5.0.19 has added a field to the MYSQL struct that was not there before. For clients that use a MYSQL struct they have allocated themselves (rather than allowing mysql_init to allocate it), this breaks binary compatibility, because they won't have made the struct large enough.

$ diff -c mysql-5.0.18/include/mysql.h mysql-5.0.19/include/mysql.h ... *************** *** 287,292 **** --- 287,294 ---- from mysql_stmt_close if close had to cancel result set of this object. */ my_bool *unbuffered_fetch_owner; + /* needed for embedded server - no net buffer to store the 'info' */ + char *info_buffer; } MYSQL;

typedef struct st_mysql_res { ***************

I don't know why the MYSQL struct wasn't made abstract in the first place, but given that mysql has chosen to expose it to client apps, they can't change it (not this much anyway) without creating an ABI break.

No kidding. Red Hat has backed off our intention to ship 5.0.19 at all. What I want to know is, should I file a bug report or is this change intentional? (And if the latter, why wasn't there a .so version bump?)