- developers - lists.mariadb.org

Re: [Maria-developers] 88f8aa20bba: MDEV-23486 RBR can bypass secure_timestamp=YES
by Sergei Golubchik 26 Mar '22

26 Mar '22

Hi, Aleksey! On Jan 09, Aleksey Midenkov wrote: > revision-id: 88f8aa20bba (mariadb-10.3.31-79-g88f8aa20bba) > parent(s): d5451867af1 > author: Aleksey Midenkov > committer: Aleksey Midenkov > timestamp: 2021-12-23 04:15:39 +0300 > message: > > MDEV-23486 RBR can bypass secure_timestamp=YES > > Pass to a slave whether master table is system-versioned via > TM_SYSTEM_VERSIONED flag. That deprecates m_vers_from_plain > detection. If secure_timestamp == YES or master did not supply row_end > data treat the row as current data and generate timestamps for it. I don't understand why does it matter whether the master is system versioned or not. secure_timestamp=YES means the table has to store actual timestamps when rows were changed. It doesn't trust the master to set timestamps. So it should not matter whether master table is versioned or not, if you don't trust the master, you cannot trust its TM_SYSTEM_VERSIONED flag either. Even if the table was system versioned on the master, the master could've had secure_timestamp=NO and a completely fake history. Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 1

Re: [Maria-developers] MDEV-24560 SIGSEGV in st_join_table::cleanup
by Sergey Petrunia 25 Mar '22

25 Mar '22

Hi Oleg, > commit 4f41909c3ee1d7e42c56947c9a9156fb78c22d8a > Author: Oleg Smirnov <olernov(a)gmail.com> > Date: Fri Mar 11 21:18:34 2022 +0700 > > MDEV-24560 SIGSEGV in st_join_table::cleanup > > If JOIN::create_postjoin_aggr_table encounters errors during execution > then free_tmp_table() is then called twice for JOIN_TAB::aggr. > The solution is to initialize JOIN_TAB::aggr and JOIN_TAB::table only > on successful completion of JOIN::create_postjoin_aggr_table > > diff --git a/sql/sql_select.cc b/sql/sql_select.cc > index c044656e916..8c07014d981 100644 > --- a/sql/sql_select.cc > +++ b/sql/sql_select.cc > @@ -4037,10 +4037,6 @@ JOIN::create_postjoin_aggr_table(JOIN_TAB *tab, List<Item> *table_fields, > !tables_list); > if (tab > join_tab) > (tab - 1)->next_select= sub_select_postjoin_aggr; > - if (!(tab->aggr= new (thd->mem_root) AGGR_OP(tab))) > - goto err; > - tab->table= table; > - table->reginfo.join_tab= tab; > > /* if group or order on first table, sort first */ > if ((group_list && simple_group) || > @@ -4089,7 +4085,10 @@ JOIN::create_postjoin_aggr_table(JOIN_TAB *tab, List<Item> *table_fields, > order= NULL; > } > } > - > + if (!(tab->aggr= new (thd->mem_root) AGGR_OP(tab))) > + goto err; > + tab->table= table; > + table->reginfo.join_tab= tab; > DBUG_RETURN(false); I see a potential issue in the patch: The moved code does this assignment: > tab->table= table; Let's look inside this if-branch: >> if ((group_list && simple_group) || >> (implicit_grouping && select_lex->have_window_funcs())) The code inside if() { ... } part seems to be fine: it does not touch the contents of 'tab', in particular tab->table. Let's look at the coed inside the else {... } part: It has this condition: >> if (!group_list && !table->distinct && order && simple_order && >> tab == join_tab + const_tables) >> { Note that tab==join_tab+const_tables. Then, a few lines below: >> add_sorting_to_table(join_tab + const_tables, order)) This is essentially add_sorting_to_table(tab, order). Inside JOIN::add_sorting_to_table one can find this: TABLE *table= tab->table; if ((tab == join_tab + const_tables) && table->pos_in_table_list && table->pos_in_table_list->is_sjm_scan_table()) that is, we access tab->table and tab->table->pos_in_table_list. Which are not assigned yet (because your patch moves the assignment to happen later). I've tried to construct an example to demonstrate this issue, but was not successful. I've made this change: diff --git a/sql/sql_select.cc b/sql/sql_select.cc index c044656e916..5a9a7d7e73f 100644 --- a/sql/sql_select.cc +++ b/sql/sql_select.cc @@ -4080,6 +4080,8 @@ JOIN::create_postjoin_aggr_table(JOIN_TAB *tab, List<Item> *table_fields, tab == join_tab + const_tables) { DBUG_PRINT("info",("Sorting for order")); + fprintf(stderr,"AAABBBCCC:%s\n", thd->query()); + DBUG_ASSERT(0); THD_STAGE_INFO(thd, stage_sorting_for_order); if (ordered_index_usage != ordered_index_order_by && and ran the entire testsuite with it. There were no failures. I'm not sure if this is deadcode or we just don't have enough test coverage. I suggest to assume the latter and change the patch to avoid this issue. BR Sergei -- Sergei Petrunia, Software Developer MariaDB Corporation | Skype: sergefp | Blog: http://petrunia.net

1 0

Re: [Maria-developers] 3be1e491fb5: MDEV-27743 Remove Lex::charset
by Sergei Golubchik 22 Mar '22

22 Mar '22

Hi, Alexander, All comments from the previous review were addressed and there were no other changes. So, ok to push. On Mar 22, Alexander Barkov wrote: > revision-id: 3be1e491fb5 (mariadb-10.6.1-383-g3be1e491fb5) > parent(s): c69defd3b61 > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2022-03-21 21:33:30 +0400 > message: > > MDEV-27743 Remove Lex::charset > > This patch also fixes: > > MDEV-27690 Crash on `CHARACTER SET csname COLLATE DEFAULT` in column definition > MDEV-27853 Wrong data type on column `COLLATE DEFAULT` and table `COLLATE some_non_default_collation` > MDEV-28067 Multiple conflicting column COLLATE clauses are not rejected > MDEV-28118 Wrong collation of `CAST(.. AS CHAR COLLATE DEFAULT)` > MDEV-28119 Wrong column collation on MODIFY + CONVERT > Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

1 0

Re: [Maria-developers] 026e52bad84: MDEV-27743 Remove Lex::charset
by Sergei Golubchik 21 Mar '22

21 Mar '22

Hi, Alexander, Please, see below On Mar 18, Alexander Barkov wrote: > revision-id: 026e52bad84 (mariadb-10.6.1-369-g026e52bad84) > parent(s): 369498d6b8c > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2022-03-18 12:26:03 +0400 > message: > > MDEV-27743 Remove Lex::charset > > This patch also fixes: > > MDEV-27690 Crash on `CHARACTER SET csname COLLATE DEFAULT` in column definition > MDEV-27853 Wrong data type on column `COLLATE DEFAULT` and table `COLLATE some_non_default_collation` > MDEV-28067 Multiple conflicting column COLLATE clauses are not rejected > MDEV-28118 Wrong collation of `CAST(.. AS CHAR COLLATE DEFAULT)` > MDEV-28119 Wrong column collation on MODIFY + CONVERT > > diff --git a/mysql-test/main/ctype_collate_column.test b/mysql-test/main/ctype_collate_column.test > new file mode 100644 > index 00000000000..6a7cf2cc84c > --- /dev/null > +++ b/mysql-test/main/ctype_collate_column.test > @@ -0,0 +1,637 @@ ... > + > +DELIMITER $$; > +CREATE PROCEDURE p1(dt TEXT, cs TEXT, cl0 TEXT, cl1 TEXT, cl2 TEXT, tcs TEXT) > +BEGIN > + DECLARE errstate TEXT DEFAULT NULL; > + DECLARE errno INT DEFAULT NULL; > + DECLARE errmsg TEXT DEFAULT NULL; > + DECLARE query TEXT; > + DECLARE result TEXT; > + DECLARE CONTINUE HANDLER FOR SQLEXCEPTION > + BEGIN > + GET DIAGNOSTICS CONDITION 1 errstate = RETURNED_SQLSTATE, > + errno = MYSQL_ERRNO, errmsg = MESSAGE_TEXT; > + END; > + SET query= CONCAT('CREATE TABLE t1 (a ', dt, ' ', cs, ' ', cl0, > + ' NOT NULL ',cl1, > + ' DEFAULT '''' ', cl2, > + ') ', tcs, ' ENGINE=Memory'); > + EXECUTE IMMEDIATE query; > + IF errmsg IS NOT NULL > + THEN > + SET result=CONCAT('ERROR: ', > + COALESCE(errstate,'<NULL>'), ' ', > + COALESCE(errno,'<NULL>'), ' ', > + COALESCE(errmsg,'<NULL>')); > + INSERT INTO results (dt,cs,cl0,cl1,cl2,tcs,query,result) > + VALUES (dt,cs,cl0,cl1,cl2,tcs,query,result); > + ELSE > + FOR row IN (SELECT CONCAT(COLUMN_TYPE, > + ' CHARACTER SET ', CHARACTER_SET_NAME, > + ' COLLATE ', COLLATION_NAME) AS result > + FROM INFORMATION_SCHEMA.COLUMNS > + WHERE TABLE_SCHEMA='test' AND TABLE_NAME='t1') > + DO > + INSERT INTO results (dt,cs,cl0,cl1,cl2,tcs,query,result) > + VALUES (dt,cs,cl0,cl1,cl2,tcs,query,row.result); > + END FOR; > + DROP TABLE t1; > + END IF; > +END; > +$$ > +DELIMITER ;$$ > + > + > +DELIMITER $$; > +CREATE PROCEDURE p3(dt TEXT) > +BEGIN > + FOR row IN ( > + SELECT cs, cl0, cl1.cl1 AS cl1, cl2.cl1 AS cl2, tcs > + FROM cs, cl0, cl1, cl1 AS cl2, tcs > + ORDER BY cs, cl0, cl1, cl2, tcs > + ) > + DO > + CALL p1('char(10)', row.cs, row.cl0, row.cl1, row.cl2, row.tcs); minor detail: here ^^^ you want to use dt, not 'char(10)' > + END FOR; > +END; > +$$ > +DELIMITER ;$$ > + > + > +--disable_column_names > +CALL p3('char(10)'); > +--enable_column_names > + > +SELECT * FROM results_statistics; isn't it redundant? you select all rows from result anyway. If something will change, the result of that select will already cause the the to fail. Why do you need second mismatch in the result/reject. > + > +--vertical_results > +SELECT query, result, '' AS `` FROM results > +ORDER BY dt, cs, cl0, cl1, cl2, tcs; > +--horizontal_results > + > +DROP PROCEDURE p1; > +DROP PROCEDURE p3; > + > +DROP TABLE cs, cl0, cl1, tcs; ... > +# CREATE TABLE t1 > +# ( > +# a CHAR(10) BINARY NOT NULL DEFAULT '' > +# ) CHARACTER SET cs COLLATE cs_bin; > + > +DELETE FROM results > +WHERE cs='' > + AND cl0='BINARY' > + AND cl1='' > + AND cl2='' > + AND tcs LIKE 'CHARACTER SET%' > + AND result NOT LIKE 'ERROR%' > + AND result RLIKE CONCAT('COLLATE ', tcs_character_set_name, '_bin'); > +SELECT ROW_COUNT(); ... > +DROP FUNCTION is_collate_clause_with_explicit_default_collation; > +DROP FUNCTION is_collate_clause_with_explicit_collation; > +DROP FUNCTION is_conflicting_charset_explicit_collate_explicit; > +DROP FUNCTION is_conflicting_collate_explicit2; > +DROP FUNCTION is_conflicting_collate_default_collate_explicit; > +DROP FUNCTION collate_cs_default_collation; I think the main result is `select * from results` (with order by, etc). It's great, that you added that. Everything else is redundant, I suggest to remove other selects and deletes. > + > + > +--echo # > +--echo # End of 10.9 tests > +--echo # > diff --git a/sql/lex_charset.h b/sql/lex_charset.h > new file mode 100644 > index 00000000000..d95f0bc1920 > --- /dev/null > +++ b/sql/lex_charset.h > @@ -0,0 +1,202 @@ ... > +#define LEX_CHARSET_COLLATION_TYPE_BITS 2 > + static_assert(((1<<LEX_CHARSET_COLLATION_TYPE_BITS)-1) >= > + TYPE_COLLATE_CONTEXTUALLY_TYPED, > + "Lex_charset_collation_st::Type bits check"); > + > +protected: > + CHARSET_INFO *m_ci; > + Type m_type; > +public: > + static CHARSET_INFO *find_bin_collation(CHARSET_INFO *cs); > + static CHARSET_INFO *find_default_collation(CHARSET_INFO *cs); > +public: ... > + void set_contextually_typed_binary_style() > + { > + m_ci= &my_collation_contextually_typed_binary; > + m_type= TYPE_COLLATE_CONTEXTUALLY_TYPED; > + } > + bool is_contextually_typed_collate_default() const > + { > + return m_ci == &my_collation_contextually_typed_default && > + m_type == TYPE_COLLATE_CONTEXTUALLY_TYPED; what does it mean when only one of these two expressions is true? Say, the first true, the second false? Or the first false, the second true? > + } > + bool is_contextually_typed_binary_style() const > + { > + return m_ci == &my_collation_contextually_typed_binary && > + m_type == TYPE_COLLATE_CONTEXTUALLY_TYPED; > + } ... > diff --git a/strings/ctype-bin.c b/strings/ctype-bin.c > index bb746ad90b0..a5c9129ebea 100644 > --- a/strings/ctype-bin.c > +++ b/strings/ctype-bin.c > @@ -630,3 +630,69 @@ struct charset_info_st my_charset_bin = > &my_charset_handler, > &my_collation_binary_handler > }; > + > + > +struct charset_info_st my_collation_contextually_typed_binary= > +{ > + 0,0,0, /* number */ I thought you will never need to resove the &my_collation_contextually_typed_binary pointer, that is, I thought the values in the my_collation_contextually_typed_binary struct should not matter as they'll never be used. Was that wrong? Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 2

Re: [Maria-developers] d8e915a88fc: MDEV-26009 Server crash when calling twice procedure using FOR-loop
by Sergei Golubchik 20 Mar '22

20 Mar '22

Hi, Oleksandr, ok to push also, see below On Mar 20, Oleksandr Byelkin wrote: > revision-id: d8e915a88fc (mariadb-10.3.33-32-gd8e915a88fc) > parent(s): 6a2d88c1322 > author: Oleksandr Byelkin > committer: Oleksandr Byelkin > timestamp: 2022-03-18 11:13:09 +0100 > message: > > MDEV-26009 Server crash when calling twice procedure using FOR-loop > > The problem was that instructions sp_instr_cursor_copy_struct and > sp_instr_copen uses the same lex, adding and removing "tail" of > prelocked tables and forgetting that tail of all tables is kept in > LEX::query_tables_last. If the LEX used only by one instruction > or the query do not have prelocked tables it is not important. > But to work correctly in all cases LEX::query_tables_last should > be reset to make new tables added in the correct list (after last > table in the LEX instead after last table of the prelocking "tail" > which was cut). > > diff --git a/sql/sp_head.cc b/sql/sp_head.cc > index 57ab31d9edf..96dc78dcf49 100644 > --- a/sql/sp_head.cc > +++ b/sql/sp_head.cc > @@ -3486,6 +3486,11 @@ sp_lex_keeper::reset_lex_and_exec_core(THD *thd, uint *nextp, > lex_query_tables_own_last= m_lex->query_tables_own_last; > prelocking_tables= *lex_query_tables_own_last; > *lex_query_tables_own_last= NULL; > + /* > + Here we cut "tail" of prelocked tables, so have to return last table > + to the original position (wuthout the tail > + */ > + m_lex->query_tables_last= m_lex->query_tables_own_last; 1. the comment looks truncated 2. I don't even think this line needs a comment, this code block removes the list of prelocked tables from the query_tables list. Looking at how query_tables_last and query_tables_own_last are documented: /* Pointer to next_global member of last element in the previous list. */ TABLE_LIST **query_tables_last; /* If non-0 then indicates that query requires prelocking and points to next_global member of last own element in query table list (i.e. last table which was not added to it as part of preparation to prelocking). 0 - indicates that this query does not need prelocking. */ TABLE_LIST **query_tables_own_last; it is pretty clear that after the "tail" is removed, one has to set query_tables_last to point to the end of the list. The fact that it wasn't done was clearly an omission. That is, existing code and existing comments document it pretty well, your comment looks redundant and even confusing to me. So, please, remove the comment and ok to push > m_lex->mark_as_requiring_prelocking(NULL); > } > thd->rollback_item_tree_changes(); Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

1 0

Re: [Maria-developers] Google Summer of Code 2022
by Vicențiu Ciorbaru 20 Mar '22

20 Mar '22

Hi Sergey! I'm happy to see that you have experience in MariaDB code base already. The MDEV-18827 is definitely up for grabs for GSoC 2022. The project is mostly a part-time project (requiring a commitment of around 175 hours), however depending on your availability we can increase the scope of the project and bump it up to a full time project. My suggestion for this task is to start looking at code parsing FRM files. Search for the function TABLE_SHARE::init_from_binary_frm_image to get an idea how an .frm is structured. After that, think how you would like to implement the feature described in MDEV-18827. Feel free to submit a draft proposal and we can iterate on it. Not 100% sure how the GSoC interface works this year, but last time you were able to share with potential mentors a Google Doc with the possibility to comment. Have a look at the proposal guidelines https://mariadb.org/gsoc2022-instructions/ and start writing it. When the program application opens you can submit it (4th of April). In the meantime, feel free to ask any questions you might have here. I encourage you to keep all discussions on the mailing list as much as possible so others can chime in. Vicențiu On Sat, 19 Mar 2022 at 09:54, Сергей Ваниславский <serzhmn(a)gmail.com> wrote: > Hello Vicentiu! > My name is Vanislavsky Sergey. Your colleague Nikita Malyavin > recommended contacting you about GSoC.I would like to do your task > MDEV-18827. I am a 3rd year student at FEFU. Here is my last pull request: > https://github.com/MariaDB/server/pull/2016 > Thank you for your consideration. >

1 0

[Maria-developers] Relevance of MDEV-19556
by Vladisav Kakurin 17 Mar '22

17 Mar '22

Hi, Vicențiu! With my mentor Nikita Malyavin we chose the topic of my bachelor's graduation work, and I was really interested in the MDEV-19556 <https://jira.mariadb.org/browse/MDEV-19556> ticket. Could you say some words about this, maybe the ticket ain’t actual, or something about work that was done by one of the developers to this direction?

1 0

Re: [Maria-developers] 2467eb2d314: MDEV-27743 Remove Lex::charset
by Sergei Golubchik 16 Mar '22

16 Mar '22

Hi, Alexander, I had only a couple of questions/comments, see below: On Mar 11, Alexander Barkov wrote: > revision-id: 2467eb2d314 (mariadb-10.6.1-330-g2467eb2d314) > parent(s): 61b72ed3dde > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2022-02-10 10:27:55 +0400 > message: > > MDEV-27743 Remove Lex::charset > > This patch also fixes: > MDEV-27690 Crash on `CHARACTER SET csname COLLATE DEFAULT` in column definition how do you plan to fix it in 10.2-10.8 ? > > diff --git a/mysql-test/main/ctype_collate_column.result b/mysql-test/main/ctype_collate_column.result > new file mode 100644 > index 00000000000..b20be5d3a48 > --- /dev/null > +++ b/mysql-test/main/ctype_collate_column.result > @@ -0,0 +1,450 @@ > +# > +# Start of 10.9 tests > +# > +# > +# MDEV-27743 Remove Lex::charset 1. Jira doesn't explain what user-visible changes one can expect. If there are none, it's not clear why you needed this big test case 2. the test is totally unreadable. Could you do a normal test instead? Like, put `select` instead of `execute immediate` and convert the output to a .test file? > +# > +CREATE TABLE cs (cs VARCHAR(64) NOT NULL); > +INSERT INTO cs (cs) VALUES > +(''), > +('CHARACTER SET latin1'), > +('CHARACTER SET utf8mb4'); > +CREATE TABLE cl0 (cl0 VARCHAR(64) NOT NULL); > +INSERT INTO cl0 (cl0) VALUES > +(''), > +('BINARY'), > +('COLLATE DEFAULT'), > +('COLLATE latin1_bin'), > +('COLLATE latin1_swedish_ci'), > +('COLLATE utf8mb4_bin'), > +('COLLATE utf8mb4_general_ci'); > +CREATE TABLE cl1 (cl1 VARCHAR(64) NOT NULL); > +INSERT INTO cl1 (cl1) VALUES > +(''), > +('COLLATE DEFAULT'), > +('COLLATE latin1_bin'), > +('COLLATE latin1_swedish_ci'), > +('COLLATE utf8mb4_bin'), > +('COLLATE utf8mb4_general_ci'); ... > diff --git a/mysql-test/main/ctype_latin1.result b/mysql-test/main/ctype_latin1.result > index 0b16d1cf6f2..70577f2c22c 100644 > --- a/mysql-test/main/ctype_latin1.result > +++ b/mysql-test/main/ctype_latin1.result > @@ -8889,6 +8889,38 @@ a b > 111 111 > DROP TABLE t1; > # > +# MDEV-27690 Crash on `CHARACTER SET csname COLLATE DEFAULT` in column definition > +# > +CREATE TABLE t1 (a CHAR(10) CHARACTER SET latin1 COLLATE DEFAULT); > +SHOW CREATE TABLE t1; > +Table Create Table > +t1 CREATE TABLE `t1` ( > + `a` char(10) DEFAULT NULL > +) ENGINE=MyISAM DEFAULT CHARSET=latin1 > +DROP TABLE t1; > +SELECT CAST('a' AS CHAR(10) CHARACTER SET latin1 COLLATE DEFAULT); > +CAST('a' AS CHAR(10) CHARACTER SET latin1 COLLATE DEFAULT) > +a > +CREATE TABLE t1 AS > +SELECT CAST('a' AS CHAR(10) CHARACTER SET latin1 COLLATE DEFAULT) AS c1; > +SHOW CREATE TABLE t1; > +Table Create Table > +t1 CREATE TABLE `t1` ( > + `c1` varchar(10) DEFAULT NULL > +) ENGINE=MyISAM DEFAULT CHARSET=latin1 > +DROP TABLE t1; > +SELECT COLUMN_GET(COLUMN_CREATE(0, 'string'),0 AS CHAR CHARACTER SET latin1 COLLATE DEFAULT) AS c1; > +c1 > +string > +CREATE TABLE t1 AS > +SELECT COLUMN_GET(COLUMN_CREATE(0, 'string'),0 AS CHAR CHARACTER SET latin1 COLLATE DEFAULT) AS c1; > +SHOW CREATE TABLE t1; > +Table Create Table > +t1 CREATE TABLE `t1` ( > + `c1` longtext DEFAULT NULL > +) ENGINE=MyISAM DEFAULT CHARSET=latin1 > +DROP TABLE t1; 1. why is it in "10.2 tests" ? 2. shouldn't here be at least one test that COLLATE DEFAULT is not a no-op? Like CREATE TABLE (... COLLATE DEFAULT ...) COLLATE _non_default ? Or is it in your generated ctype_collate_column.test? > +# > # End of 10.2 tests > # > # > diff --git a/mysql-test/suite/roles/create_and_drop_role_invalid_user_table.test b/mysql-test/suite/roles/create_and_drop_role_invalid_user_table.test > index 9cf0d7b4aff..dac6eab21e9 100644 > --- a/mysql-test/suite/roles/create_and_drop_role_invalid_user_table.test > +++ b/mysql-test/suite/roles/create_and_drop_role_invalid_user_table.test > @@ -36,7 +36,7 @@ set default role test_role for root@localhost; > drop role test_role; > drop user test_user@localhost; > > -alter table user add column default_role char(80) binary default '' not null > +alter table user add column default_role char(80) default '' not null > COLLATE utf8_general_ci Heh, interesting that it worked before good catch > after is_role; > alter table user add max_statement_time decimal(12,6) default 0 not null > diff --git a/sql/structs.h b/sql/structs.h > index 9a5006e8b47..db065c49931 100644 > --- a/sql/structs.h > +++ b/sql/structs.h > @@ -599,11 +599,140 @@ class DDL_options: public DDL_options_st ... > +class Lex_collation: public Lex_collation_st > +{ > +public: > + Lex_collation(CHARSET_INFO *collation, Type type) > + { > + m_collation= collation; > + m_type= type; > + } > + static Lex_collation national(bool bin_mod) > + { > + if (bin_mod) > + return Lex_collation(&my_charset_utf8mb3_bin, TYPE_EXPLICIT); > + return Lex_collation(&my_charset_utf8mb3_general_ci, TYPE_IMPLICIT); utf8mb3? not utf8? > + } > +}; > + > + > struct Lex_length_and_dec_st > { > -private: > +protected: > uint32 m_length; > uint8 m_dec; > + uint8 m_collation_type:2; > bool m_has_explicit_length:1; > bool m_has_explicit_dec:1; > public: > diff --git a/sql/sql_lex.cc b/sql/sql_lex.cc > index ae6ae9a0cc0..6ca10267187 100644 > --- a/sql/sql_lex.cc > +++ b/sql/sql_lex.cc > @@ -11816,3 +11784,257 @@ bool SELECT_LEX_UNIT::explainable() const ... > +CHARSET_INFO * > +Lex_collation_st::resolved_to_character_set(CHARSET_INFO *def) const > +{ > + DBUG_ASSERT(def); > + if (m_type != TYPE_CONTEXTUALLY_TYPED) > + { > + if (!m_collation) > + return def; // Empty - not typed at all > + return m_collation; // Explicitly typed > + } > + > + // Contextually typed > + DBUG_ASSERT(m_collation); > + > + if (m_collation == &my_charset_bin) // CHAR(10) BINARY > + return find_bin_collation(def); > + > + if (m_collation == &my_charset_latin1) // CHAR(10) COLLATE DEFAULT > + return find_default_collation(def); Uhm. Could you create two dummy CHARSET_INFO variables for that instead? Like CHARSET_INFO binary_collaton, default_collation; no need to initialize them, but to have distinct values for binary/default and not reuse my_charset_bin and my_charset_latin1. > + > + /* > + Non-binary and non-default contextually typed collation. > + We don't have such yet - the parser cannot produce this. > + But will have soon, e.g. "uca1400_as_ci". > + */ > + DBUG_ASSERT(0); > + return NULL; > +} > + > + ... > diff --git a/sql/field.h b/sql/field.h > index 2ed02b37cfd..fe5ba435202 100644 > --- a/sql/field.h > +++ b/sql/field.h > @@ -5493,6 +5492,22 @@ class Column_definition: public Sql_alloc, > { return compression_method_ptr; } > > bool check_vcol_for_key(THD *thd) const; > + > + void set_lex_collation(const Lex_collation_st &lc) > + { > + charset= lc.collation(); > + if (lc.is_contextually_typed_collation()) > + flags|= BINCMP_FLAG; > + else > + flags&= ~BINCMP_FLAG; Isn't COLLATE DEFAULT also contextually typed? > + } > + Lex_collation lex_collation() const > + { > + return Lex_collation(charset, > + flags & BINCMP_FLAG ? > + Lex_collation_st::TYPE_CONTEXTUALLY_TYPED : > + Lex_collation_st::TYPE_IMPLICIT); > + } > }; > > Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

2 2

Re: [Maria-developers] 75aa95ada6b: Tricky static asserts did not compile on some platforms. Commenting out.
by Sergei Golubchik 16 Mar '22

16 Mar '22

Hi, Alexander, On Mar 16, Alexander Barkov wrote: > revision-id: 75aa95ada6b (mariadb-10.6.1-338-g75aa95ada6b) > parent(s): e90062c20f1 > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2022-03-09 13:07:35 +0400 > message: > > Tricky static asserts did not compile on some platforms. Commenting out. Details? What doesn't compile and where? I'd suggest to try to keep the assert in some form > diff --git a/sql/structs.h b/sql/structs.h > index abfafb635b1..23958490c7b 100644 > --- a/sql/structs.h > +++ b/sql/structs.h > @@ -602,13 +602,16 @@ class DDL_options: public DDL_options_st > > struct Lex_length_and_dec_st > { > + /* > + Does not compile on some platforms: > static constexpr uint collation_type_bits= 2; > static_assert(((1<<collation_type_bits)-1) >= > Lex_charset_collation_st::TYPE_LAST); > + */ > protected: > uint32 m_length; > uint8 m_dec; > - uint8 m_collation_type:collation_type_bits; > + uint8 m_collation_type:2; /*collation_type_bits;*/ > bool m_has_explicit_length:1; > bool m_has_explicit_dec:1; > public: Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

1 0

Re: [Maria-developers] e90062c20f1: MDEV-27266 Improve UCA collation performance for utf8mb3 and utf8mb4
by Sergei Golubchik 16 Mar '22

16 Mar '22

Hi, Alexander, also ok On Mar 14, Alexander Barkov wrote: > revision-id: e90062c20f1 (mariadb-10.6.1-337-ge90062c20f1) > parent(s): 6a3201d3769 > author: Alexander Barkov > committer: Alexander Barkov > timestamp: 2022-02-28 14:43:43 +0400 > message: > > MDEV-27266 Improve UCA collation performance for utf8mb3 and utf8mb4 > > Adding two levels of optimization: > > include/m_ctype.h | 53 ++++ > strings/ctype-uca-scanner_next.inl | 39 +++ > strings/ctype-uca.c | 548 ++++++++++++++++++++++++++++++++++++- > strings/ctype-uca.ic | 30 ++ I thought all *.ic files were renamed to *.inl ? > unittest/strings/strings-t.c | 2 +- > 5 files changed, 660 insertions(+), 12 deletions(-) Regards, Sergei VP of MariaDB Server Engineering and security(a)mariadb.org

1 0