3.0.17 Oct 4, 2007
  - Fix t/11 to cope with Class::Trigger no longer supporting multiple
    triggers in one shot

3.0.16 Nov 05, 2006
  - Better error reporting from has_a setup failure

3.0.15 Aug 19 2006
  - Fix ignorage with mutator_name and accessor_name (Ask Bjørn Hansen)

3.0.14 Jan 03 2006
  - Fix breakage with mutator_name (Ask Bjørn Hansen)

3.0.13 Dec 15 2005
  - Use DBI's last_insert_id() where available (David Steinbrunner)
  - Cope better with deleting through a might_have (Rob Brown)
  - Allow setting a false value or NULL in a might_have (Brad Bowman)

3.0.12 Nov 04 2005
  - Use Clone instead of dclone for cloning meta info to reduce required
    perl version (Juan Camacho) and allow for closures (RT#15498)
  - Remove lots of old deprecations

3.0.11 Oct 23 2005

  - search through a has_many can now take hash_ref
  - Fix documentation for SQL wildcards (RT#15145)

3.0.10 Oct 7 2005

  - Set Storable::Deparse so that has_a subrefs can be cloned (Will Ross)
  - Use shorter version of mk_classdata
  - Include t/24 which was accidentally left out of MANIFEST in 3.0.9

3.0.9 Sep 23 2005

  - Fixed bug with shared meta_info (Will Ross)

  - create() has been renamed to insert() to make it much clearer that
    it corresponds to an SQL INSERT rather than lots of different ideas
    as to what create() might mean. create() still (silently) works. It
    will give 'deprecated' warnings from 3.2.0 and will be removed no
    earlier than 3.4.0.

    The before/after create triggers have NOT been renamed yet as I'm
    rethinking some of that. We'll probably end up with more trigger
    points. Comments and suggestions welcome.

3.0.8 Sep 20 2005

  - constraint exceptions now set their 'data' (Dan Collis Puro)
  - method created by has_many can now override method in parent class
     (reported by Christopher H. Laco)
  - factored out transform_sql for easier modification

3.0.7 Sep 17 2005

  - Make Column responsible for accessor() and mutator() so they can be
    set up declaratively as well as programmatically (this means the
    second argument to accessor_name and mutator_name is now a Column
    object rather than its name, so you may need to check your case
    sensitivity)
  - Rename accessor/mutator_name to accessor/mutator_name_for
  - Document that order_by clauses in sort() are passed through
  - Make one-shot 'Essential' set up slightly more explicit in docs
    (Adam Kennedy)
  - Fix HasMany warning typo (argumemt) (Dan Friedman)
  - Remove obsolete SearchGenerator code which had leaked out

3.0.6 Sep 16 2005

  - constrain_column can now take subref constraint
  - Document DBD::AnyData and FreeTDS issues (Matt Trout)
  - Factor out database error handling to _db_error()
  - Switch to Class::Accessor::Fast in Column and Relationship 

3.0.5 Sep 14 2005

  - has_many can take compile time constraints (Cees Hek)
  - has_many can take a cascading delete strategy (deprecating the old,
    undocumented, 'no_cascade_delete' option)
  - columns() can take Class::DBI::Column objects directly
    which can now in turn take options, thus allowing things like:
      __PACKAGE__->columns(dates => Class::DBI::Column->new(
        tdate => { placeholder => 'IF(1, CURDATE(), ?)' }
      )

3.0.4 Sep 13 2005

  Pre-Reqs
    - Note requirement for Scalar::Util 1.08+ (for refaddr)

  Refactorings
    - Move to pluggable Search interface 
       (includes reworking of search approach per Tim Bunce)

3.0.3 Sep 11 2005

  Bug Fixes
    - Ensure object is removed from index when delete()d (Tim Bunce)
    - clear_object_index when new relationship set up (Tim Bunce)
    - Fixed bug where PK values got auto-vivified 
       (Tatsuhiko Miyagawa, Christopher L. Everett, Tim Bunce)
    - Removed 'AS' when aliasing tables; some databases don't like that
    - Properly return -1 from unchanged object updates (Kingsley Kerce)
    - Fixed problems with overloaded stringification of related classes
       (Tim Bunce)
    - Fixed bug where Essential might contain the PK twice 

  Refactorings
    - Split live_object_key for easy subclassing (Tim Bunce)
    - Split out _as_hash() to return underlying data hash
    - Optimised _mk_column_accessors (Maurice Aubrey)
    - Don't hard-code relationship names (Peter Speltz)

  Internals
    - Changed error message when setting up has_a with incorrect column
       (Drew Taylor)

  Documentation
    - Fixed docs for after_update trigger and update (Kingsley Kerce)

3.0.2 Sep 11 2005

  Code
    - No changes

  Pre-Reqs
    - Fixed code to explicitly need 5.6 (rather than just Makefile)
    - Require 'version' for new 3 part versions

  Tests
    - Fixed t/01 to check mutator_name better
    - Added NOT NULL to Primary in t/Blurb to avoid 0.95 regression
    - Added new Test base class Class::DBI::Test::SQLite
    - Changed all remaining uses of eq_set() to is_deeply()

  Documentation
    - Documented $obj->id() in list context (William McKee)
    - Documented cascading delete for might_have (Tom Hukins)
    - Documented MCFK better (plus fixed lots of typos etc) (Tom Hukins)
    - Documented DBIx::ContextualFetch better
    - select_val better (Dave Howorth)
    - Fixed misspelling of Perrin Harkins
    - Fixed documentation for the year constraint (Andy Lester)
    - Fixed set_sql documentation to explain when it creates a method
    - Fixed new_music documentation (Carl Johnstone)
    - Fixed docs for CD columns to show 'reldate' (Mark Thomas)
    - Fixed lots of other tiny doc issues
    - Fixed docs for Essential (defaults to Primary, rather than All)
    - Fixed docs for what gets passed to triggers (Ryan Tate)

3.0.1 Sep 11 2005

    - Code is identical to 0.96
    - New section added to documentation on Release Philosophy
    - Replace eq_set with is_deeply in t/04 to work around Test::More bug

3.0.0 UNRELEASED

    - IDENTICAL to 0.96. Only difference is version renumbering.

0.96 30 April 2004

  New Functionality
    - Maintain live object index so subsequent requests for same row return 
      the same object (Perrin Harkins)
    - connection() is now preferred over set_db()
    - New relationship architecture (see Class::DBI::Relationship)
    - meta_info now returns more information about relationships 
    - search($column => undef) now does an ISNULL search (Bart Lateur,
      Drew Wilson, William McKee)
    - A slice of an iterator in scalar context now gives another
      iterator (Thomas Klausner)
    - Attribute methods can now take slices (Charles Bailey)
    - pass $self to inflation/deflation methods, allowing (amongst other
      things, cross-column processing in a custom method) (Charles Bailey)
    - deflation now happens at distinct trigger points that happen after
      normal before_create/update triggers (Dave Cash)
    - Columns may now set their placeholders:
        Foo->find_column('tdate')->placeholder("IF(1, CURDATE(), ?)");
    - Columns know if they have been constrained

  Documentation
    - add note about needing a RootClass (and fix tests for this)
    - Explain interaction of is_changed and auto-update (William McKee)
    - Better examples for select_val and sql_single (Paul Makepeace)
    - Add note about sprintf interpolation and wildcard searches in
      home-grown queries (Christopher L. Everett)
    - Mention the wiki

  Bug Fixes
    - Better error reporting from failed inflations [Tatsuhiko Miyagawa]
    - You can now have a column named 'first' [Tatsuhiko Miyagawa]
    - inflation and deflation handle edge cases better [Charles Bailey]
    - search_ methods are only generated for SELECT calls
    - add_constructor is now propery subclassable (Thomas Klausner)

  Deprecations
    - Class::DBI::Query is now deprecated. The SQL substitutions are a
      much better approach.
    - sort argument to has_many is now order_by to match search
    - __hasa_rels and __hasa_list are deprecated in favour of meta_info

  Other
    - construct() is no longer a protected method. 
    - skip t/17 on error. There's a bug here I can't find, or fix, yet.

0.95 Fri Jan  9 2004

  Pre-Requisites
    - Now requires Perl 5.6 or greater

  New Functionality
    - constrain_column() adds simple per-column constraints
    - has_many now attempts to work out the reference column by
      examining the foreign class's has_a() declarations, making the 3rd
      argument to has_many mostly obsolete.
    - __TABLE(Other::Class)__ now interpolates the table from the other
      class, allowing avoidance of hard-coding other table names.
    - __ESSENTIAL(alias)__ now prepends the table alias supplied to the
      start of each column name, for avoiding column name clashes.
    - __JOIN(alias alias)__ now inserts the SQL to join the two tables.
    - might_have now allows creating the remote object through a simple
      set of the accessor. [David Jack Olrik]
    - Iterators can now be reset() [Tatsuhiko Miyagawa]

  Documentation
    - count_all, maximum_value_of and minimum_value_of are now
      documented (along with how to implement similar methods)

  Bug Fixes
    - Allow retrieve() to take unnormalised columns, as with search [Connie]
    - Handle stringification before Primary Key(s) are filled 
    - search() can take an overridden accessor name as a key
    - MultiColumn primary keys working better with ColumnGrouper [Perrin
      Harkins, Paul Makepeace, and Charles Bailey]

  Deprecations
    - _single_value_select and _single_row_select deprecated in favour
      of new select_val and select_row methods in Ima::DBI

  Other
    - Class::Accessor can no longer be used directly to add other attributes.
      TEMP columns should be used for this.
    - Ensure all croaks internally go through _croak() [Richard Piacentini]
    - deleted objects now blessed into Class::DBI::Object::Has::Been::Deleted
    - Lots of doc patches [Jesse Sheidlower, Paul Makepeace]
    - All internal attribute accesses now go through helper methods,
      as first step to changing how all this works.
    - set_sql now passes all arguments up to Ima::DBI (so you can turn
      off caching if required) [gfalck]
    - We now warn about column names clashing with any inherited methods,
      not just ones defined in Class::DBI itself [Dan Thill]
    - Silence some test warnings reported by Paul Makepeace

0.94  Wed Aug 27 2003

  New Functionality
    - allow has_a() columns to hold NULL values (Dominic Mitchell)
    - cascading deletions can be turned off by passing argument to
      has_many setup: no_cascade_delete => 1 (undocumented/untested)
      [now implemented as a trigger]
    - classes and objects can be marked as read_only (experimental and
      undocumented, although tested)

  Documentation
    - no longer refer to Class::DBI::Join, which is dead
    - warn about inflating a primary key using has_a
    - rewrite of docs on transactions
    - numerous small tweaks from Jay Strauss, Alexander Karelas and Brad
      Bowman
    - redocument dbi_commit and dbi_rollback that have vanished...

  Bug Fixes
    - make sequences work again (Dominic Mitchell) [now with tests!]
    - interpolation of __TABLE__ etc now happens globally, rather than
      only once [Corion]
    - might_have now properly does a cascading delete
    - Fixed problem where we can't update an object which might_have a
      relationship, but doesn't.
    - might_have's tentative, but broken, support for MCPK has been
      removed.
    - Increase Class::Accessor pre-requisite to 0.18 to fix strange bug
      reported by Casey West.
    - Make sure we don't alter the hashref being passed to create()
    - The after_create() trigger is now called after the object is
      emptied, and will thus be able to see objects in has_a columns
      (by reloading from the DB) [If you were using this
      trigger to alter the columns to be cleaned out this
      can (for now) be done using the 'create' trigger,
      although that's deprecated. Let me know if you really
      need this ability.

  Deprecations
    - move() is deprecated. The concept is broken and will be removed
      unless someone convinces me to keep it.
    - normalize* are no longer necessary, due to the new Column objects,
      which store their normalized and unnormalize form. (Calling these
      will have no effect, other than a warning).
    - has_column deprecated in favour of find_column

  Other
    - init() now takes hash to instantiate with (Tim Bunce)
    - Rewrite of _require_class, with better short-circuiting and error
      message reporting (Tim Bunce)
    - new Class::DBI::Column to represent a column. Some functionality
      moved from ColumnGrouper to here.
    - mutator and accessor name overrides now stored in column, and not
      looked up every time.
    - attempt to trap and propogate all DBI failures 

0.93  Wed Jul 2 2003

  New Functionality
    - Multi-column primary keys are now supported. [with huge thanks to Tim
      Bunce]
    - New syntax for inbuilt SQL
    - classes can control stringification via as_string() and
      stringify_column().
    - set_sql automatically translates __TABLE__, __ESSENTIAL__, and
      __IDENTIFIER__

  Performance Improvements
    - use Storable's dclone() rather than Data::Dumper's deepcopy

  Documentation
    - document the order_by option to search() [Drew Taylor]
    - fix the documentation of the DBI connect string [Phil Crow]
    - document working with pre_create triggers
    - add a note on dynamically generating a database connection
    - avoid using (shift => $val) due to quoting issues! [Simon Wilcox]
    - refer to Class::DBI::Oracle for sequences [Jay Strauss]
    - give better explanation for construct() [Jay Strauss]
    - document get/set with multiple arguments [Jay Strauss]
    - more detailed description of arguments to has_a()

  Bug Fixes
    - Bug with setting driver defaults [Jay Strauss + Schwern]
    - Bug where selecting a column that was only in 'All' group also
      attempted to fetch any TEMP columns. [Dominic Mitchell]

  Removal of Functionality
    - Undocumented ordered_search method has been removed in favour of
      order_by option to search.
    - Undocumented make_filter method removed in favour of add_constructor
    - add_constructor no longer does (undocumented) %s substitutions
    - (undocumented) single_value_select() method no longer takes raw
      SQL fragments.

0.92 Sat Apr 12 2003

  New Functionality
    - classes named after reserved SQL words can now supply a
      table_alias either directly, or as a second argument to table().

  Deprecations
    - hasa() now calls has_a() with a warning. It will disappear
      completely in a forthcoming release.
    - issue warning on use of old hasa_list

  Bug Fixes
    - class data properly propagates (i.e. two classes can once again
      have has_a relationships of the same name) [Miyagawa + Marty]
    - only deflate columns that are changing on update.

  Other
    - croak if update() does not change exactly one row [Tim Bunce].
      (Not yet live)
    - The Essential column group now defaults to Primary rather than
      All. 

0.91 Sat Mar  8 2003

  New Functionality
    - has_many can now take an extra set of search parameters at execution
    - has_many mapping method can now be a list
    - Class->delete(@search_args) deletes all results matching search
      criteria.  
    - search can take a final hash of arguments. Currently the only one
      honoured is 'order_by'
    - searches deflate objects passed as search values as with has_a()
    - iterators can be slice()d
    - classes can provide their own iterators 
    - columns in the TEMP column group will be non-persistent
    - set_sql automatically sets up a search_ method for that SQL
    - constraints can cross-check using new syntax, and per_column
      trigger points.
    - the after_update trigger can modify the list of changed columns
      to change refresh behaviour 
    - Can delete via the iterator: $cd->artists->delete_all
    - New Class::DBI::Query module for dynamically constructing queries
    - Standardised exception handling (thanks to Tim Bunce)
      - on_failed_create no longer needed
    - overloading of objects
    - autoincrements work with SQLite

  Deprecations
    - commit() and autocommit() are now update() and autoupdate()
    - rollback() is now discard_changes()
    - hasa() now warns if used
    - trigger points create() and delete() warn when called
    - trigger point on_setting no longer exists

  Optimisation
    - has_a lazy inflates other Class::DBI instances 
    - roll our own require rather than depending on UNIVERSAL::require
     (thanks to Tim Bunce)
    - Don't add __Changed = {} to every object on creation,
      just autovivify it on demand. (Tim)
    - Avoid calling $self->primary_column in a loop in create() (Tim)
    - Use fetchrow_arrayref instead of fetchrow_array. (Tim)
    - Return from _deflated_column earlier if value is not a ref. (Tim)
    - Use $sth->{NAME_lc} instead of $sth->{NAME}. (Tim)
    - Optmize _normalized and normalize. (Tim)
    - optimise various calls to columns() (Tim)
    - don't reflesh after an update - defer for lazy loading
    - Iterator no longer inherits from Class::DBI itself.

  Bug Fixes
    - has_many maps now work as iterators
    - columns only report as being in group 'All' if they're in no
      other columns (so lazy loading works in those cases!)
    - calling create with a primary key of zero no longer attempts to 
      use auto_increment (Tim Bunce)

  Other
    - reference Class::DBI::AbstractSearch
    - retrieve_from_sql removes any leading 'WHERE' 
    - inflate handles overloading better (thanks to Tim Bunce)
    - Avoid boolean test on object, use defined instead.
    - Tweak create() to prepare for possible caching of colmap info.
    - better documentation on transactions
    - test suite now uses DBD::SQLite instead of DBD::CSV
    - consolidated some test scripts

0.90 Wed Nov 27 GMT 2002
    - hasa and associated_class merged (and deprecated) into has_a
      - has_many therefore no longer sets up reciprocal hasa
    - allow search() and search_like() to take multiple columns
    - added find_or_create()
    - add_constructor() is now preferred to make_filter()
    - added retrieve_from_sql() for inline SQL
    - documented running arbitrary SQL and using sth_to_objects to 
      convert these to objects. (Making sth_to_objects public too)
    - has_many can now call a mapping method on the results
      (for simpler many-to-many joins, for example)
    - has_many adds an 'add_to_<whatevers>' method 
      e.g.  CD->has_many(tracks => Track);
      now adds not just 'tracks' method, but also 'add_to_tracks'.
      (thanks to Michael Styer)
    - try to guess table name if none given
    - speed up iterators (thanks to Tom Renfro)
    - added (undocumented) data_type method
      (thanks to Tatsuhiko Miyagawa)
    - renamed column_type() to associated_class()
    - added references to Class::DBI::SQLite, Class::DBI::Pg,
      and Class::Join
    - before_create trigger can now modify object itself
    - uses UNIVERSAL::require instead of rolling our own
    - marked primary_key(), is_column(), add_hook() as deprecated
    - changed _ids_to_objects to receieve listref, rather than list
      to cope with weird bug in 5.005_03 (thanks to Tatsuhiko Miyagawa)
    - warn if column name clashes with built-in method
    - third argument to has_many is now optional, defaulting to our
      'class-name' (undocumented)
    - provide normalised method names if appropriate (i.e. a Film column
      will give us methods $obj->Film and $obj->film). This previously
      happened inconsistently.
    - split most of the column related code to Class::DBI::Columns
    - remove dependency on Class::Fields
    - reference the new Class::DBI mailing lists
    - Give better errors if no database connection set up 
      (thanks to Simon Cozens)
    - A failed create now calls $class->on_failed_create, which by
      default dies, but can be made to do whatever you like. 

0.89 Mon Jun 24 2002
    - allow has_many to not have a relationship 
      (Thanks to Brian Parker)
    - renamed (undocumented) min() and max() to 
      minimum_value_of() and maximum_value_of()
    - croak() and carp() deprecated in favour of _croak() and _carp()
    - primary() and essential() deprecated in favour of 
      primary_key() and _essential()/columns(essential);
    - normalize_one() deprecated in favour of _normalize_one
    - If a 'might-have' link points to nothing, don't give a 'Can't call
      method' error
    - better reporting on errors from create etc.
    - better reporting when you incorrectly set through hasa()
    - handle primary/foreign key value of zero better
      (reported by Jim O'Brien)
    - added (undocumented) 'column_type' based on ideas from
      Matthew Simon Cavalletto and Tatsuhiko Miyagawa.
    - brought delete triggers in line with documentation
      (reported by Barry Hoggard, fixed by Tatsuhiko Miyagawa)

0.88 Beltane 2000
    - fix for hasa() and create() with modified accessor/mutator names
      (Thanks to Schwern)

0.87 Fri Mar 29 2002
    * added might_have method
    - better MySQL autoincrement code (fixes occassional problems
       under mod_perl) [Thanks to Tatsuhiko Miyagawa]
    - fixed a test that was failing on 5.005 (but not 5.6.1) due to weird 
      interaction between overloaded iterator and Test::More's ok() 
      prototype  [Thanks to Tatsuhiko Miyagawa]

0.86 Fri Mar  8 GMT 2002
    * removed support for pseudo-hashes
    * hasa_list is now has_many 
      (although has_many still exists for backwards compatability)
    + has_many auto-generates reciprocal hasa decalaration
       (unless called as hasa_list, or with { nohasa => 1 })
    * searches now return iterators when used in scalar context
    - hooks are now called 'triggers' (using Class::Trigger)
    * new trigger for SELECT
    * added basic constraints
    + added support for filters on the same column
      (%s >= ? OR %s <= ?) 
    + added 'between' filter which provides this
    - deal better with the case where the only column group is 'All'
    - much internal twiddly stuff

0.36  Wed Nov 28 2001
    + tests no loner die horribly if you don't have MySQL installed
    - minor tweaks to some error messages
    - bugfixes for problems with auto-deleting objects
    - bugfix for mis-normalized accessor warning
       (thanks to Schwern)

0.35  Sun Oct  7 2001
    + added hooks for create, update and delete
    - split commit() for easier subclassing

0.34  Sat Oct  6 2001
    - Don't die if a value is a reference. (Ima::DBI does this for us,
       and better in case of overloaded objects)
    - fix minor problem with mutual hasa / hasa_list referencing
    - better diagnostics if hasa_list is miscalled

0.33  Sat Sep 15 2001
    + Added create_filter(), and with it retrieve_all()
    + added docs on how to set-up many-to-many relationships
    + _cascade delete now split out to allow overriding
    + copy() and move() can now take multiple arguments to change
       (thanks to Jonathan Swartz)

0.32  Sun Sep  9 2001
    + delete() now removes any foreign elements, to avoid orphans

0.31  Sun Sep  9 2001
    + split out _column_placeholder (thanks to Jonathan Swartz)
    + added hasa() checks for orphaned rows

0.29  Wed Sep  5 10:24:54 BST 2001
    + split docs into Class::DBI::Tutorial
    + renamed is_column to has_column (is_column still provided as alias)
    + added accessor_name and mutator_name form template methods
    - _load_class didn't work. Now fixed.
    - _next_in_sequence was broken, and would warn even if it wasn't.
    - better diagnostics when you call retrieve incorrectly

0.28  Wed Aug 29 20:02:28 BST 2001
    + Tony Bowden took over maintenance
    * Added hasa_list
    - Fixed bug with inherited columns (reported by Boris Mouzykantskii)
    - move() and copy() were broken with auto_increments. Now fixed.
    - set() did bad things if you'd set up new fields with
      mk_accessors (reported by Boris Mouzykantskii). Now fixed.
    - Better errors when calling create() incorrectly
    - Split tests into distinct files (DBD::CSV now required)
    - General internal tidying

0.27  Mon Apr 23 09:04:51 BST 2001
    * Class::DBI->new() deprecated in favor of create()
    - Fixed the 'Primary before All' bug reported by Tony Bowden

0.26  Mon Apr  9 14:32:17 BST 2001
    - Class->columns('All', @cols) now assumes the first column to be
      your primary.
    - Class->columns('Essential', @cols) now automatically includes
      the primary column.
    * move() was broken.  Works and tested.
    - Updated our base requirement to get at its bugfix
    - Updated our Ima::DBI requirement to get at commit() and rollback()
    * Documented construct()
    * Added docs about transactions
    * Added dbi_commit() and dbi_rollback()
    - Added docs about Class::DBI and mod_perl

0.25  Wed Jan 10 01:54:27 EST 2001
    - The new ID is now optional for copy()
    * Added move() to move objects between classes/databases.
    - Updated PG's default attributes.
    - Made the way classes are loaded with hasa() safer

0.24  Thu Oct  5 19:07:21 EDT 2000      *UNRELEASED*
    - Fixed some $@ naughtiness.  Errors should propigate properly.
    * Added default database attributes for Pg, MySQL, Oracle, DBD::CSV and
      DBD::RAM.  Class::DBI should work out of the box with them now.

0.23  Tue Sep 12 00:33:38 EDT 2000
    - rollback() needed to normalize its data (thanks to Greg Bartlett)

0.22  Sun Sep 10 05:36:51 GMT 2000      *UNRELEASED*
    - Added a warning if the primary column is not in the essential group

0.21  Sun Sep 10 01:03:01 EDT 2000      *UNRELEASED*
    - Now requiring Ima::DBI 0.24
    - sequence() fixed, tested and working
    - Now working with DBD::Pg and PostgreSQL 7.0

0.20  Sun Sep 10 00:29:15 EDT 2000      *UNRELEASED*
    - *bug_fix* is_column() died if false.

0.19  Fri Sep  8 15:07:49 EDT 2000
    - Forgot to update the Changes file.

0.18  Fri Sep  9 18:51:45 GMT 2000
    * new() now accepts Class::DBI objects as values
    - hasa() is now inherited properly
    - normalize() will now take an empty array

0.17  Mon Sep  4 02:46:25 GMT 2000      *UNRELEASED*
    - Docs forgot to turn AutoCommit on in MySQL examples
    - Documented the behavior of new() and AUTO_INCREMENT
    - Added support for sequences.  *Untested*
    - *bug fix* columns() can now be added after the class has been used
    - *bug fix* Columns added by hasa() had a chance of overwriting each 
      other.

0.16  Sun Sep  3 17:49:08 GMT 2000      *UNRELEASED*
    * hasa() will now attempt to require the foreign class for you
    * hasa() will setup the necessary columns for you.

0.15  Sun Jul 16 23:14:20 PDT 2000
    - Fixed minor bug in DESTROY message.
    * Added hasa() object relationships.

0.14  Sun Jul  9 05:25:53 EDT 2000
    - Pseudohashes as objects now work and basicly tested
    - The rollback() mechanism changed to use less memory.
    - Expanded the TODO list

0.13  Wed May 24 02:45:43 EDT 2000
    - Accessors were not being generated properly in certain cases.

0.12  Tue May 23 ish...                 * Unreleased*
0.11
    - Aborted attempts to fix accessor misgeneration.

0.10  Thu May 18 00:59:34 EDT 2000
    - new() was not working with objects which autogenerated primary keys

0.09  Tue May  2 00:42:06 EDT 2000
    - README, INSTALL and Makefile.PL have been updated

0.08  Mon May  1 20:27:13 EDT 2000
    - Documented lazy column population.

0.07  Mon May  1 02:18:20 EDT 2000      * Unreleased *
    * Added lazy population of columns.
    - columns('All') can now be autogenerated.

0.06  Mon May  1 00:56:52 EDT 2000      * Unreleased *
    * **API CHANGE** make_sql() is gone in favor of set_sql()
    * normalization has been added in all the right places.
    - retreive now handles autoincremented primary keys.
    - columns() internally normalizes column names.

0.05  Sat Apr 15 01:48:25 EDT 2000
    * MAJOR BUG - simple subclasses of subclasses of Class::DBI would not 
      work.  This has been fixed.
    * Autoloader eliminated.
    * API CHANGE - columns must now be explictly declared, the class's
      public data members are not used.  That was silly.
    * rollback() now works.
    - is_changed() now returns a list of changed columns

0.03  Mon Dec 20 10:04:23 EST 1999
    * MAJOR BUG - commit() was not committing!!!
    * added is_changed()
    - columns('Essential') now autogenerated from columns('All') if not 
      already present.
    - columns('Essential') is now prefered for most SELECT statements 
      instead of just columns().

0.02
    * First working version
    * First version sent to CPAN