(TRAILING ' ' FROM a)) VIRTUAL, KEY(v(100))); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(BOTH ' ' FROM a)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v TEXT AS (TRIM(BOTH ' ' FROM a)) VIRTUAL, KEY(v(100))); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(TRAILING NULL FROM a)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(BOTH NULL FROM a)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +# PAD_CHAR_TO_FULL_LENGTH + TRIM not resolving dependency > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(LEADING ' ' FROM a)) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'trim(leading ' ' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` nice :) > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(leading ' ' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v TEXT AS (TRIM(LEADING ' ' FROM a)) VIRTUAL, KEY(v(100))); > +ERROR HY000: Function or expression 'trim(leading ' ' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(leading ' ' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(TRAILING '' FROM a)) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'trim(trailing '' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(trailing '' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(BOTH '' FROM a)) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'trim(both '' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(both '' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(TRAILING 'x' FROM a)) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'trim(trailing 'x' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(trailing 'x' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (TRIM(BOTH 'x' FROM a)) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'trim(both 'x' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(both 'x' from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# PAD_CHAR_TO_FULL_LENGTH + TRIM(... non_constant FROM a) > +CREATE TABLE t1 ( > +a CHAR(5), > +b CHAR(5), > +v TEXT AS (TRIM(TRAILING b FROM a)) VIRTUAL, KEY(v(100))); > +ERROR HY000: Function or expression 'trim(trailing `b` from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'trim(trailing `b` from `a`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# PAD_CHAR_TO_FULL_LENGTH + RPAD resolving dependency > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RPAD(a,5,' ')) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RPAD(a,6,' ')) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RPAD(a,6,NULL)) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RPAD(a,NULL,' ')) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +# PAD_CHAR_TO_FULL_LENGTH + RPAD not resolving dependency > +CREATE TABLE t1 (a CHAR(5), v VARCHAR(5) AS (RPAD(a,4,' ')) VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression 'rpad(`a`,4,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'rpad(`a`,4,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 ( > +a CHAR(5), > +b CHAR(5), > +v VARCHAR(5) AS (RPAD(a,NULL,b)) VIRTUAL, > +KEY(v) > +); > +ERROR HY000: Function or expression 'rpad(`a`,NULL,`b`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'rpad(`a`,NULL,`b`)' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# PAD_CHAR_TO_FULL_LENGTH + comparison > +CREATE TABLE t1 (a CHAR(5), v INT AS (a='a') VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 ( > +a CHAR(5) CHARACTER SET latin1 COLLATE latin1_nopad_bin, > +v INT AS (a='a') VIRTUAL, KEY(v) > +); > +ERROR HY000: Function or expression '`a` = 'a'' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a` = 'a'' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# PAD_CHAR_TO_FULL_LENGTH + LIKE > +CREATE TABLE t1 (a CHAR(5), v INT AS (a LIKE 'a%') VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v INT AS (a LIKE NULL) VIRTUAL, KEY(v)); > +DROP TABLE t1; > +CREATE TABLE t1 (a CHAR(5), v INT AS (a LIKE 'a') VIRTUAL, KEY(v)); > +ERROR HY000: Function or expression '`a` like 'a'' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a` like 'a'' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# > +# Testing NO_UNSIGNED_SUBTRACTION > +# > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c INT GENERATED ALWAYS AS (a-b) VIRTUAL, > +KEY (c) > +); > +ERROR HY000: Function or expression '`a` - `b`' cannot be used in the GENERATED ALWAYS AS clause of `c` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a` - `b`' cannot be used in the GENERATED ALWAYS AS clause of `c` > +Warning 1105 depends on system variable @@sql_mode value NO_UNSIGNED_SUBTRACTION > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c INT GENERATED ALWAYS AS (CAST(a AS SIGNED)-b) VIRTUAL, > +KEY (c) > +); > +ERROR HY000: Function or expression 'cast(`a` as signed) - `b`' cannot be used in the GENERATED ALWAYS AS clause of `c` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'cast(`a` as signed) - `b`' cannot be used in the GENERATED ALWAYS AS clause of `c` > +Warning 1105 depends on system variable @@sql_mode value NO_UNSIGNED_SUBTRACTION > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c INT GENERATED ALWAYS AS (a-CAST(b AS SIGNED)) VIRTUAL, > +KEY (c) > +); > +ERROR HY000: Function or expression '`a` - cast(`b` as signed)' cannot be used in the GENERATED ALWAYS AS clause of `c` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression '`a` - cast(`b` as signed)' cannot be used in the GENERATED ALWAYS AS clause of `c` > +Warning 1105 depends on system variable @@sql_mode value NO_UNSIGNED_SUBTRACTION > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c INT GENERATED ALWAYS AS (CAST(a AS SIGNED)-CAST(b AS SIGNED)) VIRTUAL, > +KEY (c) > +); > +DROP TABLE t1; > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c INT GENERATED ALWAYS AS (CAST(a AS DECIMAL(20,0))-CAST(b AS DECIMAL(20,0))) VIRTUAL, > +KEY (c) > +); > +DROP TABLE t1; > +# > +# Comnination: PAD_CHAR_TO_FULL_LENGTH + NO_UNSIGNED_SUBTRACTION > +# > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c CHAR(5), > +v VARCHAR(5) GENERATED ALWAYS AS (RPAD(c,a-b,' ')) VIRTUAL, > +KEY (v) > +); > +ERROR HY000: Function or expression 'rpad(`c`,`a` - `b`,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'rpad(`c`,`a` - `b`,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value NO_UNSIGNED_SUBTRACTION > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +CREATE TABLE t1 ( > +a INT UNSIGNED, > +b INT UNSIGNED, > +c CHAR(5), > +v VARCHAR(5) GENERATED ALWAYS AS (RPAD(c,CAST(a AS DECIMAL(20,1))-b,' ')) VIRTUAL, > +KEY (v) > +); > +ERROR HY000: Function or expression 'rpad(`c`,cast(`a` as decimal(20,1)) - `b`,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +SHOW WARNINGS; > +Level Code Message > +Error 1901 Function or expression 'rpad(`c`,cast(`a` as decimal(20,1)) - `b`,' ')' cannot be used in the GENERATED ALWAYS AS clause of `v` > +Warning 1105 depends on system variable @@sql_mode value PAD_CHAR_TO_FULL_LENGTH > +# > +# End of 10.2 tests > +# > diff --git a/sql/field.cc b/sql/field.cc > index e9eaa440952..28caea78c7d 100644 > --- a/sql/field.cc > +++ b/sql/field.cc > @@ -1428,6 +1428,35 @@ void Field::load_data_set_value(const char *pos, uint length, > } > > > +void Field::error_generated_column_function_is_not_allowed(THD *thd) const > +{ > + StringBuffer<64> tmp; > + vcol_info->expr->print(&tmp, QT_TO_SYSTEM_CHARSET); > + my_error(ER_GENERATED_COLUMN_FUNCTION_IS_NOT_ALLOWED, MYF(0), > + tmp.c_ptr(), vcol_info->get_vcol_type_name(), > + (const char *) field_name /*add .str on merge to 10.3 */); better const_cast<const char*>(field_name) and it won't need a comment - the compiler will catch it > +} > + > + this needs a comment. May be as simple as "see sql_mode.h" btw, you might need this one-line comment in few other places too > +bool Field::check_vcol_sql_mode_dependency(THD *thd) const > +{ > + DBUG_ASSERT(vcol_info); > + if (!(part_of_key.is_clear_all() && key_start.is_clear_all())) 1. why not for persistent? 2. why both part_of_key and key_start? I though part_of_key means "anywhere in the key", that is it's a superset of key_start. > + { > + Sql_mode_dependency dep= > + vcol_info->expr->value_depends_on_sql_mode() & > + Sql_mode_dependency(~0, ~can_handle_sql_mode_dependency_on_store()); > + if (dep) > + { > + error_generated_column_function_is_not_allowed(thd); > + dep.push_dependency_warnings(thd); > + return true; > + } > + } > + return false; > +} > + > + > /** > Numeric fields base class constructor. > */ > diff --git a/sql/item_func.h b/sql/item_func.h > index 81a7e948085..5be62427852 100644 > --- a/sql/item_func.h > +++ b/sql/item_func.h > @@ -773,11 +773,16 @@ class Item_func_plus :public Item_func_additive_op > > class Item_func_minus :public Item_func_additive_op > { > + Sql_mode_dependency m_sql_mode_dependency; > public: > Item_func_minus(THD *thd, Item *a, Item *b): > Item_func_additive_op(thd, a, b) {} > const char *func_name() const { return "-"; } > enum precedence precedence() const { return ADD_PRECEDENCE; } > + Sql_mode_dependency value_depends_on_sql_mode() const > + { > + return m_sql_mode_dependency; 1. no soft_to_hard here? 2. If you do need soft_to_hard here, then, may be, better to do soft_to_hard not in value_depends_on_sql_mode() at all, but just in one place - in Field::check_vcol_sql_mode_dependency() ? > + } > longlong int_op(); > double real_op(); > my_decimal *decimal_op(my_decimal *); > diff --git a/sql/item_strfunc.cc b/sql/item_strfunc.cc > index 753b6134419..846920bc8c2 100644 > --- a/sql/item_strfunc.cc > +++ b/sql/item_strfunc.cc > @@ -2103,6 +2103,38 @@ void Item_func_trim::print(String *str, enum_query_type query_type) > } > > > +/* > + RTRIM(expr) > + TRIM(TRAILING ' ' FROM expr) > + remove argument's soft dependency on PAD_CHAR_TO_FULL_LENGTH: > +*/ > +Sql_mode_dependency Item_func_trim::value_depends_on_sql_mode() const > +{ > + DBUG_ASSERT(fixed); > + if (arg_count == 1) // RTRIM(expr) > + return (args[0]->value_depends_on_sql_mode() & > + Sql_mode_dependency(~0, ~MODE_PAD_CHAR_TO_FULL_LENGTH)). > + soft_to_hard(); > + // TRIM(... FROM expr) > + DBUG_ASSERT(arg_count == 2); > + if (!args[1]->value_depends_on_sql_mode_const_item()) > + return Item_func::value_depends_on_sql_mode(); > + StringBuffer<64> trimstrbuf; > + String *trimstr= args[1]->val_str(&trimstrbuf); > + if (!trimstr) > + return Sql_mode_dependency(); // will return NULL > + if (trimstr->length() == 0) > + return Item_func::value_depends_on_sql_mode(); // will trim nothing > + if (trimstr->lengthsp() != 0) > + return Item_func::value_depends_on_sql_mode(); // will trim not only spaces what about TRIM(TRAILING ' ' FROM expr) ? it'll trim an even number of spaces, might leave one space untrimmed. > + // TRIM(TRAILING ' ' FROM expr) > + return ((args[0]->value_depends_on_sql_mode() | > + args[1]->value_depends_on_sql_mode()) & > + Sql_mode_dependency(~0, ~MODE_PAD_CHAR_TO_FULL_LENGTH)). > + soft_to_hard(); > +} > + > + > /* Item_func_password */ > > bool Item_func_password::fix_fields(THD *thd, Item **ref) Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 1