This migration requires a complete dump of the 6.2 or 6.2.1 database and a
restore of the database in 6.3.

There are some general 6.3 issues that I want to mention.  These are
only the big items that can not be described in one sentence.  A review
of the HISTORY files is still needed.

First, we now have subselects.  Now that we have them, I would like to
mention that without subselects, SQL is a very limited language.
Subselects are a major feature, and you should review your code for
places where subselects provide a better solution for your queries.  I
think you will find that there are more uses for subselects than you may
think.  Vadim has put us on the big SQL map with subselects, and fully
functional ones too.  The only thing you can't do with subselects is to
use them in the target list.

Second, 6.3 uses unix domain sockets rather than TCP/IP by default.  To
enable connections from other machines, you have to use the new
postmaster -i option, and of course edit pg_hba.conf.  Also, for this
reason, the format of pg_hba.conf has changed.

Third, char() fields will now allow faster access than varchar() or
text. Specifically, the text and varchar() have a penalty for access to
any columns after the first column of this type.  char() used to also
have this access penalty, but it no longer does.  This may suggest that
you redesign some of your tables, especially if you have short character
columns that you have defined as varchar() or text.  This and other
changes make 6.3 even faster than earlier releases.

We now have passwords definable independent of any Unix file.  There are
new SQL USER commands.  See the pg_hba.conf manual page for more
information.  There is a new table, pg_shadow, which is used to store
user information and user passwords, and it by default only SELECT-able
by the postgres super-user.  pg_user is now a view of pg_shadow, and is
SELECT-able by PUBLIC.  You should keep using pg_user in your
application without changes.

User-created tables now no longer have SELECT permission to PUBLIC by
default.  This was done because the ANSI standard requires it.  You can
of course GRANT any permissions you want after the table is created. 
System tables continue to be SELECT-able by PUBLIC.

We also have real deadlock detection code.  No more sixty-second
timeouts.  And the new locking code implements a FIFO better, so there
should be less resource starvation during heavy use.

Many complaints have been made about inadequate documenation in previous
releases.  Thomas has put much effort into many new manuals for this
release.  Check out the /doc directory.

For performance reasons, time travel is gone, but can be implemented
using triggers (see pgsql/contrib/spi/README).  Please check out the new
\d command for types, operators, etc.  Also, views have their own
permissions now, not based on the underlying tables, so permissions on
them have to be set separately.  Check /pgsql/interfaces for some new
ways to talk to PostgreSQL.

This is the first release that really required an explaination for
existing users.  In many ways, this was necessary because the new
release removes many limitations, and the work-arounds people were using
are no longer needed.

Long live PostgreSQL.

-- Bruce Momjian