Wrong comparision of BLOB's empty preffix with non-preffixed BLOB causes rows count mismatch for clustered and secondary indexes during non-locking read

origin/10.7 commit 3351dfaab0599268eaf25f9d6995ef128910a8b9

SET GLOBAL default_storage_engine = 'InnoDB' ;

SET SESSION sql_mode = 'NO_ENGINE_SUBSTITUTION' ;

CREATE TABLE IF NOT EXISTS t1 ( col_text TEXT NOT NULL, KEY ( col_text(9) ) ) ENGINE = InnoDB ROW_FORMAT = Dynamic ;

SET @fill_amount = (@@innodb_page_size / 2 ) + 1 ;

INSERT INTO t1 (col_text) VALUES ( REPEAT(SUBSTR(CAST( 24 - 1 AS CHAR),1,1), @fill_amount) ) ;

UPDATE t1 SET col_text = REPEAT(SUBSTR(CAST( NULL AS CHAR),1,1), @fill_amount) ;

Warnings:

Warning 1048    Column 'col_text' cannot be null

UPDATE t1 SET col_text = REPEAT(SUBSTR(CAST( NULL AS CHAR),1,1), @fill_amount) ;

Warnings:

Warning 1048    Column 'col_text' cannot be null

UPDATE t1 SET col_text = REPEAT(SUBSTR(CAST( 18 AS CHAR),1,1), @fill_amount) ;

CHECK TABLE t1 ;

Table   Op      Msg_type        Msg_text

test.t1 check   Warning InnoDB: Index 'col_text' contains 2 entries, should be 1.

test.t1 check   error   Corrupt

I cannot exclude that the uploaded MTR test contains more SQL than required for showing the problem.

Warning:

The layout of the table with one column only comes from the attempt to get the smallest test possible.

In case the table contains additional columns of type INT and/or CHAR/VARCHAR I get the same bad effect.