NAME
     OOPS - Object Oriented Persistent Store

SYNOPSIS
     use OOPS;

     transaction(sub {
            $oops = new OOPS 
                    dbi_dsn => $DBI_DSN, 
                    username => $username, 
                    password => $password,
                    table_prefix => "MY";

            $oops->{h} = { a => b};
            $oops->{foobar} = $myobject;
            $oops->{abcd} = getref(%{$oops->{h}}, 'a');

            $oops->commit;
     });

DESCRIPTION
    The goal of OOPS is to make perl objects transparently persistent. OOPS
    handles deeply nested and cross-linked objects -- even object
    hierarchies that are too large to fit in memory and (with a hint)
    individual hash tables that are too large for memory. Objects will be
    demand-loaded into memory as they are accessed. All changes to your
    object hierarchy will be saved with a single commit().

    Full transactional consistenty is the only operational mode. Either all
    of your changes are saved or none of them are. During your program, you
    will see a consistent view of the data: no other running transactions
    will change the data you see. If another transaction changes data that
    you are using then at least one of the transactions must abort. This
    happens automatically.

    OOPS maps all perl objects to the same RDBMS schema. No advance schema
    definition is required on the part of the user of OOPS. The name comes
    from the realization that perl's data model is much more complicated
    than I initially understood. Internally, the RDBMS schema uses four
    tables: a table of objects, a table of attributes (keys and values), a
    table of large attributes that big the normal column widths, and a table
    of counters.

    At this time, OOPS is expecting a web-like workflow:

     create OOPS instance

     access some ojbects

     modify some objects

     commit

     exit

    If you need more than one transaction in a program, create more than one
    OOPS instance.

    To make your data persistent, make a reference to your data from the
    OOPS object. To later retrieve your data, simply access it through the
    OOPS object.

EXAMPLE PROGRAM
     use OOPS;

     transaction(sub {
            my $oops = new OOPS 
                    dbi_dsn => 'DBI:mysql:database=MY-DATABASE-NAME;host=localhost', 
                    username => 'MY-USERNAME', 
                    password => 'MY-PASSWORD',
                    table_prefix => "MY-TABLE-PREFIX";

            my $p = $oops->{pages}{"/some/path"};

            $p->{next} = $oops->{pages}{"/some/other/path"};
            $p->{jpgs} = [ read_file("x.jpg"), read_file("y.jpg") ];

            $oops->commit;
     });

     exit;

SUPPORTED DATA TYPES
    Perl HASHes, REFs, SCALARs, and ARRAYs are supported. Currently, HASH
    keys may not be longer than 255 characters. Class names may not be more
    than 128 characters long. References to hash keys and array elements are
    supported.

    At the current time, large ARRAYs are not efficient. Use HASHes instead
    if this matters to you. References to array elements and hash values are
    not efficient.

    Large HASHes are supported by only loading keys that are accessed.

    HASHes, array elements, and REFs are implmented with tie(). ARRAYs are
    not currently tie()d because of bugs in perl. Multiple references to the
    same scalar are supported. References to array elements and hash values
    are supported. Persistent data is reference counted and cycles must be
    manually broken to assure de-allocation.

SUPPORTED PLATFORMS
    The following RDBMSs are supported in the code:

    PostgreSQL     OOPS has been tested with PostgreSQL version 7.4.2 and
                   7.3.5. 7.4.2 on Linux 2.4.23 performs okay. 7.3.5 on
                   pre-release DragonflyBSD is slow.

    mysql          OOPS has been tested with mysql 4.0.16 and 4.0.18 using
                   InnoDB tables.

    SQLite         OOPS has been tested with DBD::SQLite 0.31.

                   SQLite is not 8-bit clean with respect to "\0". OOPS uses
                   a DBD::SQLite feature to translate binary nulls. A
                   side-effect is that backslash will be doubled "\" ->
                   "\\". OOPS does not use the "counters" table with SQLite.

                   <http://www.hwaci.com/sw/sqlite/>

    Perl versions 5.8.2 through 5.8.4 are supported. Prior to 5.8.2, it
    wasn't possible to untie scalars from within the a tied scalar access
    method. An ugly workaround is possible if there is enough interest.

    OOPS has been tested on Linux 2.4.23 (Debian unstable as of April '04);
    on FreeBSD 4.9; and on DragonflyBSD prerelease.

    As far as performance goes, mysql and SQLite are both about twice as
    fast as PostgreSQL for applications that only have one transaction at at
    time. SQLite is particularly slow when there are multiple transactions
    as its lock grangularity is the entire database. Mysql also slows down
    when there are multiple simultaneous transactions. I don't know yet
    which backend is fastest for applications with many simultanous
    transactions.

    The RDBMS schema for SQLite is different than for mysql or PostgreSQL
    (which are the same as each other).

FUNCTIONS
    "transaction($funcref, @args)"
        "transaction()" is a wrapper for a complete transaction.
        Transactions that fail due to deadlock with other processes will be
        re-run automatically.

        The first parameter is a reference to a function. Any additional
        parameters will be passed as parameters to that function. The return
        value of "transaction()" is the return value of "&$funcref()".

        It is not necessary to use the "transaction()" method. Beware that
        nearly any operation on persistent date (even read operations) can
        cause deadlock.

        Any use of persistent data can trigger a deadlock. The
        "transaction()" function catches this and retries automatically upto
        $OOPS::transaction_maxtries times (15 times unless you change it).
        If you don't use "transaction()" you might want to catch the
        exceptions that "transaction()" catches. To do this, you can regex
        match $@ against $OOPS::transfailrx.

        Basically, transaction is a slightly fancier version of the
        following:

         sub transaction
         {
                my ($sub, @args) = @_;
                for (0..15) {
                        eval {
                                &$sub(@args);
                        }
                        last unless $@;
                        next if $@ =~ /$transfailrx/;
                        die $@;
                }
         }

        The important thing to notice is that your code will be called over
        and over until the transaction suceeeds. This means you should write
        your code so that it doesn't have any external side effects until
        after it calls commit().

    "getref(%hash, $key)"
        References to tied hash keys are buggy in all perls through 5.8.3.
        Use getref(%hash, $key) to create your reference to a tied hash key.
        See: <http://rt.perl.org/rt3/Ticket/Display.html?id=27555> and
        <http://rt.perl.org/rt3/Ticket/Display.html?id=29224>.

         $ref = getref(%hash, $key);

        Alternatively, use "$oops->workaround27555($ref)".

        Getref() and workaround27555() work around all the perl bugs with
        tied hash key references. Failure to use them may result in
        unexpected and inconsistent results.

PUBLIC CLASS METHODS
    "OOPS->new(%parameters)"
        Creates a OOPS object instance. More than one object instance is
        allowed at the same time. Instances are unaware of each other.
        Making a reference from a persistent object in one instance to a
        persistent object in another instance will cause recursive copying
        from one instance to the other.

        The %parameters are documented in the next section.

    "OOPS->initial_setup(%parameters)"
        Drops and recreates the database tables. Don't use it too often :-)
        The regression suite drops and re-creates the tables many times. Be
        careful.

        The %parameters are documented in the next section.

PARAMETERS FOR NEW
    The "new()" and "initial_setup()" methods take a hash specification for
    their behavior. Here are the parameters allowed:

    "dbi_dsn" / $ENV{OOPS_DSN} / $ENV{OOPS_DRIVER} / $ENV{DBI_DSN} /
    "$ENV{DBI_DRIVER"
        Many ways to specify the DSN for DBI: as an argument; as an OOPS
        environment variable; as a DBI environment variable. Use at least
        one :-) See the DBI documentation for more details.

    "user" / $ENV{OOPS_USER} / $ENV{DBI_USER}
        Three ways to specify the user for DBI: as an argument; as an OOPS
        environment variable; as a DBI environment variable. Not required
        for all databases.

    "password" / $ENV{OOPS_PASS} / $ENV{DBI_PASS}
        Three ways to specify the password for DBI: as an argument; as an
        OOPS environment variable; as a DBI environment variable. Not
        required for all databases.

    "table_prefix" / $ENV{OOPS_PREFIX}
        OOPS allows a prefix to be supplied for it's internal table names.
        If you set a prefix of "FOO_" then it will use a "FOO_object" table
        instead of an "object" table. This can be set as an argument to
        "new()" or it can be set with the environment variable $OOPS_PREFIX.
        This allows multiple separate object spaces to exist within the same
        SQL database. It's intended use is to support testing vs. production
        environments.

    "auto_upgrade" / $ENV{OOPS_UPGRADE}
        The relational schema for OOPS is not same for all versions of OOPS.
        If this is set, then an older schema will be upgraded to the current
        schema. If this is not set, then OOPS will use the older schema
        unchanged. In most cases, this means that OOPS will use a historical
        version of itself rather than the current version. See the SCHEMA
        VERSIONS section of this document.

    "auto_initialize" / $ENV{OOPS_INIT}
        The DBMS must be initialized before it can be used: tables created,
        a few rows inserted, etc. This can be handled by calling
        "initial_setup()" or it can be handled automatically by "new()" if
        "auto_initialize" or $OOPS_AUTO_INIT is set.

    "default_synchronous" / $ENV{OOPS_SYNC}
        With SQLite, an additional parameter to "OOPS->new()" is recognized:
        "default_synchronous". Possible values are:

        FULL    Sync() all transactions to disk before returning.

        NORMAL  The default: sync() at critical moments only - protects
                against program failure, but not all power or OS failures.

        OFF     Don't sync() at all and go really fast.

PUBLIC OBJECT METHODS
    "->commit()"
        Writes any changed objects back to the database and commits the
        transaction. Currently only one commit() call is allowed. Do not
        access your persistent data after commit() -- it may work but this
        is not covered well in the regresssion suite.

    "->virtual_object(\%hash [,$new_value])"
        Queries [and sets] the load-virtual flag on a persistent hash.
        Hashes that load virtual will do separate queries for each key
        rather than load the entire hash. This is a good thing if your has
        has many keys. This flag takes effect the next time the hash is
        loaded. The value is a perl boolean.

        This may be handled automatically in the future.

    "->workaround27555($reference)"
        References to tied hash keys are buggy in all perls through 5.8.3.
        Use workaround27555($reference) to register your new tied hash key
        references so that they can be transformed into references that
        actually work correctly.

         $ref = \%hash{$key};
         $oops->workaround27555($ref);

        "workaround27555()" is harmless if called on other sorts of
        references so it is safe to use indescriminately. See
        <http://rt.perl.org/rt3/Ticket/Display.html?id=27555>.

        Alternatively, use "getref(%hash, $key)".

    "->dbh()"
        This returns the main DBI database handle used by OOPS. This
        function is provided for those who want to hand-write queries.
        Please note: no changes are written to the DBMS until commit().

    "->load_object($id)"
        This will load a persistent object by number. It returns the object
        or undef if the object doesn't exists. This function is provided for
        those who want to hand-write queries.

ERRATA, DEVELOPMENT STATUS, BUGS, AND BUG BOUNTY
    OOPS has been thoroughly tested. As this is written, there are no known
    users of OOPS. Don't let this stop you from using it! Someone has to go
    first. The regression suite is very well developed: there is twice as
    much code in the test suite as there is in the module itself. The suite
    does over 1.5million tests. On my fastest computer it takes over six
    hours to run. I have so much confidence in my test suite, I'm offering a
    bounty on bugs!

  Bug Bounty
    I'll pay (via paypal) US$5.00 to anyone who submits a new bug (with
    regression test) that is caused by a problem in my OOPS module. To
    qualify, bugs must be serious: they must cause data corruption, false
    results, or program failure. Bugs must not be listed here.

    I'll report the number of bounties paid in the CHANGELOG.

  Known fixable bugs in OOPS
    memory leaks
        OOPS currently has memory leaks. This may or may not matter to your
        application. The rate of leakage varies depending on which RDBMS is
        used. SQLite seems to have the most significant problems.

    Schema error with mysql
        There is an error in the schema that OOPS uses with mysql. This
        error does not cause any problems with the versions of mysql that
        OOPS has been tested with (4.0.16 and 4.0.18). This error will be
        fixed in the next version of OOPS.

    delayed DESTROY
        Additional references to the in-memory copies of persistent data are
        kept by OOPS. These extra references will prevent DESTORY methods
        from being called as soon as they otherwise would be.

    other magic
        Other perl magic attributes are not currently stored persistently.
        Many probably could be supported, but many could not. For example,
        taint does not work on tied hashes:
        <http://rt.perl.org/rt3/Ticket/Display.html?id=6758>.

    unreferenced blessed scalars
        When you bless a reference to a scalar value, the blessing is stored
        with the scalar, not the reference. The blessing remains even if
        there is no reference to the scalar. The following code prints
        "true".

         my $x = 'foobar';
         my $y = \$x;
         bless $y, 'baz';
         $y = 7;
         $y = \$x;
         my $z = ref($y);
         print "true\n" if $z eq 'baz';

        At the current time, OOPS does not store such blessings. OOPS does
        remember blessings when there is a reference.

    re-blessing the OOPS object
        If you re-bless the OOPS object, your data is likely to become
        inaccessable. I list this here so nobody claims a bounty for
        breaking OOPS in this way.

    DBD::Pg and ASCII NULL
        DBD::Pg does not easily support ASCII NULL. OOPS has only partial
        support for ASCII NULL with PostgreSQL: don't have ASCII NULL in
        your class names.

  Bugs in perl that effect OOPS
    references to hash keys
        Persistent hashes are implemented with tie. There are bugs with
        perl's implementation of references to tied hash keys. These bugs
        will be triggered in several situations: creating a reference to
        tied hash key that doesn't exist yet; deleting a key that has a
        reference tied to it; assigning through a reference to a key that
        has multiple references.

        All of the above can either be avoided or you can workaround them by
        either calling "workaround27555"($YOUR_REFERENCE) whenver you create
        a tied hash key reference or by using "getref(%hash, $key)" to
        create your reference.

        The perl bugs are documented in:
        <http://rt.perl.org/rt3/Ticket/Display.html?id=27555> and
        <http://rt.perl.org/rt3/Ticket/Display.html?id=29224>.

    "local" and tie
        "local(%some_tied_hash)" doesn't work right. Thus
        "local(%some_persistent_hash)" won't work right either:
        <http://rt.perl.org/rt3/Ticket/Display.html?id=6017>.

    "scalar(%hash)"
        "scalar(%hash)" support was added in perl 5.8.3 and does not exist
        in 5.8.2.

    tied arrays don't work right
        There are a couple of bugs with tied arrays that prevent OOPS from
        using them: <http://rt.perl.org/rt3/Ticket/Display.html?id=22570>
        and <http://rt.perl.org/rt3/Ticket/Display.html?id=22571>. This
        isn't a big deal unless you've got big arrays.

    SQLite and perl's malloc().
        If SQLite is used with a perl that has been compiled to use perl's
        "malloc()", it will report LOTS of "Bad free() ignored (PERL_CORE)"
        errors. It is not currently known if these errors are harmfull
        beyond generating lots of output to STDERR. The default perl
        configuration on FreeBSD uses perl's "malloc()".

SCHEMA VERSIONS
    As OOPS is developed, the schema that OOPS uses changes. At the current
    time, all versions of OOPS use forward and backwards compatable (though
    not identical) schemas.

    OOPS notices if the version of the schema in the DBMS is different than
    the version the code currently supports. When this happens there are two
    possibilities: either OOPS will upgrade the schema to the current
    version or it will use an older version of OOPS to access the data.

    This behavior allows the OOPS module to be upgraded without distrupting
    installed applications.

FUTURE DIRECTIONS
    OOPS isn't done. There are a bunch of things that I am considering
    adding to it. If any of these things is important to you, speak up so
    that I know there is interest...

    fix the bugs
        There are bugs listed in the DEVELOPMENT STATUS section that could
        be fixed. First up is fixing the memory leak.

    code cleanup and general performance enhancements
        The initial release of OOPS concentrated on correct behavior and
        other aspects of the module were somewhat ignored. The code could be
        cleaned up a bunch.

    better concurency
        With the transaction model of SERIALIABLE on mysql, there are
        serious performance issues surrounding access to the main hash.
        Change to a different transaction model or DBMS.

    perl-syntax sql query translator
         SELECT Employee WHERE $Employee->{salary} > 5000

        It's possible to translate perl-syntax queries into real SQL that
        can be used to query the object store.

    better grouping
        Objects are loaded in groups rather than individually. There is much
        room for improvement in choosing how groups are formed. This is
        largely undeveloped as yet.

    support for databases other than mysql
        PostgreSQL is the next likely supported backend.

    caching
        Many possibilities. A cache-invalidation daemon to note when objects
        have changed. Re-verification of touched data from the database.
        Ability to call commit() more than once.

    weak references
        Support for persistent weak references is possible.

    external references to objects
        Currently objects are reference counted internally. You must have a
        reference to something from an already existing object for it to
        continue to exist.

    contracts
        OOPS has to do a lot of scanning of objects to see if they've
        changed. Explicit notification of changes would improve performance.

        OOPS could call functions before saving and after loading to
        transform objects for a better or cleaner on-disk representation.

    support for 'base' & accessor methods
        This isn't something that I care about but maybe someone else does?

    schema enforcement
        Allow explicit schemas to be defined. Do not save objects that don't
        conform.

    RDBMS -> object mapping
        Map existing RDBMS schemas into objects.

    data viewer
        Viewing large datasets of deep and cross-linked data is difficult.
        Perhaps a CGI-based or Tk-based data navigator would help.

    support for tied data structures
        It is possible to support storing tied data. The tied object is what
        would need to be persistent. This would only work on some kinds of
        ties.

    garbage collect circular references
        Like perl, OOPS uses reference counting to know when to delete an
        object. Unlike perl, restarting your program does not clear out the
        circular references.

    support for other base types.
        Right now, just HASH, SCALAR, REF, and ARRAY are supported. Regular
        expressions, file handles, I don't know it's possible to support
        code references.

    on-line data migration
        By doing double updates, OOPS could support live migration from one
        DBMS to another.

WRITING SQL QUERIES BY HAND
    If you want to query your data, then until a translator is written, your
    only choice for making queries is to write them by hand. Using your data
    does not require a query: anything you've got a reference to will be
    loaded as you access it. Queries are for performing searches that don't
    have a perl-object index.

    Each perl HASH, REF/SCALAR, or ARRAY has a row in the "object" table and
    multiple rows in the "attribute" and "big" tables.

    Here are the columns you'll care about:

    object      There is one row per perl object.

                id        The object id.

                class     The blessed class name (limited to 255
                          characters).

                otype     The type of object:

                          'H'  A HASH.

                          'A'  An ARRAY.

                          'S'  A SCALAR or REF.

    attribute   This is a table of key/value pairs. The keys corrospond to
                perl hash keys and perl array indexes. The values corrospond
                to perl hash values and array element values.

                id        The object id.

                pkey      The hash key or array index.

                pval      he hash value or array value. Limited to 255
                          characters.

                ptype     Flags the type of the value. Possible values are:

                          '0'  A normal value. Numeric or string.

                          'B'  An big value. "pval" will be a copy of the
                               start of the value for the first N
                               characters. The end of "pval" will be a MD5
                               checksum of the full big value.

                          'R'  A reference to another object.

    big         This is a table of values that were too large for the normal
                columns. Even with databases that support wide columns, a
                separate big table is used so that you don't load large
                scalars unless you actually need the value.

                id        The object id.

                pkey      The hash key or array index.

                pval      The hash value or array value. Limited to whatever
                          the underlying database will support as it's
                          largest blob.

                fragno    Blob fragment number. This column only exists with
                          SQLite. SQLite has a smallish maximum row size and
                          so big values must be split into multiple rows.

    REFs are are special. There are several types of REFs: references to
    scalar values; references to objects; secondary references to scalar
    values; references to scalar values that are part of another object
    (references to hash keys and references to array elements).

    The representation of references is designed so that you don't need to
    care what sort of REF it is when you're doing a query.

    The basic REF is a ref to a value inside another object. An example:

     OBJECT TABLE
     id             class           otype

     1              OOPS::NamedObj  H
     383            SCALAR          R
     384            SCALAR          R
     385            SCALAR          R
     386            REF             R
     400            HASH            H
     500            ARRAY           A

     ATTRIBUTE TABLE
     id             pkey            pval            ptype

     1              A500            500             R
     1              H400            400             R
     1              R383            383             R
     1              R384            384             R
     1              R385            385             R
     1              R386            386             R

     383            400             'a-key'         0

     384            384             'nopkey'        0
     384            'nokey'         'a-value'       0

     385            384             'nopkey'        0

     386            386             'nopkey'        0
     386            'nopkey'        500             R

     400            'a-key'         'a-value        0
     400            'another-key'   'another-value' 0
     400            'A500'          500             R

     500            0               'a-value'       0
     500            1               'another-value' 0

    HASH 1 is %$oops.

    REF 383 is a reference to the key 'a-key' in object #400 (a HASH).

    REF 384 is a ref to scalar. It uses two rows to make writing queries
    easier.

    REF 385 is a duplicate reference to a scalar value. It duplicates REF
    384. In behavior, these two REFs should be identical even though they
    are represented differently in the database.

    REF 386 is a ref to an object: #500 (an ARRAY).

    HASH 400 is a normal hash.

    ARRAY 500 is a normal hash.

    This example data is what you would end up with after running code like:

     my $oops = new OOPS 
            dbi_dsn => 'DBI:mysql:database=MY-DATABASE-NAME;host=localhost', 
            username => 'MY-USERNAME', 
            password => 'MY-PASSWORD';

     $oops->{A500} = [ 'a-value', 'another-value' ];

     $oops->{H400} = { 
            'a-key' => 'a-value',
            'another-key' => 'another-value',
            'A500' => $oops->{A500},
     };

     $oops->{R383} = \$oops->{H400}{'a-key'};
     $oops->workaround27555($oops->{R383});

     $oops->{R384} = \'a-value';

     $oops->{R385} = $oops->{R384};

     $oops->{R386} = \$oops->{A500};

     $oops->commit;

    SQL queries require a bunch of joins to link data structures together.
    Here are some examples.

    "SELECT Foobar WHERE $Foobar->{xyz} = 'abc'"
         SELECT object
         FROM   object, attribute
         WHERE  object.class = 'Foobar'
         AND    object.id = attribute.id
         AND    attribute.pkey = 'xyz'
         AND    attribute.pval = 'abc'
         AND    attribute.ptype = '0'

    "SELECT Foobar WHERE ${$Foobar->{xyz}} = 'abc'"
        This example should show why an automatic translator would be a good
        idea...

         SELECT ohash.object
         FROM   object AS ohash,
                attribute AS ahash,
                object AS oref,
                attribute AS aref,
                attribute AS target
         WHERE  ohash.class = 'Foobar'
         AND    ohash.otype = 'H'
         AND    ahash.id = ohash.id
         AND    ahash.pkey = 'xyz'
         AND    ahash.ptype = 'R'
         AND    oref.id = ahash.pval
         AND    oref.otype = 'S'                # this is the outer ref
         AND    oref.id = aref.id               
         AND    aref.pval = target.pkey         # here's the reference indirection
         AND    target.pval = 'abc'
         AND    target.ptype = '0'

    If you construct a query like these examples that return object id's,
    then use "$object = $oops->load_object($id)" to load them into memory.

    I reccomend that hand-written queries be read-only as there are
    additional columns that must be kept consistent. For example, the object
    table includes a reference count column to handle garbage collection of
    the persistent data.

RUNNING THE REGRESSION TEST SUITE
    The regression test suite empties and re-creates the persistent store
    over and over again. To prevent the accidental erasure of production
    data, all of the tests require a special environment variable to be set
    $OOPSTEST_DSN. This variable replaces the normal $DBI_DSN or $OOPS_DSN.
    Correspondingly there is a $OOPSTEST_USER, $OOPSTEST_PASS, and
    $OOPSTEST_PREFIX.

    Set these variables to something different than what you use for your
    production data!

    Most of the tests take a long time to run and are disabled by default.
    If you can run the full suite in less than six hours please tell me
    about your configuration.

    Beware mysql logging. On Debian unstable, the default configuration for
    mysql logs every SQL statement. Running the test suite to completion
    will generate several gigabytes of log file. Running out of disk space
    will cause the tests to fail. On DragonflyBSD (FreeBSD 4.9) the default
    mysql configuration includes making replication master logs.

THE COMPETITION
    There are a number of other modules that make perl objects persistent.
    For each module, I'll provde a sentence or two that explains why I wrote
    OOPS rather than using that one.

    OOPS does not require any schema defintion. Most other persistence
    package requires schema definition and allows the nature of the
    underlying RDBMS to show in both the kinds of objects you can use and
    how you use them. OOPS tries to let you work with perl objects without
    concerning yourself about how they'll be stored persistently.

    Persistence::Object::Postgres
        Uses Data::Dumper to serialize objects into BLOBs. Cannot handle
        cross-linking. No query support.

    pop Only supports Sybase. Uses a fixed schema defined in XML.

    SPOPS
        Objects are HASHes, values are scalars only. No nesting.

    DBIx::RecordSet
        Thin layer on top of DBI. Removes much pain from using DBI. Does not
        change basic relationship to data -- still relational.

    Object::Transaction
        Uses Storable to preserve objects in files. No query support. No
        cross-linking. Lightweight. Transactions.

    Class::DBI
        Must define schema and relationships to RDBMS. Accessor methods.
        Doesn't support many2many. Close tie to underlying RDBMS.

    Alzabo
        Must define schema and relationships as relates to RDBMS. Close tie
        to underlying RDBMS. Can help create RDBMS schema. Complex. Can
        cache database rows.

    Tangram
        Must define schmea. RDBMS schema is created from perl schema.

    DBM::Deep
        DBM::Deep handles deep and cross-linked data. It does not have
        support for queries or transactions.

    Pixie
        Less trasnparent. Requires user to remember a cookie to retreive
        objects. Cannot handle cross-linking of objects? No query support?

    MLDBM / Tie::MLDBM / MLDBM::Easy / MLDBM::Sync
        Uses Data::Dumper or Storable or such to serialize objects and store
        them as strings. Lock granularity is the whole file. No query
        support. Only one (or two with MLDBM::Easy) level of access is
        automated.

    ObjStore
        Proprietary. Not as transparent as OOPS.

    see also
        <http://poop.sourceforge.net/> has an overview of options.

LICENSE
    Copyright(C) 2004 David Muir Sharnoff <muir@idiom.com>

    All rights reserved except as granted by a license or contract. This
    module is licensed both commercially and as free software. The free
    software license follows.

    The Play-Fair-To-Play Software License

    Preamble.

    I wish to let people use this software for free as long as they don't
    abuse the free software community or Internet in particular ways. I want
    all derrivitives of this software to be free. I want to be able to
    license this software commercially.

    This license is not compatabile with the GPL. It was not written by a
    lawyer and the exact wording will probably change when a legal expert
    examines it. This license applies only to this version of this program:
    other versions may exist with other licenses.

    I.     This software may be used, copied, modified, and redistributed
           with no fee paid to the copyright holder (original licensor) so
           long as all the parts of this license are honored. Should any
           part be deemed invalid or otherwise contradicted, the entire
           license is invalidated and no right to use, copy, modify or
           redistribute exists.

    II.    Any redistribution of this software must retain this license. Any
           redistribution of this software must include it's original
           (source code) form. When redistributing this software, you may
           not add additional licensing terms that would prevent the new
           recipient of the software from using, copying, modifying, or
           redistributing this software themselves under the terms of this
           license.

    III.   This software is licensed free of charge. To the extent permitted
           by applicable law, there is no warranty for the software. The
           software "as is" without warranty of any kind, either expressed
           or implied, including, but not limited to, the implied warranties
           of merchantability and fitness for a particular purpose. The
           entire risk as to the quality and performance of the software is
           with you. should the software prove defective, you assume the
           cost of all necessary servicing, repair or correction.

    IV.    All useful modifications to this software must be licensed back
           to the original licensor such that the original licensor can
           relicense the modifications to third parties under terms of their
           own choosing with no restrictions whatsoever and no need to even
           acknowledge the origin of such modifications. There are two easy
           ways to accomplish this: put all such modifications into the
           public domain or assign the copyright to the original licensor.

    V.     Entities that have willfully violated an open source license are
           excluded from this license. For example, SCO for violating the
           GPL.

    VI.    Entities that have attempted to enforce a software patent (except
           as a defensive move like IBM's counter to SCO) are excluded from
           this license. For example, Unisys for enforcing a patent on GIF
           image creation.

    VII.   Entities that engage in or encourage sending SPAM emails are
           excluded from this license.

    VIII.  Entities that are or have challenged this license in court are
           excluded from this license.

    IX.    The following entities are excluded from this license: Network
           Solutions; Verisign; ICANN; and Internet-Journals.

    X.     Entities that share a 5% or more common ownership interest with
           an excluded entity are also excluded from this license.

    XI.    Entities that are excluded from this license are forbidden to
           use, copy, modify, or redistribute the covered software.

    This is version 1.1 of the Play-Fair-To-Play Sofware License.