Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
Fetching contributors…

Cannot retrieve contributors at this time

705 lines (502 sloc) 21.448 kb
#!/usr/bin/perl
use strict;
use lib '.';
use Tangram;
use Tangram::Springfield;
use Getopt::Std;
use DBI;
do {
my %opt;
getopts('pxtw', \%opt);
$Tangram::TRACE = \*STDOUT if exists $opt{t};
my @cp = qw( dbi:Pg:dbname=tour tangram tangram );
my $cp = join(', ', map { "'$_'" } @cp);
my ( $schema, $dbh, $storage, @kids, $marge, $homer, $homer_id,
$ned_id, @sisters_id, $ned, @sisters, @pairs, $patty, $selma, $burns
);
if ($opt{w}) {
$opt{p} = 1;
open STDOUT, '>Tangram/Tour.pod';
}
my $tour = join '', <DATA>;
if (exists $opt{p})
{
$tour =~ s/{{\n//gm;
$tour =~ s/}}\n//gm;
print $tour;
exit;
}
$tour =~ s[\@cp][$cp]g;
if ($opt{x}) {
system 'dropdb -q -U postgres tour 2>/dev/null';
system 'createdb -q -U postgres tour';
while ($tour =~ m[ {{ (.*?) }} ]sgx) {
my $chunk = $1;
my $show = $chunk;
my $line = 1;
$show =~ s/^/ sprintf "%03d: ", $line++/gem;
print "executing:\n\n$show";
eval "do { use strict; $chunk }";
die $@ if $@;
print "\n", '-' x 40, "\n\n";
}
%$homer = ();
%$marge = ();
}
};
print "finished!\n";
__END__
=head1 NAME
Tangram - Guided Tour
=head1 INTRODUCTION
In this tour, we add persistence to a simple Person design.
A Person is either a NaturalPerson or a LegalPerson. Persons (in
general) have a collection of addresses.
An address consists in a type (a string) and a city (also a string).
NaturalPerson - a subclass of Person - represents persons of flesh and
blood. NaturalPersons have a name and a firstName (both strings) and
an age (an integer). NaturalPersons sometimes have a partner (another
NaturalPerson) and even children (a collection of NaturalPersons).
LegalPerson - another subclass of Person - represents companies and
other entities that the law regards as 'persons'. A LegalPerson has a
name (a string) and a manager (a NaturalPerson).
All this is expressed in the following UML diagram:
+---------------------+ +--------------+
| Person | | Address |
| { abstract } |1<>-->-*|--------------|
|---------------------| | kind: string |
+---------------------+ | city: string |
| +--------------+
|
+--------------A--------------+
| |
+-------------------+ +---------------+
+--*| NaturalPerson | | LegalPerson |
| |-------------------|manager |---------------|
V | firstName: string |1---<-----1| name: string |
| | name: string | +---------------+
+--*| age: integer |
children +-------------------+
1 1
| partner
| |
+--->---+
B<Note that Tangram does I<not> create the corresponding Perl
packages!>. That's up to the user. However, to facilitate
experimentation, Tangram comes with a module that implements the
necessary classes. For more information see L<Tangram::Springfield>.
Before we can actually store objects we must complete two steps:
=over 4
=item 1
Create a Schema
=item 2
Create a database
=back
=head2 Creating a Schema
A Schema object contains information about the persistent
aspects of a system of classes.
It also gives a degree of control over the way Tangram performs the
object-relational mapping, but in this tour we will use all the defaults.
Here is the Schema for Springfield:
{{
$schema = Tangram::Relational->schema( {
classes => [
Person => {
abstract => 1,
fields => {
iarray => {
addresses => { class => 'Address', aggreg => 1 } }
}
},
Address => {
fields => {
string => [ qw( kind city ) ],
},
},
NaturalPerson => {
bases => [ qw( Person ) ],
fields => {
string => [ qw( firstName name ) ],
int => [ qw( age ) ],
ref => [ qw( partner ) ],
array => { children => 'NaturalPerson' },
}
},
LegalPerson => {
bases => [ qw( Person ) ],
fields => {
string => [ qw( name ) ],
ref => [ qw( manager ) ],
}
},
] } );
}}
The Schema lists all the classes that need persistence, along with
their attributes and the inheritance relationships. We must provide
type information for the attributes, because SQL is more typed than
Perl. We also tell Tangram that C<Person> is an abstract class, so it
wastes no time attempting to retrieve objects of that exact class.
Note that Tangram cannot deduce this information by itself. While Perl
makes it possible to extract the list of all the classes in an
application, in general not all classes will need to persist. A class
may have both persistent and non-persistent bases. As for attributes,
Perl's most typical representation for objects - a hash - even allows
two objects of the same class to have a different set of attributes.
For more information on creating Schemas, see L<Tangram::Relational>
and L<Tangram::Schema>.
=head2 Setting up a database
Now we create a database. The simplest way is to create an
empty database and let Tangram initialize it:
{{
use Tangram;
$dbh = DBI->connect(
@cp );
Tangram::Relational->deploy($schema, $dbh );
$dbh->disconnect();
}}
Tangram::Relational is the vanilla object-relational backend. It
assumes that the database understands standard SQL, and that both the
database and the related DBI driver fully implements the DBI
specification.
Tangram also comes with vendor-specific backends for Mysql and
Sybase. When a vendor-specific backend exists, it should be used in
place of the vanilla backend.
For more information, see L<Tangram::Relational>, L<Tangram::Sybase>
and L<Tangram::mysql>.
=head2 Connecting to a database
We are now ready to store objects. First we connect to the database,
using the class method Tangram::Relational::connect (or
Tangram::mysql::connect for Mysql).
The first argument of connect() the schema object; the others are
passed directly to DBI::connect. The method returns a Tangram::Storage
object that will be used to communicate with the database.
For example:
{{
$storage = Tangram::Relational->connect( $schema,
@cp );
}}
connects to a database named Springfield via the vanilla Relational
backend, using a specific account and password.
For more information on connecting to databases, see L<Tangram::Relational> and
L<Tangram::Storage>.
=head2 Inserting objects
Now we can populate the database:
{{
$storage->insert( NaturalPerson->new(
firstName => 'Montgomery', name => 'Burns' ) );
}}
This inserts a single NaturalPerson object into the database. We can
insert several objects in one call:
{{
$storage->insert(
NaturalPerson->new( firstName => 'Patty', name => 'Bouvier' ),
NaturalPerson->new( firstName => 'Selma', name => 'Bouvier' ) );
}}
Sometimes Tangram saves objects implicitly:
{{
@kids = (
NaturalPerson->new( firstName => 'Bart', name => 'Simpson' ),
NaturalPerson->new( firstName => 'Lisa', name => 'Simpson' ) );
$marge = NaturalPerson->new(
firstName => 'Marge', name => 'Simpson',
addresses => [
Address->new(
kind => 'residence', city => 'Springfield' ) ],
children => [ @kids ] );
$homer = NaturalPerson->new( firstName => 'Homer', name => 'Simpson',
addresses => [
Address->new(
kind => 'residence', city => 'Springfield' ),
Address->new(
kind => 'work', city => 'Springfield' ) ],
children => [ @kids ] );
$homer->{partner} = $marge;
$marge->{partner} = $homer;
$homer_id = $storage->insert( $homer );
}}
In the process of saving Homer, Tangram detects that it contains
references to objects that are not persistent yet (Marge, the
addresses and the kids), and inserts them automatically. Note that
Tangram can handle cycles: Homer and Marge refer to each other.
insert() returns an object id, or a list of object ids, that uniquely
identify the object(s) that have been inserted.
For more information on inserting objects, see L<Tangram::Storage>.
=head2 Updating objects
Updating works pretty much the same as inserting:
{{
my $maggie = NaturalPerson->new(
firstName => 'Maggie', name => 'Simpson' );
push @{ $homer->{children} }, $maggie;
push @{ $marge->{children} }, $maggie;
$storage->update( $homer, $marge );
}}
Here again Tangram detects that Maggie is not already persistent in
$storage and automatically inserts it. Note that we need to update
Marge explicitly because she was already persistent.
For more information on updating objects, see L<Tangram::Storage>.
=head2 Memory management
...is still up to you. Tangram won't break in-memory cycles, it's a
persistence tool, not a memory management tool. Let's make sure we
don't leak objects:
{{
$homer->{partner} = undef; # do this before $homer goes out of scope
}}
Also, when we're finished with a storage, we can explicitly disconnect it:
{{
$storage->disconnect();
}}
Whether it's important or not to disconnect the Storage depends on
what version of Perl you use. If it's prior to 5.6, you I<must>
disconnect the storage explicitly (or at least call unload())
otherwise the Storage will prevent the objects it controls from being
reclaimed by Perl. For more information see see L<Tangram::Storage>.
=head2 Finding objects
After reconnecting to Springfield, we now want to retrieve some objects.
But how do we find them? Basically there are three options
=over 4
=item *
We know their IDs.
=item *
We obtain them from another object.
=item *
We use a query.
=back
=head2 Loading by ID
When an object is inserted, Tangram assigns an identifier to it.
IDs are numbers that uniquely identify objects in the database.
C<insert> returns the ID(s) of the object(s) it was passed:
{{
$storage = Tangram::Relational->connect( $schema,
@cp );
$ned_id = $storage->insert( NaturalPerson->new(
firstNname => 'Ned', name => 'Flanders' ) );
@sisters_id = $storage->insert(
NaturalPerson->new( firstName => 'Patty', name => 'Bouvier' ),
NaturalPerson->new( firstName => 'Selma', name => 'Bouvier' ) );
}}
This enables us to retrieve the objects:
{{
$ned = $storage->load( $ned_id );
@sisters = $storage->load( @sisters_id );
}}
For more information on loading objects by id, see L<Tangram::Storage>.
=head2 Obtaining objects from other objects
Once Homer has been restored to his previous state, including his relations
with his family. Thus we can say:
{{
$storage = Tangram::Relational->connect( $schema,
@cp );
$homer = $storage->load( $homer_id ); # load by id
$marge = $homer->{partner};
@kids = @{ $homer->{children} };
}}
Actually, when Tangram loads an object that contains references to
other persistent objects, it doesn't retrieve the referenced objects
immediately. Marge is retrieved only when Homer's 'partner' field is
accessed. This mechanism is almost totally transparent, we'd have to
use C<tied> to observe a non-present collection or reference.
For more information on relationships, see L<Tangram::Schema>,
L<Tangram::Ref>, L<Tangram::Array>, L<Tangram::IntrArray>,
L<Tangram::Set> and L<Tangram::IntrSet>.
=head2 select
To retrieve all the objects of a given class, we use C<select>:
{{
$storage = Tangram::Relational->connect( $schema,
@cp );
my @people = $storage->select( 'NaturalPerson' );
}}
Tangram supports polymorphic retrieval. Let's first insert a
LegalPerson:
{{
$storage->insert( LegalPerson->new(
name => 'Springfield Nuclear Power Plant', manager => $burns ) );
}}
Now we can retrieve all the Persons - Natural or Legal - by making a
single call to select(), passing it the base class name:
{{
my @all = $storage->select( 'Person' );
}}
For more information on select(), see L<Tangram::Storage>.
=head2 Filtering
Usually we won't want to load I<all> the NaturalPersons, only those
objects that satisfy some condition. Say, for example, that we want to
load only the NaturalPersons whose name field is 'Simpson'. Here's how
this can be done:
{{
my $person = $storage->remote( 'NaturalPerson' );
my @simpsons = $storage->select( $person, $person->{name} eq 'Simpson' );
}}
This will bring in memory only the Simpsons; Burns or the Bouvier
sisters won't turn up. The filtering happens on the database server
side, not in Perl space. Internally, Tangram translates the
C<$person->{name} eq 'Simpson'> clause into a piece of SQL code that
is passed down to the database.
The above example only begins to scratch the surface of Tangram's
filtering capabilities. The following examples are all legal and working code:
{{
# find all the persons *not* named Simpson
my $person = $storage->remote( 'NaturalPerson' );
my @others = $storage->select( $person, $person->{name} ne 'Simpson' );
# same thing in a different way
my $person = $storage->remote( 'NaturalPerson' );
my @others = $storage->select( $person, !($person->{name} eq 'Simpson') );
# find all the persons who are older than me
my $person = $storage->remote( 'NaturalPerson' );
my @elders = $storage->select( $person, $person->{age} > 35 );
# find all the Simpsons older than me
my $person = $storage->remote( 'NaturalPerson' );
my @simpsons = $storage->select( $person,
$person->{name} eq 'Simpson' & $person->{age} > 35 );
# find Homer's wife - note that select *must* be called in list context
my ($person1, $person2) = $storage->remote(
qw( NaturalPerson NaturalPerson ));
my ($marge) = $storage->select( $person1,
$person1->{partner} == $person2
& $person2->{firstName} eq 'Homer' & $person2->{name} eq 'Simpson' );
# find Homer's wife - this time Homer is already in memory
my $homer = $storage->load( $homer_id );
my $person = $storage->remote( 'NaturalPerson' );
my ($marge) = $storage->select( $person,
$person->{partner} == $homer );
# find everybody who works in Springfield
my $address = $storage->remote( 'Address' );
my @population = $storage->select( $person,
$person->{addresses}->includes( $address )
& $address->{kind} eq 'work'
& $address->{city} eq 'Springfield');
# find the parents of Bart Simpson
my ($person1, $person2) = $storage->remote(
qw( NaturalPerson NaturalPerson ));
my @parents = $storage->select( $person1,
$person1->{children}->includes( $person2 )
& $person2->{firstName} eq 'Bart'
& $person2->{name} eq 'Simpson' );
# load Bart
my ($bart) = $storage->select( $person1, $person1->{firstName} eq 'Bart');
# find the parents of Bart, this time given an object already loaded
my $person = $storage->remote( 'NaturalPerson' );
@parents = $storage->select( $person,
$person->{children}->includes( $bart ) );
}}
Note that Tangram uses a single ampersand (&) or vertical bar (|) to
represent logical conjunction or disjunction, not the usual && or
||. This is due to a limitation in Perl's operator overloading
mechanism. Make sure you never forget this, because, unfortunately,
using && or || in place of & or | is not even a syntax error :(
Finally, Tangram make it possible to retrieve tuples of related objects:
{{
my ($parent, $child) = $storage->remote('NaturalPerson', 'NaturalPerson');
@pairs = $storage->select( [ $parent, $child ],
$parent->{children}->includes($child) );
}}
@pairs contains a list of references to arrays of size two; each array
contains a pair of parent and child.
For more information on filters, see L<Tangram::Expr> and L<Tangram::Remote>.
=head2 Cursors
Cursors provide a way of retrieving objects one at a time. This is
important is the result set is potentially large. cursor() takes the
same arguments as select() and returns a Cursor objects that can be
used to iterate over the result set via methods current() and next():
{{
$storage = Tangram::Relational->connect( $schema,
@cp );
# iterate over all the NaturalPersons in storage
my $cursor = $storage->cursor( 'NaturalPerson' );
while (my $person = $cursor->current())
{
# process $person
$cursor->next();
}
$cursor->close();
}}
The Cursor will be automatically closed when $cursor is garbage-collected,
but Perl doesn't define just when that may happen :( Thus it's a good idea to
explicitly close the cursor.
Each Cursor uses a separate connection to the database. Consequently you can
have several cursors open at the same, all with pending results. Of course,
mixing reads and writes to the same tables can result in deadlocks.
For more information on cursors, see L<Tangram::Storage> and
L<Tangram::Cursor>.
=head2 Remote objects
At this point, most people wonder what $person I<exactly> is and how
it all works. This section attempts to give an idea of the mechanisms
that are used.
In Tangram terminology, $person a I<remote> object. Its Perl class is
Tangram::Remote, but it's really a placeholder for an object of class
C<NaturalPerson> I<in the database>, much like a table alias in
SQL-speak.
When you request a remote object of a given class, Tangram arranges
that the remote object I<looks like> an object of the said class. It
I<seems> to have the same fields as a regular object, but don't be
misled, it's not the real thing, it's just a way of providing a nice
syntax.
If you dig it, you'll find out that a Remote is just a hash of
Tangram::Expr objects. When you say $homer->{name}, an Expr is
returned, which, most of the time, can be used like any ordinary Perl
scalar. However, an Expr represents a value I<in the database>, it's
the equivalent of Remote, only for expressions, not for objects.
Expr objects that represent scalar values (e.g. ints, floats, strings)
can be compared between them, or compared with straight Perl
scalars. Reference-like Exprs can be compared between themselves and
with references
Expr objects that represent collections have an C<include> methods
that take a persistent object, a Remote object or an ID.
The result of comparing Exprs (or calling C<include>) is a
Tangram::Expr::Filter that will translate into part of the SQL
where-clause that will be passed to the RDBMS.
For more information on remote objects, see L<Tangram::Remote>.
=head2 Multiple loads
What happens when we load the same object twice? Consider:
{{
my $person = $storage->remote( 'NaturalPerson' );
my @simpsons = $storage->select( $person, $person->{name} eq 'Simpson' );
my @people = $storage->select( 'NaturalPerson' );
}}
Obviously Homer Simpson will be retrieved by both selects. Are there
two Homers in memory now? Fortunately not. There is only one copy of
Homer in memory. When Tangram load an object, it checks whether an
object with the same ID is alredy present. If yes, it keeps the old
copy, which is desirable, since we may have changed it already.
Incidentally, this explains why a Storage will hold objects in memory
- until disconnected (again, this will change when Perl supports weak
references).
=head2 Transactions
Tangram wraps database transactions in a object-oriented interface:
$storage->tx_start();
$homer->{partner} = $marge;
$marge->{partner} = $homer;
$storage->update( $homer, $marge );
$storage->tx_commit();
Both Marge and Homer will be updated, or none will. tx_rollback() drops
the changes.
Tangram does not emulate transactions for databases that do not
support them (like earlier versions of mySql).
Unlike DBI, Tangram allows the nested transactions:
$storage->tx_start();
{
$storage->tx_start();
$patty->{partner} = $selma;
$selma->{partner} = $patty;
$storage->tx_commit();
}
$homer->{partner} = $marge;
$marge->{partner} = $homer;
$storage->update( $homer, $marge );
$storage->tx_commit();
Tangram uses a single database transaction, but commits it only when
the tx_commit()s exactly balance the tx_start()s. Thanks to this
feature any piece of code can open all the transactions it needs and
still cooperate smoothly with the rest of the application. If a DBI
transaction is already active, it will be reused; otherwise a new one
will be started.
Tangram offer a more robust alternative to the start/commit code
sandwich. tx_do() calls CODEREF in a transaction. If the CODEREF
dies, the transaction is rolled back; otherwise it's committed. The
first example can be rewritten:
$storage->tx_do( sub {
$homer->{partner} = $marge;
$marge->{partner} = $homer;
$storage->update( $homer, $marge };
} );
For more information on transactions, see L<Tangram::Storage>.
=cut
Jump to Line
Something went wrong with that request. Please try again.