New (backward-compatible) connect protocol that allows you to specify the database to use when connecting, to get much faster connections to a specific database.

The mysql_real_connect() call is changed to:

mysql_real_connect(MYSQL *mysql, const char *host, const char *user,
                   const char *passwd, const char *db, uint port,
                   const char *unix_socket, uint client_flag)

Each connection is handled by its own thread, rather than by the master accept() thread. This fixes permanently the telnet bug that was a topic on the mail list some time ago.

All TCP/IP connections are now checked with backward-resolution of the hostname to get better security. mysqld now has a local hostname resolver cache so connections should actually be faster than before, even with this feature.

A site automatically will be blocked from future connections if someone repeatedly connects with an ``improper header'' (like when one uses telnet).

You can now refer to tables in different databases with references of the form tbl_name@db_name or db_name.tbl_name. This makes it possible to give a user read access to some tables and write access to others simply by keeping them in different databases!

Added --user option to mysqld, to allow it to run as another Unix user (if it is started as the Unix root user).

Added caching of users and access rights (for faster access rights checking)

Normal users (not anonymous ones) can change their password with mysqladmin password "new_password". This uses encrypted passwords that are not logged in the normal MySQL log!

All important string functions are now coded in assembler for x86 Linux machines. This gives a speedup of 10% in many cases.

For tables that have many columns, the column names are now hashed for much faster column name lookup (this will speed up some benchmark tests a lot!)

Some benchmarks are changed to get better individual timing. (Some loops were so short that a specific test took < 2 seconds. The loops have been changed to take about 20 seconds to make it easier to compare different databases. A test that took 1-2 seconds before now takes 11-24 seconds, which is much better)

Re-arranged SELECT code to handle some very specific queries involving group functions (like COUNT(*)) without a GROUP BY but with HAVING. The following now works:

mysql> SELECT COUNT(*) as C FROM table HAVING C > 1;

Changed the protocol for field functions to be faster and avoid some calls to malloc().

Added -T32 option to mysqld, for running all queries under the main thread. This makes it possible to debug mysqld under Linux with gdb!

Added optimization of not_null_column IS NULL (needed for some Access queries).

Allow STRAIGHT_JOIN to be used between two tables to force the optimizer to join them in a specific order.

String functions now return VARCHAR rather than CHAR and the column type is now VARCHAR for fields saved as VARCHAR. This should make the MyODBC driver better, but may break some old MySQL clients that don't handle FIELD_TYPE_VARCHAR the same way as FIELD_TYPE_CHAR.

CREATE INDEX and DROP INDEX are now implemented through ALTER TABLE. CREATE TABLE is still the recommended (fast) way to create indexes.

Added --set-variable option wait_timeout to mysqld.

Added time column to mysqladmin processlist to show how long a query has taken or how long a thread has slept.

Added lots of new variables to show variables and some new to show status.

Added new type YEAR. YEAR is stored in 1 byte with allowable values of 0, and 1901 to 2155.

Added new DATE type that is stored in 3 bytes rather than 4 bytes. All new tables are created with the new date type if you don't use the --old-protocol option to mysqld.

Fixed bug in record caches; for some queries, you could get Error from table handler: # on some operating systems.

Added --enable-assembler option to configure, for x86 machines (tested on Linux + gcc). This will enable assembler functions for the most important string functions for more speed!