=pod
=head1 NAME
Alzabo::Design - Documentation on Alzabo's design
=head1 DESCRIPTION
This document describes some of the Alzabo's design.
=head1 ARCHITECTURE
There are objects representing the schema, which contains table
objects. Table objects contain column, foreign key, and index
objects. Column objects contain column definition objects. A single
column definition may be shared by multiple columns, but has only one
owner.
This is a diagram of these inheritance relationships:
Alzabo::* (::Schema, ::Table, ::Column, ::ColumnDefinition, ::ForeignKey, ::Index)
/ \
is parent to
/ \
Alzabo::Create::* Alzabo::Runtime::*
This a diagram of how objects contain other objects:
Schema - makes--Alzabo::SQLMaker subclass object (many)
/ \
contains contains--Alzabo::Driver subclass object (1)
| \
Table (0 or more) Alzabo::RDBMSRules subclass object (1)
/ \ (* Alzabo::Create::Schema only)
/ \
contains--------------------
/ \ \
/ \ \
ForeignKey Column (0 or more) Index (0 or more)
(0 or more) |
contains
|
ColumnDefinition (1)
Note that more than one column I<may> share a single definition object
(this is explained in the
L<C<Alzabo::Create::ColumnDefinition>|Alzabo::Create::ColumnDefinition>
documentation). This is only relevant if you are writing a schema
creation interface.
=head2 Other classes
=over 4
=item * C<Alzabo::Driver>
These objects handle all the actual communication with the database,
using a thin wrapper over DBI. The subclasses are used to implement
functionality that must be handled uniquely for a given RDBMS, such as
creating new values for sequenced columns.
=item * C<Alzabo::SQLMaker>
These objects handle the generation of all SQL for runtime operations.
The subclasses are used to implement functionality that varies between
RDBMS's, such as outer joins.
=item * C<Alzabo::RDBMSRules>
These objects perform several funtions. First, they validate things
such as schema or table names, column type and length, etc.
Second they are used to generate SQL for creating and updating the
database and its tables.
And finally, they also handle the reverse engineering of an existing
database.
=item * C<Alzabo::Runtime::Row> and C<Alzabo::Runtime::RowState::*>
The C<Alzabo::Runtime::Row> class represents a single row. These
objects are created by
L<C<Alzabo::Runtime::Table>|Alzabo::Runtime::Table>,
L<C<Alzabo::Runtime::RowCursor>|Alzabo::Runtime::RowCursor>, and
L<C<Alzabo::Runtime::JoinCursor>|Alzabo::Runtime::JoinCursor> objects.
It is the sole interface by which actual data is retrieved, updated,
or deleted in a table.
The various C<RowState> classes are used in order to change a row's
behavior depending on whether it is live, live and cached, potential,
or deleted.
=item * C<Alzabo::Runtime::JoinCursor> and C<Alzabo::Runtime::RowCursor>
These objects are cursor that returns row objects. Using a cursor
saves a lot of memory for big selects.
=item * C<Alzabo::Runtime::UniqueRowCache>
Loading this class turns on Alzabo's simple row caching mechanism.
=item * C<Alzabo::Config>
This class is generated by Makefile.PL during installation and
contains information such as what directory contains saved schemas and
other configuration information.
=item * C<Alzabo::ChangeTracker>
This object provides a method for an object to register a series to
backout from multiple changes. This is done by providing the
ChangeTracker object with a callback after a change is succesfully
made to an object or objects. If a future change in a set of
operations fail, the tracker can be told to back the changes out. This
is used primarily in
L<C<Alzabo::Create::Schema>|Alzabo::Create::Schema>.
=item * C<Alzabo::MethodMaker>
This module can auto-generate useful methods for you schema, table,
and row objects based on the structure of your schema.
=item * C<Alzabo::Exceptions>
This object creates the exception subclasses used by Alzabo.
=back
=head1 WHY THE SUBDIVISION BETWEEN Alzabo::*, Alzabo::Create::*, and Alzabo::Runtime::*?
There are several reasons for doing this:
=over 4
=item *
In some environments (mod_perl) we would like to optimize for memory.
For an application that uses an existing schema, all we need is to be
able read object information, rather than needing to change the
schema's definition. This means there is no reason to have the
overhead of compiling all the methods used when creating and modifying
objects.
=item *
In other environments (for example, when running as a separately
spawned CGI process) compile time is important.
=item *
Many people using Alzabo will use the schema creation GUI and then
write an application using that schema. At the simplest level, they
would only need to learn how to instantiate C<Alzabo::Runtime::Row>
objects and how that class's methods work. For more sophisticated
users, they can still avoid having to ever look at documentation on
methods that alter the schema and its contained objects.
=back
=head1 RATIONALE FOR CURSORS
Using cursors is definitely more complicated. However, there are two
excellent reasons for using them: speed and memory savings. As an
example, I did a test with the old code (which returned all its
objects at once) against a table with about 8,000 rows using the
L<C<Alzabo::Runtime::Table-E<gt>all_rows>
method|Alzabo::Runtime::Table/all_rows>. Under the old
implementation, it took significantly longer to return the first row.
Even more importantly than that, the old implementation used up about
10MB of memory versus about 4MB! Now imagine that with a 1,000,000
row table.
Thus Alzabo uses cursors so it can scale better. This is a
particularly big win in the case where you are working through a long
list of rows and may stop before the end is reached. With cursors,
Alzabo creates only as many rows as you need. Plus the start up time
on your loop is much, much quicker. In the end, your program is
quicker and less of a memory hog. This is good.
=head1 AUTHOR
Dave Rolsky, <autarch@urth.org>
=cut