Bug #77473 Truncated data with subquery & UTF8

REPEAT() and RPAD() results in a subquery get truncated if the length parameter is between 32768 and 65536 for 2-byte UTF-8 sequences, or between 21845 and 65535 . Below and above that point the result is correct.

set names utf8;

select length(data) as len from ( select repeat('ä', ...) as data ) as Sub;

parameter -> result

36766 -> 65532

32767 -> 65534

32678 -> 65534

...

65535 -> 65534

65536 -> 131072

65537 -> 131074

This is not a problem of the length function in the outer query, it is already the string returned from the inner query that gets truncated as can be seen with

mysql -Ne "set names utf8; select data from (select repeat('é', 40000) as data ) as Sub;" | wc

      1       1   65535

Looks as if the intermediate result length is calculated by character count and not by byte length, and so the wrong data type seems to be used to store the intermediate result. E.g. when using a unicode character that has a 3 byte UTF-8 representation:

select length(data) as len from ( select repeat('☃', ...) as data ) as Sub;

parameter -> result

21844 -> 65532

21845 -> 65535

21846 -> 65535 

...

65535 -> 65535

65536 -> 196608