Collation aggregation for IN works differently for scalar and row arguments.

This script demonstrates the same problem:

SET NAMES utf8;

SELECT

  'i' = 'j' COLLATE utf8_roman_ci  AS c1,

   ('i') IN ('j' COLLATE utf8_roman_ci) AS c2,

   ('i',1) IN (('j' COLLATE utf8_roman_ci,1),('j' COLLATE utf8_roman_ci,1)) AS c3;

+----+----+----+

| c1 | c2 | c3 |

+----+----+----+

|  1 |  1 |  0 |

+----+----+----+

i and j are equal letters in utf8_roman_ci
So:

• c1 doing IN for scalar data returns the expected result
• c2 doing IN for ROW data with one value returns the expected result
• c3 doing IN for ROW data with more than one values returns a wrong result