Current Path : /home/usr.opt/mysql57/mysql-test/include/ |
FreeBSD hs32.drive.ne.jp 9.1-RELEASE FreeBSD 9.1-RELEASE #1: Wed Jan 14 12:18:08 JST 2015 root@hs32.drive.ne.jp:/sys/amd64/compile/hs32 amd64 |
Current File : /home/usr.opt/mysql57/mysql-test/include/group_by_fd.inc |
--source include/have_optimizer_trace.inc let $DEFAULT_TRACE_MEM_SIZE=1048576; # 1MB eval set optimizer_trace_max_mem_size=$DEFAULT_TRACE_MEM_SIZE; set end_markers_in_json=on; set optimizer_trace="enabled=on"; SET @old_sql_mode = @@sql_mode; SET @@sql_mode='ONLY_FULL_GROUP_BY'; --echo # --echo # Bug#16021396 ONLY_FULL_GROUP_BY REJECTS VALID QUERY USING VIEW --echo # create table t1(a int, b int, c int) engine=InnoDB; let $count=2; while($count) { if ($count == 2) { let $viewtype=merge; } if ($count == 1) { let $viewtype=temptable; } dec $count; eval create algorithm=$viewtype view v1 as select t1.a*2 as a, t1.b*2 as b, t1.c*2 as c from t1; show create view v1; select sin(b) as z from t1 group by sin(b); select sin(b) as z from v1 group by sin(b); select sin(b) as z from t1 group by b; select sin(b) as z from v1 group by b; select sin(b) as z from v1 group by z; drop view v1; } drop table t1; --echo # From testcase of Bug#16903135: CREATE TABLE group_by_test2 ( id int unsigned primary key, cat int unsigned not null, name varchar(10), num int unsigned ); INSERT INTO group_by_test2 (id,cat,name,num) VALUES (1,10,'foo',2), (2,11,'foo',1), (3,22,'bar',3), (4,23,'bar',7), (5,34,'test',7); let $query= SELECT cat, name, SUM(num) FROM group_by_test2 GROUP BY cat; --error ER_WRONG_FIELD_WITH_GROUP eval $query; ALTER TABLE group_by_test2 ADD UNIQUE INDEX (cat); eval $query; --error ER_WRONG_FIELD_WITH_GROUP eval $query WITH ROLLUP; --echo # Expressions of group columns are ok: SELECT cat, length(cat), SUM(num) FROM group_by_test2 GROUP BY cat WITH ROLLUP; DROP TABLE group_by_test2; --echo # Test from Bug #18993257 SELECT AGGR + NON-AGGR FROM JOIN WITH VIEW IS NOT REJECTED BY ONLY_FULL_GROUP_BY CREATE TABLE t1 ( col_int_key INT, col_varchar_key VARCHAR(1) ); CREATE TABLE t2 ( pk INTEGER, col_int_key INTEGER ); CREATE VIEW view_b AS SELECT * FROM t2; --error ER_MIX_OF_GROUP_FUNC_AND_FIELDS SELECT MIN( alias2.col_int_key ), alias2.col_int_key FROM t1 AS alias1, t2 AS alias2 WHERE alias1.col_int_key IS NULL; --echo # Same with view: --error ER_MIX_OF_GROUP_FUNC_AND_FIELDS SELECT MIN( alias2.col_int_key ), alias2.col_int_key FROM t1 AS alias1, view_b AS alias2 WHERE alias1.col_int_key IS NULL; DROP TABLE t1, t2; DROP VIEW view_b; --echo # --echo # WL#2489; Recognizing some functional dependencies --echo # --source include/show_json_object.inc let $show_trace= SELECT show_json_object('"functional_dependencies_of_GROUP_columns": {', TRACE) FROM INFORMATION_SCHEMA.OPTIMIZER_TRACE; create table t1( a int, b int not null, c int not null, d int, unique key(b,c), unique key(b,d) ); --error ER_WRONG_FIELD_WITH_GROUP select sin(a) as z from t1 group by b; --error ER_WRONG_FIELD_WITH_GROUP select sin(a) as z from t1 group by d,b; --echo # {b,c} UNIQUE NOT NULL => {c,b}->a select sin(a) as z from t1 group by c,b; select sin(a+b*c) as z from t1 group by c,b; --echo # In PS mode, we see nothing below, because only_full_group_by --echo # checks are done at PREPARE, whereas trace below is from EXECUTE. eval $show_trace; --echo # With outer references: --error ER_WRONG_FIELD_WITH_GROUP select (select sin(a)) as z from t1 group by d,b; select (select sin(a)) as z from t1 group by c,b; --echo # If key columns are in function, functional dependency disappears --error ER_WRONG_FIELD_WITH_GROUP select sin(a) as z from t1 group by c*2,b*2; --echo # FDs are recognized, like in SQL standard: --echo # b=2 => c->{b,c} => c->a as (b,c) is unique not null. select sin(a) as z from t1 where b=2 group by c; eval $show_trace; --echo # t2.a=t1.a => {t1.b,t1.c}->t2.a select sin(t2.a) as z from t1, t1 as t2 where t2.a=t1.a group by t1.b,t1.c; eval $show_trace; --echo # t2.b=t1.b and t2.c=t1.c => {t1.b,t1.c}->{all cols of t2} select sin(t2.a) as z from t1, t1 as t2 where t2.b=t1.b and t2.c=t1.c group by t1.b,t1.c; eval $show_trace; --echo # t2.b=t1.b and t2.c=t1.c => {t1.b,t1.c}->{all cols of t2} --echo # Moreover, {t1.b,t1.c}->{t1.d}. --echo # So t3.b=t2.b and t3.c=t1.d => {t1.b,t1.c}->{all cols of t3}. select t3.a from t1, t1 as t2, t1 as t3 where t3.b=t2.b and t3.c=t1.d and t2.b=t1.b and t2.c=t1.c group by t1.b,t1.c; eval $show_trace; --echo # 3 tables: --echo # {t1.b,t1.c}->{t1.*}->{t2.b,t2.c}->{t2.*}->{t3.pk}->{t3.b} create table t3(pk int primary key, b int); select t3.b from t1,t1 as t2,t3 where t3.pk=t2.d and t2.b=t1.b and t2.c=t1.a group by t1.b,t1.c; drop table t3; --echo # With subq select (select t1.b from t1 where t2.b=t1.b group by t1.a) from t1 as t2; --echo # Outer join. create table t2 like t1; delete from t1; insert into t1 (a,b) values(1,10),(2,20); insert into t2 (a,b) values(1,-10); --echo # In result, t2.a is NULL for both rows, values of t1.a are 1 and 2 select t1.a,t2.a from t1 left join t2 on t2.a=t1.a and t2.b=t1.b; --echo # So this query would choose one arbitrary value of t1.a - wrong: --error ER_WRONG_FIELD_WITH_GROUP select t1.a from t1 left join t2 on t2.a=t1.a and t2.b=t1.b group by t2.a; --echo # Also problem for FD with constant: select t1.a,t2.a from t1 left join t2 on 42=t1.a and t2.b=t1.b; --error ER_WRONG_FIELD_WITH_GROUP select t1.a from t1 left join t2 on 42=t1.a and t2.b=t1.b group by t2.a; --error ER_WRONG_FIELD_WITH_GROUP select t1.a from t1 left join t2 on t2.b=t1.a group by t2.b; --error ER_WRONG_FIELD_WITH_GROUP select t1.a from t1 left join t2 on 42=t1.a group by t2.a; --error ER_WRONG_FIELD_WITH_GROUP select t1.c from t1 left join t2 on t1.a=t1.c group by t1.a; --error ER_WRONG_FIELD_WITH_GROUP select t1.b from t1 left join t2 on t1.c=t1.b and t1.a=t1.c group by t1.a; select t1.a,t2.c from t1 left join t2 on t1.a=t2.c and cos(t2.c+t2.b)>0.5 and sin(t1.a+t2.d)<0.9 group by t1.a; --echo # with keys: select t1.a,t2.d from t1 left join t2 on t1.a=t2.c and t1.d=t2.b and cos(t2.c+t2.b)>0.5 and sin(t1.a+t2.d)<0.9 group by t1.a,t1.d; --echo # with non-determinism: --error ER_WRONG_FIELD_WITH_GROUP select t1.a,t2.c from t1 left join t2 on t1.a=t2.c and cos(t2.c+rand())>0.5 group by t1.a; select t1.a, ANY_VALUE(t2.c) from t1 left join t2 on t1.a=t2.c and cos(t2.c+rand())>0.5 group by t1.a; --echo # with parameter: prepare s from 'select t1.a,t2.c from t1 left join t2 on t1.a=t2.c and cos(t2.c+ ? )>0.5 group by t1.a'; execute s using @dummy; --echo # No OR --error ER_WRONG_FIELD_WITH_GROUP select t1.a,t2.c from t1 left join t2 on t1.a=t2.c and cos(t2.c+t2.b)>0.5 OR sin(t2.d)<0.9 group by t1.a; --echo # With subq --error ER_WRONG_FIELD_WITH_GROUP select t2.b from t1 left join t1 as t2 on t1.a=t2.b and t1.b group by t1.a; --error ER_WRONG_FIELD_WITH_GROUP select t2.b from t1 left join t1 as t2 on t1.a=t2.b and (select t1.b) group by t1.a; --echo # Test ANY_VALUE: select ANY_VALUE(t1.b) from t1 left join t2 on t1.c=t1.b and t1.a=t1.c group by t1.a; --error ER_WRONG_FIELD_WITH_GROUP select 3+(5*t1.b) from t1 left join t2 on t1.c=t1.b and t1.a=t1.c group by t1.a; select 3+(5*ANY_VALUE(t1.b)) from t1 left join t2 on t1.c=t1.b and t1.a=t1.c group by t1.a; delete from t1; insert into t1 (a,b) values(1,10),(1,20),(2,30),(2,40); --error ER_MIX_OF_GROUP_FUNC_AND_FIELDS select a, sum(b) from t1; select any_value(a), sum(b) from t1; --echo # different order of input rows, different "any_value": select any_value(a), sum(b) from (select * from t1 order by a desc) as d; --error ER_WRONG_FIELD_WITH_GROUP select a,b,sum(c) from t1 group by a; select a,any_value(b),sum(c) from t1 group by a; select a,any_value(b),sum(c) from (select * from t1 order by a desc, b desc) as d group by a; --echo # With view. let $count=3; while($count) { set @optimizer_switch_saved=@@optimizer_switch; let $source_query=select t1.a*2 as a, t1.b as b, t1.c as c, t1.d as d, t1.a as e from t1; if ($count == 3) { --echo # Merged view eval create algorithm=merge view v1 as $source_query; show create view v1; let $source=v1; let $source_no_alias=v1; } if ($count == 2) { --echo # Materialized view eval create algorithm=temptable view v1 as $source_query; show create view v1; let $source=v1; let $source_no_alias=v1; } if ($count == 1) { --echo # Materialized derived table set optimizer_switch='derived_merge=off'; let $source=($source_query) as v1; let $source_no_alias=($source_query); } --error ER_WRONG_FIELD_WITH_GROUP eval select sin(a) as z from $source group by b; --error ER_WRONG_FIELD_WITH_GROUP eval select sin(a) as z from $source group by d,b; eval select sin(a) as z from $source group by c,b; --error ER_WRONG_FIELD_WITH_GROUP eval select (select sin(a)) as z from $source group by d,b; eval select (select sin(a)) as z from $source group by c,b; --error ER_WRONG_FIELD_WITH_GROUP eval select sin(a) as z from t1 group by c*2,b*2; eval select sin(a+b*c) as z from $source group by c,b; eval $show_trace; eval select sin(a) as z from $source where b=2 group by c; --echo # {v1.b,v1.c}->v1.a->t2.a eval select sin(t2.a) as z from $source, $source_no_alias as t2 where t2.a=v1.a group by v1.b,v1.c; eval $show_trace; eval select sin(t2.a) as z from $source, $source_no_alias as t2 where t2.b=v1.b and t2.c=v1.c group by v1.b,v1.c; --echo # With materialized view we get more dependencies than needed, due to implementation. eval select t3.a from $source, $source_no_alias as t2, $source_no_alias as t3 where t3.b=t2.b and t3.c=v1.d and t2.b=v1.b and t2.c=v1.c group by v1.b,v1.c; eval $show_trace; --echo # If we simply went to real_item(), we would have WHERE 2*a=b, or --echo # GROUP BY 2*t1.a, so we would not find FDs. The original item --echo # (direct_view_ref here) must also be considered! eval select a from $source where a=b group by b; eval $show_trace; eval select b from $source where a=b group by a; eval $show_trace; eval select v1.c from $source where v1.c=v1.a group by v1.a; eval $show_trace; eval select v1.a from $source group by v1.e; eval $show_trace; eval select v1.c from $source where v1.c=v1.a group by v1.e; eval $show_trace; --echo # View appears only in WHERE eval select t2.d from $source, t1 as t2 where v1.a=t2.d and v1.e=t2.a group by t2.a; eval $show_trace; drop view if exists v1; --echo # FD due to view's WHERE: let $source_query=select t1.a*2 as a, t1.b as b from t1; if ($count == 3) { eval create algorithm=merge view v1 as $source_query; } if ($count == 2) { eval create algorithm=temptable view v1 as $source_query; } if ($count == 1) { let $source=($source_query) as v1; let $source_no_alias=($source_query); } --error ER_WRONG_FIELD_WITH_GROUP eval select a from $source group by b; drop view if exists v1; let $source_query=select t1.a*2 as a, t1.b as b from t1 where t1.a=t1.b; if ($count == 3) { --echo # Merged view eval create algorithm=merge view v1 as $source_query; } if ($count == 2) { --echo # Materialized view eval create algorithm=temptable view v1 as $source_query; } if ($count == 1) { --echo # Derived table let $source=($source_query) as v1; let $source_no_alias=($source_query); } eval select a from $source group by b; eval $show_trace; drop view if exists v1; --echo # Aggregates in view let $source_query=select sum(t1.a) as a, sum(t1.b) as b from t1; if ($count == 3) { --echo # Aggregates => skipping Merged view } if ($count == 2) { --echo # Materialized view eval create algorithm=temptable view v1 as $source_query; } if ($count == 1) { --echo # Derived table let $source=($source_query) as v1; let $source_no_alias=($source_query); } if ($count <= 2) { eval select a from $source group by b; eval $show_trace; } drop view if exists v1; --echo # Aggregates + GROUP BY in view --echo # We group by b*5, to show that it works with GROUP expressions, not only fields. let $source_query=select a, b*5 as b, sum(t1.c) as c, sum(t1.d) as d from t1 group by a,b*5; if ($count == 3) { --echo # Aggregates => skipping Merged view } if ($count == 2) { --echo # Materialized view eval create algorithm=temptable view v1 as $source_query; } if ($count == 1) { --echo # Derived table let $source=($source_query) as v1; let $source_no_alias=($source_query); } if ($count <= 2) { --error ER_WRONG_FIELD_WITH_GROUP eval select a from $source group by b; --error ER_WRONG_FIELD_WITH_GROUP eval select c from $source group by b; --error ER_WRONG_FIELD_WITH_GROUP eval select c from $source group by b,d; --error ER_WRONG_FIELD_WITH_GROUP eval select a,c from $source group by a; eval select d,c from $source group by b,a; eval $show_trace; } drop view if exists v1; dec $count; set optimizer_switch=@optimizer_switch_saved; } --echo # Derived table in merged view create algorithm=temptable view v2 as select a as a, 2*a as b from t1; create algorithm=merge view v1 as select v2.a as a, 3*v2.b as b from v2; select v1.b from v1 group by v1.a; eval $show_trace; select t2.a from t1 as t2, v1 where t2.a=v1.b group by v1.a; eval $show_trace; --echo # FDs in a view are those of the underlying query expression. --echo # FDs in a query expression: expressions in the SELECT list must be --echo # deterministic. drop view v1; create algorithm=merge view v1 as select v2.a as a, rand()*v2.b as b from v2; --error ER_WRONG_FIELD_WITH_GROUP select v1.b from v1 group by v1.a; drop view v1; create algorithm=temptable view v1 as select v2.a as a, rand()*v2.b as b from v2; --error ER_WRONG_FIELD_WITH_GROUP select v1.b from v1 group by v1.a; drop view v1,v2; --echo # Item_direct_view_ref pointing to Item_direct_view_ref pointing to --echo # Item_field (a rare case!) create algorithm=merge view v2 as select 2*a as a, 2*b as b from t1; create algorithm=merge view v1 as select a, 3*b as b from v2 where a=b; select 1 from (select a,b+0 from v1 group by a) as d; drop view v1,v2; --echo # Some fun cases with aggregates in derived table. --echo # Inner query is valid: t1.a -> t1.b (equality in WHERE). Outer query: --echo # d.b -> t1.b (underlying column of d.b) -> t1.a (equality) --echo # -> sum(1) (because t1.a is all group columns so determines --echo # sum()) -> d.s (because sum() is underlying of d.s) select d.s from (select b, sum(1) as s from t1 where a=b group by a) as d group by d.b; alter table t1 add column pk int primary key auto_increment; --echo # Inner query: t1.pk -> t1.* (pk). Outer query: --echo # d.b,d.c -> t1.b,t1.c (their underlying columns) -> t1.pk (because --echo # t1.b,t1.c is unique) -> sum(1) (because t1.pk is all group columns so --echo # determines sum()) -> d.s (because sum() is underlying of d.s) select d.s from (select b, c, sum(d) as s from t1 group by pk) as d group by d.b,d.c; --echo # Outer query: --echo # d.c -> t1.b*3 (underlying column of d.c) -> sum(a) (because t1.b*3 --echo # is all group expressions) -> d.s select d.s from (select b*3 as c, sum(a) as s from t1 group by b*3) as d group by d.c; drop table t1,t2; --echo # Testcase from Reviewers create table customer1(pk int primary key, a int); create table customer2(pk int primary key, b int); CREATE algorithm=merge VIEW customer as SELECT pk,a,b FROM customer1 JOIN customer2 USING (pk); let $query=select customer.pk, customer.b from customer group by customer.pk; eval $query; --echo # View is merged. Show FDs. Note that in --ps-protocol, the trace --echo # is that of execution, so contains no group-by checks. eval $show_trace; drop view customer; CREATE algorithm=temptable VIEW customer as SELECT pk,a,b FROM customer1 JOIN customer2 USING (pk); eval $query; eval $show_trace; --echo # Benefit from outer-join-to-inner conversion. insert into customer1 values(0,10),(1,20); insert into customer2 values(0,10),(1,20); --echo # 1) no conversion, no FD from customer2.b to customer1.a. --error ER_WRONG_FIELD_WITH_GROUP explain select customer1.a, count(*) from customer1 left join customer2 on customer1.a=customer2.b where customer1.pk in (7,9) group by customer2.b; --echo # 2) null-complemented row can't pass WHERE => conversion --echo # => FD from customer2.b to customer1.a. explain select customer1.a, count(*) from customer1 left join customer2 on customer1.a=customer2.b where customer2.pk in (7,9) group by customer2.b; drop table customer1,customer2; drop view customer; --echo # FDs of JOIN...USING and NATURAL JOIN create table t1(pk int primary key, a int); create table t2(pk int primary key, b int); select t1.pk, t1.a from t1 join t2 on t1.pk=t2.pk group by t1.pk; select t1.pk, t1.a from t1 join t2 using(pk) group by t1.pk; select t1.pk, t1.a from t1 natural join t2 group by t1.pk; select t1.pk, t1.a from t1 left join t2 using(pk) group by t1.pk; select t1.pk, t1.a from t1 natural left join t2 group by t1.pk; select t1.pk, t2.b from t1 join t2 on t1.pk=t2.pk group by t1.pk; select t1.pk, t2.b from t1 join t2 using(pk) group by t1.pk; select t1.pk, t2.b from t1 natural join t2 group by t1.pk; select t1.pk, t2.b from t1 left join t2 using(pk) group by t1.pk; eval $show_trace; select t1.pk, t2.b from t1 natural left join t2 group by t1.pk; --echo # Equivalent queries, with RIGHT JOIN select t1.pk, t2.b from t2 right join t1 using(pk) group by t1.pk; select t1.pk, t2.b from t2 natural right join t1 group by t1.pk; --error ER_WRONG_FIELD_WITH_GROUP select t1.pk, t2.b from t1 left join t2 on t1.pk>t2.pk group by t1.pk; --echo # Even in --ps-protocol we see the group-by checks in trace because --echo # there has been no execution (due to error). eval $show_trace; --error ER_WRONG_FIELD_WITH_GROUP select t1.pk, t2.b from t2 right join t1 on t1.pk>t2.pk group by t1.pk; drop table t1,t2; --echo # Cases where FDs from weak side do not propagate create table t1(a int, b int); insert into t1 values(null,0),(null,1); let $query= select d.a,d.c from t1 left join (select a, coalesce(a,3) as c from t1) as d on t1.b>0; eval $query; --echo # Now group it by d.a: --error ER_WRONG_FIELD_WITH_GROUP eval $query group by d.a; eval $show_trace; let $query= select d.a,d.c from t1 left join (select a, count(a) as c from t1 group by a) as d on t1.b+d.c>0; eval $query; --echo # Now group it by d.a: --error ER_WRONG_FIELD_WITH_GROUP eval $query group by d.a; eval $show_trace; let $query= select d.m,d.c from t1 left join (select max(a) as m, count(a) as c from t1) as d on t1.b+d.c>0; eval $query; --echo # Now group it by d.m: --error ER_WRONG_FIELD_WITH_GROUP eval $query group by d.m; eval $show_trace; --echo # Now group it by d.c which is non-nullable: eval $query group by d.c; eval $show_trace; drop table t1; create table t1(pk int primary key, a int); --echo # Not valid per the standard, because: --echo # 1) In t3, t3.pk->t3.a holds. --echo # 2) In R1 the result of "(t2 left join t3 on 1)", t3.pk->t3.a --echo # holds, by application of: there is a functional dependency in the --echo # weak side t3, and t3.pk is not nullable in t3. --echo # 3) In R2 the result of "t1 left join (t2 left join t3 on 1) on 1", --echo # t3.pk->t3.a doesn't hold anymore, because: it's a dependency in the --echo # weak side (weak side is R1), and t3.pk is nullable _when --echo # seen as a column of R1_ (in R1 t3.pk can be NULL, if the row of t3 --echo # is actually a null-complemented one). --echo # But for us it is valid, because we have refined the logic: the --echo # pk-based FD satisfies the requirement that a NULL value of t3.pk --echo # implies a NULL value of t3.a (indeed, the NULL value of t3.pk can --echo # only come from null-complementing of the row of t3 in R1, in which --echo # case t3.a is also NULL). select t3.a from t1 left join (t1 as t2 left join t1 as t3 on 1) on 1 group by t3.pk; eval $show_trace; --echo # Outer reference - why we use resolved_used_tables(): select (select t1.a from t1 as t2 limit 1) from t1 group by pk; --echo # We don't build the FD list if not needed select t1.a*3 from t1 group by t1.a; eval $show_trace; drop table t1; --echo # Tricky cases with "ON col=literal" propagating. create table t1(a int, b int); insert into t1 values(); insert into t1 values(10,11); create table t2(c int, d int); insert into t2 values(2,3); let $query= select t4.d from t1 left join (t2 as t3 join t2 as t4 on t4.d=3) on t1.a=10; eval $query; --echo # Equivalent to T1 LJ (T2, T3) ON T4.D=3 AND T1.A=10 --echo # (this is what simplify_joins() does). --echo # For T4.D=3, DJS is {T1.A} which is not group column. --error ER_WRONG_FIELD_WITH_GROUP eval $query group by ""; let $query= select t4.d from t1 left join (t2 as t3 left join t2 as t4 on t4.d=3) on t1.a=10; eval $query; --echo # For T4.D=3, DJS={}, not NULL-friendly, and embedding is on weak side --echo # so FD cannot propagate. --error ER_WRONG_FIELD_WITH_GROUP eval $query group by ""; let $query= select t4.d from t1 join (t2 as t3 left join t2 as t4 on t4.d=3) on t1.a=10; eval $query; --echo # For T4.D=3, DJS={}, not NULL-friendly, but embedding is on weak side --echo # so FD can propagate. eval $query group by ""; --echo # With a view create view v1 as select a as a, 2*a as b, coalesce(a,3) as c from t1; let $query= select v1.b from t1 left join v1 on 1; eval $query; --echo # If v1.a is NULL then v1.b is NULL: a->b is NULL-friendly eval $query group by v1.a; let $query= select v1.c from t1 left join v1 on 1; eval $query; --echo # If v1.a is NULL then v1.c may not be NULL: a->c is not NULL-friendly --error ER_WRONG_FIELD_WITH_GROUP eval $query group by v1.a; drop view v1; --echo # Constant view item create view v1 as select a as a, 2 as b from t1; let $query= select t1.a, v1.b from t1 left join v1 on t1.a is not null; --echo # Because of BUG#17023060, the result is wrong, should be --echo # [10,2 ; 10,2 ; NULL,NULL], which would show that {}->{v1.b} does not --echo # hold in the result, even though v1.b is constant (=2) in v1. eval $query; --echo # We correctly reject this: --error ER_WRONG_FIELD_WITH_GROUP eval $query group by v1.a; drop view v1; drop table t1,t2; create table emp(empno int, ename char(1), deptno int); create table dept(deptno int primary key, dname char(1)); let $view_rest= VIEW empdept AS SELECT emp.empno, emp.ename, dept.deptno, dept.dname FROM emp LEFT OUTER JOIN dept ON (emp.deptno = dept.deptno); let $query= SELECT dname, COUNT(*) FROM empdept GROUP BY deptno; eval CREATE algorithm=merge $view_rest; eval EXPLAIN $query; --echo # There is pk-based FD dept.Depno->dept.dname in dept --echo # and it propagates in the view-which-became-nest because it is --echo # NULL-friendly. eval $query; drop view empdept; eval CREATE algorithm=temptable $view_rest; eval EXPLAIN $query; --echo # There is pk-based FD dept.Depno->dept.dname in dept --echo # and it propagates in the materialized view because it is --echo # NULL-friendly, and then in the top query because the view is not in --echo # the weak side of an outer join. eval $query; --echo # More tests, for code coverage. --echo # UNION in derived table --error ER_WRONG_FIELD_WITH_GROUP select emp.ename from (select 1 as empno union select 2) deriv, emp where emp.empno=deriv.empno group by emp.empno; drop view empdept; --echo # Make the key-searching loop meet view columns which don't wrap a --echo # column (CONCAT). CREATE VIEW empdept AS SELECT emp.empno, emp.ename, dept.deptno, concat("",dept.dname) as dname FROM emp LEFT JOIN dept ON (emp.deptno = dept.deptno); --error ER_WRONG_FIELD_WITH_GROUP SELECT ename, COUNT(*) FROM empdept WHERE empno=dname and empno=deptno GROUP BY empno; drop table emp,dept; drop view empdept; --echo # --echo # Bug#19636980 ASSERT `TABLE->OUTER_JOIN' FAILED IN GROUP_CHECK::FIND_FD_IN_JOINED_TABLE --echo # CREATE TABLE t1 ( c1 INT, c2 INT, c4 DATE, c5 VARCHAR(1) ); CREATE TABLE t2 ( c1 INT, c2 INT, c3 INT, c5 VARCHAR(1) ); --echo # alias1.c5 is not FD, the error is detected at SELECT time CREATE VIEW v1 AS SELECT alias1.c4 AS field1 FROM t1 AS alias1 INNER JOIN t1 AS alias2 ON 1 GROUP BY field1 ORDER BY alias1.c5; --error ER_WRONG_FIELD_WITH_GROUP SELECT * FROM v1; DROP VIEW v1; --echo # alias1.c5 is FD (WHERE contains: alias1.c5='d') CREATE VIEW v1 AS SELECT alias1.c4 AS field1, alias1.c4 AS field2 FROM t1 AS alias1 INNER JOIN t1 AS alias2 ON (alias2.c1 = alias1.c2) WHERE ( NOT EXISTS ( SELECT SQ1_alias1.c5 AS SQ1_field1 FROM t2 AS SQ1_alias1 WHERE SQ1_alias1.c3 < alias1.c1 )) AND (alias1.c5 = alias1.c5 AND alias1.c5 = 'd' ) GROUP BY field1, field2 ORDER BY alias1.c5, field1, field2 ; SELECT * FROM v1; DROP VIEW v1; DROP TABLE t1,t2; --echo # --echo # Bug#19636409 ASSERT `(MAP_OF_NEW_EQ_FDS...` IN GROUP_CHECK::IS_FD_ON_SOURCE ON SELECT --echo # CREATE TABLE t1 ( pk int NOT NULL, c1 datetime, c2 varchar(1), c3 date, c4 date, c5 varchar(1), PRIMARY KEY (pk) ); CREATE VIEW v1 AS SELECT c3 AS subfield11, pk AS subfield12, c2 AS subfield13 FROM t1 GROUP BY subfield11, subfield12 ; --echo # This query is valid. Indeed: field3 i.e. t1.c2 is part of GROUP BY, --echo # so, in a group: --echo # - all rows have the same value of t1.c2, noted val_c2, --echo # - so all rows, which necessarily match WHERE, matched with a row of --echo # v1 having subfield11=val_c2 and subfield12=val_c2 (due to IN --echo # predicate), --echo # - thus both grouping columns of v1's query expression are constant --echo # (to val_c2) --echo # - thus v1.subfield13 is constant too (grouping columns determine the --echo # SELECT list) --echo # - so, due to IN, this determines the value of t1.pk (t1 is the top --echo # query's table). --echo # - so t1.pk is constant, so all columns of t1 are constant. --echo # Which proves that it makes sense to search FDs in semijoin --echo # equalities. --echo # In other words: a semijoin is like a join except that it eliminates --echo # duplicates, but duplicates are irrelevant to the decision whether a --echo # column is functionally dependent on a set of columns. SELECT c5 AS field1, c4 AS field2, c2 AS field3, c1 AS field4 FROM t1 WHERE ( c2, c2, pk ) IN ( SELECT * FROM v1 ) GROUP BY field1, field3, field3; --echo # --echo # Bug#19687724 FUNCTIONAL DEPENDENCIES ARE NOT RECOGNIZED IN EQUALITIES BETWEEN ROWS --echo # --echo # Inspired by query above, but with a join: SELECT c5 AS field1, c4 AS field2, c2 AS field3, c1 AS field4 FROM t1, v1 WHERE ( c2, c2, pk ) = (subfield11, subfield12, subfield13) GROUP BY field1, field3, field3; --echo # With constants: SELECT c5 AS field1, c4 AS field2, c2 AS field3, c1 AS field4 FROM t1 WHERE ( c2, c2, pk ) = (1, 2, 3) GROUP BY field1, field3, field3; DROP TABLE t1; CREATE TABLE t1(a INT, b INT, c INT, d INT); SELECT a,b,c,d FROM t1 WHERE a=c AND b=d GROUP by a,b; SELECT a,b,c,d FROM t1 WHERE (a,b)=(c,d) GROUP BY a,b; --error ER_WRONG_FIELD_WITH_GROUP SELECT a,b,c,d FROM t1 WHERE (a,b)=(c,d+1) GROUP BY a,b; DROP TABLE t1; DROP VIEW v1; --echo # --echo # Bug #20031708 ASSERT ON GROUP_CHECK::IS_FD_ON_SOURCE --echo # CREATE TABLE t1 ( col_varchar_10_utf8 VARCHAR(10) CHARACTER SET utf8, col_int_key INT, pk INT PRIMARY KEY ); CREATE TABLE t2 ( col_varchar_10_utf8 VARCHAR(10) CHARACTER SET utf8 DEFAULT NULL, col_int_key INT DEFAULT NULL, pk INT PRIMARY KEY ); CREATE VIEW v2 AS SELECT * FROM t2; --echo # The reporter's testcase: let $query= SELECT COUNT(*), t1.col_int_key FROM v2 LEFT OUTER JOIN t1 ON v2.col_varchar_10_utf8 = t1.col_varchar_10_utf8 WHERE v2.pk = 4; --error ER_MIX_OF_GROUP_FUNC_AND_FIELDS eval $query; DROP VIEW v2; --echo # A variant: one column is an expression: CREATE VIEW v2 AS SELECT CONCAT(col_varchar_10_utf8,' ') AS col_varchar_10_utf8, col_int_key, pk FROM t2; --error ER_MIX_OF_GROUP_FUNC_AND_FIELDS eval $query; DROP VIEW v2; --echo # Query used in the commit comment: view column involving two tables CREATE VIEW v2 AS SELECT t1.pk, t2.col_int_key+1 as c, t1.pk+t2.col_int_key as p FROM t1, t2; --echo # FDs will be discovered in this order: {}->v2.pk, v2.pk->v2.c, --echo # v2.c->v2.p SELECT COUNT(*), v2.p FROM v2 WHERE v2.c=v2.p and v2.c=v2.pk AND v2.pk = 4; DROP VIEW v2; --echo # If in the query specification defining a view, a base table's pk is --echo # determined, --echo # and the view's column is a function of this base table's columns, --echo # then the view's column is also determined. --echo # So, in this view's result, {v2.pk}->{v2.coa}: CREATE ALGORITHM=MERGE VIEW v2 AS SELECT t2.pk, COALESCE(t2.pk, 3) AS coa FROM t1 LEFT JOIN t2 ON 0; --echo # And thus {pk}->{coa} holds in the result of this query using the view; --echo # if there is NULL-complementing in the LEFT JOIN below then (pk,coa) --echo # will be (NULL,NULL) and if there is not it will be (non-NULL,3): --echo # v2.coa is determined by v2.pk. The key fact is that v2.pk is not --echo # NULLable so this is a NFFD. SELECT v2.pk, v2.coa FROM t1 LEFT JOIN v2 AS v2 ON 0 GROUP BY v2.pk; DROP VIEW v2; DROP TABLE t1,t2; --echo # --echo # Bug#21807579 FUNCTIONAL DEPENDENCIES ARE NOT RECOGNIZED IN GENERATED COLUMNS --echo # CREATE TABLE t ( a INT, c INT GENERATED ALWAYS AS (a+2), d INT GENERATED ALWAYS AS (c+2) ); --echo # {a}->{c}, {c}->{d} and {a}->{d} hold. SELECT c FROM t GROUP BY a; SELECT d FROM t GROUP BY c; SELECT d FROM t GROUP BY a; SELECT 1+c FROM t GROUP BY a; SELECT 1+d FROM t GROUP BY c; SELECT 1+d FROM t GROUP BY a; --echo # {t2.a}->{t2.d}->{t1.c}->{t1.d} SELECT t1.d FROM t as t1, t as t2 WHERE t2.d=t1.c GROUP BY t2.a; --error ER_WRONG_FIELD_WITH_GROUP SELECT t1.d FROM t as t1, t as t2 WHERE t2.d>t1.c GROUP BY t2.a; DROP TABLE t; --echo # A non-NULL-friendly dependency. CREATE TABLE t ( a INT, c INT GENERATED ALWAYS AS (COALESCE(a,3)) ); INSERT INTO t (a) VALUES(NULL); CREATE TABLE u ( a INT ); INSERT INTO u VALUES(0),(1); --echo # Even though {a}->{c} holds in 't', it doesn't propagate to the left --echo # join's result SELECT t.a,t.c FROM u LEFT JOIN t ON u.a>0; --error ER_WRONG_FIELD_WITH_GROUP SELECT t.a,t.c FROM u LEFT JOIN t ON u.a>0 GROUP BY t.a; DROP TABLE t,u; DROP FUNCTION show_json_object;