Hi Sergei, I tested comparison behaviour for various different situations. Please find the comment with the summary table in the end of: https://mariadb.atlassian.net/browse/MDEV-7649 Currently there are 5 possible reactions on bad bytes on comparison, depending on the collation, presence of an index, and character_set_connection value: - truncate on bad byte, compare only the well-formed prefix (#1) - treat bad bytes as '?' (#7 and #8) - empty set with no warning (#3) - empty set with a warning (#2) - error (#5 and #6) I'm in doubts what to dofor 10.1 and for 5.5. Please suggest. 1. For 10.1 or 10.2 I think it would be nice to make all these cases work exactly the same way. I am not sure which way would be the best. Any ideas? 2. For 5.5 we need to fix the bug with minimal changes. Do you agree? I suggest we fix only #1 and maybe #2. How we fix 5.5 now should probably depend on how we'll fix 10.1, to avoid changing behaviour many times. Please suggest. Thanks.