Bug #25251 libmysql.c:4322: setup_one_fetch_function: Assertion `param->buffer_length != 0'

Description:
Hello,

The following statements will not execute properly from DBI:

SELECT '';
SELECT SUBSTR('a', 2);
SELECT CONCAT_WS('','');
SELECT REPEAT(-1,'a');

DBI fails with:
perl: libmysql.c:4322: setup_one_fetch_function: Assertion `param->buffer_length != 0' failed.

How to repeat:
my $sth = $dbh->prepare("SELECT ''");
$sth->execute();

$dbh->do() does not trigger this behavoir (possibly because the new-style MySQL binary protocol does not kick in).

Suggested fix:
What happens is that DBD's dbd_describe() calls mysql_fetch_fields() which returns a  fields[0].length of 0. This is then passed to mysql_stmt_bind_result which calls setup_one_fetch_function() which causes the assertion since buffer_length can not be zero if type is MYSQL_TYPE_STRING.

It appears that the length is zero because all those functions in those special cases return &my_empty_string which probably has a length of zero which gets propagated all the way up to the caller.

IF it is OK to have a buffer MYSQL_BIND structure with buffer_length == 0 (that is, if it is trailing \0 safe), then the assertion should be removed.

server_side_prepare must be enabled and working for this bug to manifest itself via DBI (though it does not look like a bug of DBD::mysql). Please make sure you have the latest DBD::mysql and that DBI tracing shows that dbd_describe() was called indeede.

Thank you for the report.

Verified as described using DBD-4.00 and last BK MySQL server.

With simple C test program error is not repeatable.

This bug can't be reproduced anymore. In the libmysql sources the assert is missing.

The test cases work as expected.