MDEV-24813 Signal full scan to storage engines.

[mariadb-YuchenPei]

This is an initial patch for discussion.

Will work in the following minimal example

--source include/have_innodb.inc
CREATE TABLE t (a INT PRIMARY KEY) ENGINE=InnoDB;
insert into t values (42);

BEGIN;
SELECT * FROM t LOCK IN SHARE MODE; # will acquire S lock
COMMIT;

DROP TABLE t;

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[spetrunia]

Also, I've tried this testcase

create table t10 (a int, b varchar(100), index(a));
insert into t10 select seq, uuid() from seq_1_to_10000;
create table t11(a int);
explain insert into t11 select a from t10;
insert into t11 select a from t10;

and init_read_record() isn't invoked for it.

The index scan on t10 is initialized here:

#0  ha_innobase::index_init (this=0x7fffb80b6eb0, keynr=0)   at /home/psergey/dev-git/12.3-no-lock/storage/innobase/handler/ha_innodb.cc:8767
#1  0x0000555555f0a5a0 in handler::ha_index_init (this=0x7fffb80b6eb0,     idx=0, sorted=true)    at /home/psergey/dev-git/12.3-no-lock/sql/handler.h:3527
#2  0x0000555556125194 in join_read_first (tab=0x7fffb80cae40)    at /home/psergey/dev-git/12.3-no-lock/sql/sql_select.cc:24768

[spetrunia]

With this patch, we have "Using where with pushed condition" all over the place:

MariaDB [test]> explain select * from t1 where b <1;
+------+-------------+-------+------+---------------+------+---------+------+-------+-----------------------------------+
| id   | select_type | table | type | possible_keys | key  | key_len | ref  | rows  | Extra                             |
+------+-------------+-------+------+---------------+------+---------+------+-------+-----------------------------------+
|    1 | SIMPLE      | t1    | ALL  | NULL          | NULL | NULL    | NULL | 10157 | Using where with pushed condition |
+------+-------------+-------+------+---------------+------+---------+------+-------+-----------------------------------+
1 row in set (0.002 sec)

[mariadb-YuchenPei]

Also, I've tried this testcase

create table t10 (a int, b varchar(100), index(a));
insert into t10 select seq, uuid() from seq_1_to_10000;
create table t11(a int);
explain insert into t11 select a from t10;
insert into t11 select a from t10;

and init_read_record() isn't invoked for it.

The index scan on t10 is initialized here:

#0  ha_innobase::index_init (this=0x7fffb80b6eb0, keynr=0)   at /home/psergey/dev-git/12.3-no-lock/storage/innobase/handler/ha_innodb.cc:8767
#1  0x0000555555f0a5a0 in handler::ha_index_init (this=0x7fffb80b6eb0,     idx=0, sorted=true)    at /home/psergey/dev-git/12.3-no-lock/sql/handler.h:3527
#2  0x0000555556125194 in join_read_first (tab=0x7fffb80cae40)    at /home/psergey/dev-git/12.3-no-lock/sql/sql_select.cc:24768

You are right. I had forgotten to make the call in join_read_first (and join_read_last). Added the calls to ha_extra in a new commit now.

[mariadb-YuchenPei]

With this patch, we have "Using where with pushed condition" all over the place:

No longer the case with the new patch that does not comment out the cond_push if condition

[iMineLink]

Please check my few minor comments, especially the ones around innodb_full_scan.test, thanks.

[iMineLink]

Thanks for addressing my review comments, LGTM.