Class::PObject - Simple framework for programming persistent objects


After loading Class::PObject with use, we can declare a class:

    pobject Person => {
        columns     => ['id', 'name', 'email'],
        datasource  => './data'

We can also declare the class in its own .pm file:

    package Person;
    use Class::PObject;
    pobject {
        columns     => ['id', 'name', 'email'],
        datasource  => './data'

We can now create an instance of above Person, fill it with data, and save it:

    $person = new Person();
    $new_id = $person->save()

We can access the saved Person later, make necessary changes and save back:

    $person = Person->load($new_id);
    $person->name('Sherzod Ruzmetov (The Geek)');

We can load multiple objects as well:

    @people = Person->load();
    for $person ( @people ) {
        printf("[%02d] %s <%s>\n", $person->id, $person->name, $person->email)

or we can load all the objects based on some criteria and sort the list by column name in descending order, and limit the results to only the first 3 objects:

    @people = Person->load(
                    {name => "Sherzod"},
                    {sort => "name", direction => "desc", limit=>3});

We can also seek into a specific point of the result set:

    @people = Person->load(undef, {offset=>10, limit=>10});


Class::PObject is a simple class framework for programming persistent objects in Perl. Such objects can store themselves into disk, and restore themselves from disk.


This section has been moved to Class::PObject::Philosophy.


Programming style of Class::PObject is very similar to that of Class::Struct. Instead of exporting struct(), however, Class::PObject exports pobject() function. Another visual difference is the way you declare classes. In Class::PObject, each property of the class is represented as a column.

Suppose, you have a database called "person" with the following records:

    # person
    | id  | name           | email                  |
    | 217 | Sherzod        | sherzodr[AT]   |


Let's declare a class to represent the above data as a Persistent Object (pobject for short). To do this, we first load Class::PObject with use and declare a class with pobject() construct:

    use Class::PObject;
    pobject Person => {
        columns => ["id", "name", "email"]

Above construct is declaring a class representing a Person object. Person object has 3 attributes - id, name and email.

Above is called in-line declaration, because you are creating an inline class - the one that doesn't need to be in its own .pm file. You can declare inline classes almost anywhere inside your Perl code.

In-line declarations are not very useful, because you cannot access them separately from within another application without having to re-declare identical class several times in each of your programs.

Another, more recommended way of declaring classes is in their own .pm files. For example, inside a file we say:

    # lib/
    package Person;
    use Class::PObject;
    pobject Person => {
        columns => ["id", "name", "email"]


Now, from any other application all we need to do is to load, and access all the nifty things it has to offer:

    # inside our app.cgi, for example:
    use Person;


From the above class declaration you may be wondering, how does it now how and where the object data are stored? The fact is, it doesn't. That's why by default it stores your objects in your system's temporary folder - wherever it may be - using default file driver. To control this behavior you can define driver and/or datasource attributes in addition to above columns:

    pobject Person => {
        columns     => ["id", "name", "email"],
        datasource  => './data'

Now, it's still using the default file driver, but storing the objects in your custom, ./data folder.

NOTE: set_datasource() and set_driver() methods can also be used to set datasource and object drivers respectively. This is useful when datasource will not be known until later, in which case it can be set either from within pobject_init() method, or called directly as object method.

You could've also chosen to store your objects in a DBM file, or in mysql tables. That's where you will need to define your driver attribute.

To store them in DBM files using DB_File:

    pobject Person => {
        columns     => ["id", "name", "email"],
        driver      => 'db_file',
        datasource  => './data'

To store them in comma separated text files using DBD::CSV:

    pobject Person => {
        columns => ["id", "name", "email"],
        driver  => 'csv',
        datasource {
            Dir => './data'

Or, to store them in a mysql database using DBD::mysql:

    pobject Person => {
        columns => ["id", "name", "email"],
        driver  => 'mysql',
        datasource => {
            DSN      => "dbi:mysql:people",
            User     => "sherzodr",
            Password => "secret"

So forth. For more options you should refer to respective driver manuals. Following object drivers are available in this distribution:


Default driver. Stores objects of the same type in a dedicated folder. Individual objects are stored in their own files.


Stores objects of the same type in a single file. Each object is separated by a new line. Object attributes are separated with comma. These files can be easily loaded to your spreadsheet applications, such as Excel, Access or even RDBMS tables.


Stores objects of the same type in a single file using DB_File.


SQLite driver. Stores objects using DBD::SQLite. Depending on the datasource attribute, all the objects can be stored in a single file, or on different files.


Stores objects in MySQL tables. Objects of the same type are stored in a single table. Individual rows of these tables represent individuals objects. Each column type corresponds to object attributes.


After having the above Person class declared, we can now create an instance of a new Person with the following syntax:

    $person = new Person();

Now what we need is to fill in the $person's attributes, and save it into disk

    $person->name("Sherzod Ruzmetov");

As soon as you call save() method of the $person, all the records will be saved into the disk.

Notice, we didn't give any value for the id column. Underlying object drivers automatically generate a new ID for newly created objects and save() method returns this ID for you.

If you assign a value to id, you better make sure that ID doesn't already exist. If it does, the old object with that ID will be replaced with new object. To be safe, just don't bother defining any values for your ID columns, unless you have a really good reason.

Sometimes, if you have objects with few attributes, you may choose both to create the Person and to assign its attributes at the same time. You can do so by passing your column values while creating the new person:

    $person = new Person(name=>"Sherzod Ruzmetov", email=>"sherzodr[AT]");


PObjects support load() class method, which allows you to retrieve your objects from the disk. You can retrieve objects in many ways. The easiest, and the most efficient way of loading an object from the disk is by its id:

    $person = Person->load(217);

Now, assuming $person could be retrieved successfully, we can access the attributes of the object like so:

    printf( "Hello %s!\n", $person->name )

Notice, we are using the same method names to access them as the ones we used to assign values with, but this time with no arguments.

Above, instead of displaying $person's name, we could've also edited his name and save it back:

    $person->name("Sherzod The Geek");


Sometimes you may choose to load multiple objects. Using the same load() method, we could assign all the result set into an array:

    @people = Person->load();

Above code is retrieving all the objects from the database, and assigning them to @people - array. Each element of @people is a $person object. We can list all of them with the following syntax:

    for my $person ( @people ) {
        printf("[%d] - %s <%s>\n", $person->id, $person->name, $person->email)

Notice two different contexts load() was used in. If you call load() in scalar context, regardless of the number of matching objects, you will always retrieve the first object in the data set. For added efficiency, Class::PObject will add limit=>1 argument to load() even if it's missing, or exists with a different value. We will talk more about attributes very soon.

If you called load() in array context, you will always receive an array of objects, even if result set consists of a single object.


Sometimes you just want to load objects matching a specific criteria, say, you want all the people whose name are John. You can achieve this by passing a hash ref as the first argument to load():

    @johns = Person->load({name=>"John"});

Sets of key/value pairs passed to load() as the first argument are called terms.

You can also apply post-result filtering to your list, such as sorting by a specific column in a specific order, and limit the list to n number of objects and start the listing at object n of the result set. All these attributes can be passed as the second argument to load() in the form of a has href and are called arguments :

    @people = Person->load(undef, {sort=>'name', direction=>'desc', limit=>100});

Above @people holds 100 $person objects, all sorted by name in descending order. We could use both terms and arguments at the same time and in any combination.

Arguments are the second set of key/value pairs passed to load(). Some drivers may look at this set as post-result filters.


Defines which column the list should be sorted by. Value of this attribute can be any valid column name this particular class has.


Denotes direction of the sort. Possible values are asc meaning ascending sort, and desc, meaning descending sort. If sort is defined, but no direction is available, asc is implied.


Denotes the number of objects to be returned.


Denotes the offset of the result set to be returned. It can be combined with limit to retrieve a sub-set of the result set, in which case, result set starting at offset value and up to limit number will be returned to the caller.


load() may be all you need most of the time. If your objects are of larger size, or if you need to operate on large amount of objects, your program may not have enough memory to hold them all, because load() tends to load all the matching objects to the memory.

If this is your concern, you are better off using fetch() method instead. Syntax of fetch() is almost identical to load(), with a single exception: it doesn't accept object id as the first argument. You can either use it without any arguments, or with any combination of \%terms and \%args as needed, just like with load().

Another important difference is, it does not return any objects. It's return value is an iterator object. Iterator helps you to iterate through large data sets by loading them one at a time inside a while-loop:

    $result = Person->fetch();
    while ( my $person = $result->next ) {
    # or
    $result = Person->fetch({name=>"John"}, {limit=>100});
    while ( my $person = $result->next ) {

For the list of methods available for iterator object refer to its manual.


Counting objects is very frequent task in many projects. You want to be able to display how many people are in your database in total, or how many "John"s are there.

You can of course do it with a syntax similar to:

    @all = People->load();
    $count = scalar @all;

This also means you will be loading all the objects to memory at the same time.

Even if we could've done it using an iterator, as discussed earlier, some database engines may provide more optimized way of retrieving this information without having to load() any objects, by consulting available meta information. That's where count() class method comes in:

    $count = Person->count();

count() can accept \%terms as the first argument, just like above load() does. Using \%terms you can define conditions:

    $njohns = Person->count({name=>"John"});


PObjects support remove() and remove_all() methods. remove() is an object method. It is used only to remove one object at a time.

To remove a person with id 217, we first need to create an object of that Person, and only then call remove() method:

    $person = Person->load(217);

remove_all() is a static class method, and is used for removing all the objects from the disk:


remove_all() can also be used for removing objects selectively without having to load them first. To do this, you can pass \%terms as the first argument to remove_all(). These \%terms are the same as the ones we used for load(), fetch() and count():


Notice, if we wanted to, we still could've used a code similar to the following to remove all the objects:

    $result = Person->fetch();
    while ( $person = $result->next ) {

However, this will require loading the object to the memory one at a time, and then removing one at a time. Most object drivers may offer a better, efficient way of removing objects from the disk without having to load() them. That's why you should rely on remove_all().


Class::PObject lets you define types for your columns, also known as type-maps. First off, what for?

Type-maps can be looked at as properties of each column. If you are familiar with RDBMS, you are already familiar with them. They read as CHAR, VARCHAR, INTEGER, TEXT, BLOB etc.

Column types can be used for declaring a property for a column, defining a constraint or input-output filtering of columns. This same feature can also be used for defining object relationships, but let's talk about it later.

You can define types for your columns by tmap pobject attribute (tmap stands for type map):

    pobject User => {
        columns => ['id', 'login', 'psswd', 'name'],
        tmap    => {
            id      => 'INTEGER',
            login   => 'CHAR(18)',
            psswd   => 'MD5',
            name    => 'VARCHAR(40)'

Above class declaration is defining a User class, with 4 columns, and defining the type of each column, such as id as an INTEGER column, login as a VARCHAR column and so forth.

You normally never have to declare any column types. If you don't, id column will always default to INTEGER, and all other columns will always default to VARCHAR(255). So you should declare your column types only if you don't agree with defaults.

In above example I could've chosen to say:

    pobject User => {
        columns => ['id', 'login', 'psswd', 'name'],
        tmap    => {
            psswd   => 'MD5'

Notice, I am accepting default types for all the columns except for psswd column, which should be encrypted using MD5 algorithm before being stored into disk.

Of course, if I didn't care about security as much, I could've chosen not to define type-map for psswd column either.

As of this release, available built-in column types are:


Stands for INTEGER type. This type is handled through Class::PObject::Type::INTEGER. Currently Class::PObject doesn't enforce any constraints, but this may (and most likely will) change in the future releases, so be cautious.


Stands for CHARacter type with length n. This type is handled through Class::PObject::Type::CHAR class, which currently doesn't enforce any constraints to the length of the column. This may change in future releases, so be cautious.


Stands for VARIABLE CHARacter n characters long. Highest allowed value for n is usually 255 characters, but not yet enforced. This may change in future releases, so be cautious. Handled through Class::PObject::Type::VARCHAR.


Stands for TEXT column, which normally denotes values longer than 255 characters. As of this release TEXT columns are handled the same way as CHAR and VARCHAR columns. This may change in future releases, so be cautious.


Stands for ENCRYPTed. It denotes a value that need to be encrypted using UNIX's crypt() before being stored into disk. Smart overloading allows to do something like the following:

    $user = new User();
    print $user->psswd; # prints 'ZG9nqo.9bPjGA'
    if ( $user->psswd eq "marley01") {
        print "Good!\n";
    } else {
        print "Bad!\n";


The above example will print "Good". Notice, that even if psswd method returns encrypted password ('ZG9nqo.9bPjGA'), we could still compare it with an un-encrypted string using Perl's built-in eq operator to see if "marley01" would really equal to 'ZG9nqo.9bPjGA' if encrypted.


Stands for MD5 - Message Digest algorithm. It denotes value that need to be encrypted using MD5 algorithm before being stored into disk. Smart overloading allows it to be used in similar way as ENCRYPT column type.

Up to this point you noticed that objects' accessor methods have been returning strings, right? Wrong! They have been returning objects of their appropriate types (CHAR, VARCHAR, ENCRYPT and etc.). But you didn't notice them and kept treating them as strings because of smart operator overloading feature of those objects.

This means, if at any time in your program you want to discover the type of a specific column, you could simply use Perl's built-in ref() function:

    print ref( $user->psswd), "\n"; # <-- prints ENCRYPT
    print ref( $user->name ), "\n"; # <-- print VARCHAR


Objects returned by calling accessor methods support several built-in methods, also known as member functions. These member functions are usually meant to manipulate the value of the column before returning to the caller, but are not meant to modify the actual value of the column. So you can fearlessly call any of the member functions without modifying their values in the object.

Although Typemap Specifications and respective type classes cover all the necessary details about each of these member functions, in this section we'll just cover uc as an example, which returns the upper-cased version of the string:

    $user = User->load( $user_id );
    printf( "Hello %s!\n", $user->name()->uc() );

Assuming value of $user->name() was Sherzod, $user->name()->uc() returns SHERZOD by uppercasing all the characters of the word.

After $user->name()->uc() is called, $user->name() keeps returning Sherzod. As we mentioned early initial value of the column would not be altered.

See Typemap Specifications and respective type classes for the list of all the available member functions available for all and specific column types.


In some cases accessor methods are not all the methods your class may ever need. It may need some other behaviors. In cases like these, you can extend your class with custom methods.

For example, assume you have a User object, which needs to be authenticated before they can access certain parts of the web site. It may be a good idea to add authenticate() method into your User class, which either returns a User object if he/she is logged in properly, or returns undef, meaning the user isn't logged in yet.

To do this we can simply define additional method, authenticate() inside our .pm file. Consider the following example:

    package User;
    use Class::PObject;

    pobject {
        columns     => ['id', 'login', 'psswd', 'email'],
        datasource  => './data'

    sub authenticate {
        my $class = shift;
        my ($cgi, $session) = @_;

        unless ( $cgi->isa('CGI') && $session->isa('CGI::Session') ) {
            croak "authenticate(): usage error"

        # if user's session indicates he/she is logged in, we load the User object
        # by that id and return it
        if ( my $user_id = $session->param('_logged_in') ) {
            return $class->load( $user_id )

        # if we come this far, the user is not logged in yet, but still
        # might've submitted the login form:
        my $login     = $cgi->param('login')    or return 0;
        my $password  = $cgi->param('password') or return 0;

        # if we come this far, both 'login' and 'password' fields were submitted
        # in the form. So we try to load() the matching object:
        my $user = $class->load({login=>$login, psswd=>$password}) or return undef;

        # we store the user's Id in our session parameter, and return the user
        # object
        $session->param('_logged_in', $user->id);
        return $user


Now, we can check if the user is logged into our web site with the following code:

    use User;
    my $user = User->authenticate($cgi, $session);
    unless ( $user ) {
        die "You need to login to the web site before you can access this page!"
    printf "<h2>Hello %s</h2>", $user->login;

Notice, we're passing CGI and CGI::Session objects to authenticate(). You can do it differently depending on the tools you're using.


Some available tools prefer calling this feature as HAS-A relationship, but Class::PObject prefers the term type-mapping. For the sake of familiarity, let's look at it as HAS-A relationship for now.

HAS-A relationship says that a value in a specific column is an index of another object. In RDBMS this particular column is known as foreighn key, because it holds the primary key of another (foreign) table.

You can define this relationship the same way as we did column types using tmap pobject attribute.

Suppose, we have two classes, Article and Author:

    pobject Author => {
        columns        => ['id', 'name', 'email']

    pobject Article => {
        columns        => ['id', 'author', 'title', 'body'],
        tmap        => {
            author        => 'Author'

Notice, in the above Article class, we're defining its author column to be of Author type. This type should match the name of the existing class.

Now, we can create a new article with the following syntax:

    $author = Author->load({name=>'Sherzod Ruzmetov'});

    $article = new Article();
    $article->title("Class::PObject now supports HAS-A relationships");
    $article->body("***body of the article***");
    $article->author( $author );

Notice, we passed $author object to article's author() method. Class::PObject retrieves the ID of the author and stores it into Article's author field.

When you choose to access author field of the article, Class::PObject will automatically load respective Author object from database.

    $article = Article->load(32);
    my $author = $article->author;
    printf "Author: %s\n", $author->name;

We could've also passed $author as part of the \%terms while loading articles:

    my $author = Author->load({name=>"Sherzod Ruzmetov"});
    @my_articles = Article->load({author=>$author});

As you noticed, defining HAS-A relationship is the same as defining built-in column types, it's because the engine behind both column types and column relationships are handled through the same API.


PObjects try never to die(), and lets the programer to decide what to do on failure, (unless of course, you insult it with wrong syntax).

Methods that may fail are the ones to do with disk access, namely, save(), load(), fetch(), count(), remove() and remove_all(). So it's advised you check these methods' return values before you assume any success. If an error occurs, the above methods return undef. More verbose error message will be accessible through errstr() static class method. In addition, save() method should always return the object id on success:

    my $new_id = $person->save();
    unless ( defined $new_id ) {
        die "save() failed: " . $person->errstr
    Person->remove_all() or die "remove_all() failed: " . Person->errstr;


In addition to the above described methods, pobjects support the following few useful ones:

  • columns() - returns hash-reference to all the columns of the object. Keys of the hash hold column names, and their values hold respective column values. All the objects will be stringified:

        my $columns = $person->columns();
        while ( my ($k, $v) = each %$columns ) {
            printf "%s => %s\n", $k, $v
  • dump() - dumps the object as a chunk of visually formatted data structure using standard Data::Dumper. This method is mainly useful for debugging.

  • errstr() - class method. Returns the error message from last I/O operations, if any. This error message is also available through $CLASS::errstr global variable:

        $person = new Person() or die Person->errstr;
        # or
        $person->save() or $person->errstr;
        # or
        $person->save() or  die $Person::errstr;
  • __props() - returns class properties. Class properties are usually whatever was passed to pobject() as a has href, either as the first argument or the second. This information is usually helpful for driver authors, and for those who want to access certain information about this class.

  • __driver() - returns either already available driver object, or creates a new object and returns it. Although not recommended, you can use this driver object to access driver's low-level functionality, as long as you know what you are doing. For available driver methods consult with specific driver manual, or contact the vendor.

  • drop_datasource() - used when you want to drop storage device allocated for storing this particular object. Its meaning varies depending on the pobject driver being used. file driver may remove the entire directory of objects, whereas mysql driver may drop the table.


Class::PObject is not a tool for solving all the problems of the Universe. It may be its success rather than its failure. I still want it to do certain things, for I believe they are right down its alley.

  • Multiple object joins

    Currently load() and fetch() methods are used for querying a single object. It would be ideal if we could select multiple objects at a time through some join syntax.

  • Column Indexes

    While RDBMS drivers do not need this, db_file, csv and file drivers will not be suitable for production environment until they do.

    Column indexes are needed to retrieve more complex queries much faster, without having to perform linear search. With aforementioned drivers (especially file and csv) once number of records increases, querying objects gets extremely processor intensive if more complex load() and fetch() methods are used.

  • SET Column types

    Currently column types can be configured to be of HAS-A relationship. It would be great if we could specify HAS-MANY relationships. Following syntax is still to be implemented:

        pobject Article => {
            columns => ['id', 'title', 'authors'],
            tmap    => {
                authors => ['Author']

    Notice, authors field is set to be able to hold an array of Author objects.


Pobjects try to cache the driver object for more extended periods than pobject's scope permits them to. So a global destructor should be applied to prevent unfavorable behaviors, especially under persistent environments, such as mod_perl

Global variables that may need to be cleaned up are:


Where $drivername is the name of the driver used. If more than one driver is used in your project, more of these variables may exist. This variable holds particular driver object.


Holds the properties for this particular PObject named $ClassName. For example, if you created a pobject called Person, then it's properties are stored in global variable $Person::props.

For example, if our objects were using just a mysql driver, in our main application we could've done something like:

    END {
        $Class::PObject::Driver::mysql::__O = undef;


You can learn more about pobject driver specs in the following manuals:


Base pobject driver specification. All the drivers are expected to inherit from this class.


Base pobject driver for all the DBI-related drivers. This inherits from Class::PObject::Driver, but re-defines certain methods with those more relevant to DBI drivers.


Base pobject driver for all the DBM-related drivers. It inherits from Class::PObject::Driver, and re-defines certain methods with those more relevant to DBM


You can learn more about column type specs from Class::PObject::Type, which is also a base class of all the types.



On platforms where Perl was built ithreads enabled, several tests during make test may fail. Although these particular failures are found to be related to a bug in Test::Builder, it's still not clear whether Class::PObject is fully compatible with ithreads or not. Volunteers are welcome to help us find out.

For more details of these problems refer to

SEE ALSO, although little dated, provides brief overview of available tools for programming Persistent Objects in Perl. As of this writing, Class::PObject is still not mentioned.


Sherzod B. Ruzmetov, <>,


Copyright (c) 2003 Sherzod B. Ruzmetov. All rights reserved.

This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

2005/02/20 18:05:00