Unfortunately MariaDB server allows to store invalid utf8mb3 characters. This might end up in a possible crash of C/Python during fetch, since PyUnicode_GET_LENGTH doesn't perform any validation checks.

Instead of using the more secure function PyUnicode_GetLength (it does validate the Object) we shouldn't calculate length at all - this would also avoid to "overwrite" previous exception.

How to repeat:

import mariadb

conn= mariadb.connect(db="test")

cursor= conn.cursor(binary=True)

cursor.execute("CREATE OR REPLACE TABLE t1 (a varchar(20) charset utf8mb3)")

cursor.execute("INSERT INTO t1 VALUES (0xEDA080)")

conn.commit()

cursor.execute("SELECT a FROM t1")

row= cursor.fetchone()