xref: /macosx-10.10/CPANInternal-159.1/Ima-DBI-0.35/lib/Ima/DBI.pm

package Ima::DBI;

$VERSION = '0.35';

use strict;
use base 'Class::Data::Inheritable';
use DBI;

# Some class data to store a per-class list of handles.
Ima::DBI->mk_classdata('__Database_Names');
Ima::DBI->mk_classdata('__Statement_Names');

=head1 NAME

Ima::DBI - Database connection caching and organization

=head1 SYNOPSIS

    package Foo;
    use base 'Ima::DBI';

    # Class-wide methods.
    Foo->set_db($db_name, $data_source, $user, $password);
    Foo->set_db($db_name, $data_source, $user, $password, \%attr);

    my @database_names   = Foo->db_names;
    my @database_handles = Foo->db_handles;

    Foo->set_sql($sql_name, $statement, $db_name);
    Foo->set_sql($sql_name, $statement, $db_name, $cache);

    my @statement_names   = Foo->sql_names;

    # Object methods.
    $dbh = $obj->db_*;      # Where * is the name of the db connection.
    $sth = $obj->sql_*;     # Where * is the name of the sql statement.
    $sth = $obj->sql_*(@sql_pieces);

    $obj->DBIwarn($what, $doing);

    my $rc = $obj->commit;
    my $rc = $obj->commit(@db_names);

    my $rc = $obj->rollback;
    my $rc = $obj->rollback(@db_names);


=head1 DESCRIPTION

Ima::DBI attempts to organize and facilitate caching and more efficient
use of database connections and statement handles by storing DBI and
SQL information with your class (instead of as seperate objects).
This allows you to pass around just one object without worrying about
a trail of DBI handles behind it.

One of the things I always found annoying about writing large programs
with DBI was making sure that I didn't have duplicate database handles
open.  I was also annoyed by the somewhat wasteful nature of the
prepare/execute/finish route I'd tend to go through in my subroutines.
The new DBI->connect_cached and DBI->prepare_cached helped a lot, but
I still had to throw around global datasource, username and password
information.

So, after a while I grew a small library of DBI helper routines and
techniques.  Ima::DBI is the culmination of all this, put into a nice(?),
clean(?) class to be inherited from.

=head2 Why should I use this thing?

Ima::DBI is a little odd, and it's kinda hard to explain.  So lemme
explain why you'd want to use this thing...

=over 4

=item * Consolidation of all SQL statements and database information

No matter what, embedding one language into another is messy.
DBI alleviates this somewhat, but I've found a tendency to have that
scatter the SQL around inside the Perl code.  Ima::DBI allows you to
easily group the SQL statements in one place where they are easier to
maintain (especially if one developer is writing the SQL, another writing
the Perl).  Alternatively, you can place your SQL statement alongside
the code which uses it.  Whatever floats your boat.

Database connection information (data source, username, password,
atrributes, etc...) can also be consolidated together and tracked.

Both the SQL and the connection info are probably going to change a lot,
so having them well organized and easy to find in the code is a Big Help.

=item * Holds off opening a database connection until necessary.

While Ima::DBI is informed of all your database connections and SQL
statements at compile-time, it will not connect to the database until
you actually prepare a statement on that connection.

This is obviously very good for programs that sometimes never touch
the database.  It's also good for code that has lots of possible
connections and statements, but which typically only use a few.
Kinda like an autoloader.

=item * Easy integration of the DBI handles into your class

Ima::DBI causes each database handle to be associated with your class,
allowing you to pull handles from an instance of your object, as well
as making many oft-used DBI methods available directly from your
instance.

This gives you a cleaner OO design, since you can now just throw
around the object as usual and it will carry its associated DBI
baggage with it.

=item * Honors taint mode

It always struck me as a design deficiency that tainted SQL statements
could be passed to $sth->prepare().  For example:

    # $user is from an untrusted source and is tainted.
    $user = get_user_data_from_the_outside_world;
    $sth = $dbh->prepare('DELETE FROM Users WHERE User = $user');

Looks innocent enough... but what if $user was the string "1 OR User LIKE
'%'".  You just blew away all your users. Hope you have backups.

Ima::DBI turns on the DBI->connect Taint attribute so that all DBI
methods (except execute()) will no longer accept tainted data.
See L<DBI/Taint> for details.

=item * Taints returned data

Databases should be like any other system call.  It's the scary Outside
World, thus it should be tainted.  Simple.  Ima::DBI turns on DBI's Taint
attribute on each connection.  This feature is overridable by passing
your own Taint attribute to set_db as normal for DBI.  See L<DBI/Taint>
for details.

=item * Encapsulation of some of the more repetitive bits of everyday DBI usage

I get lazy a lot and I forget to do things I really should, like using
bind_cols(), or rigorous error checking.  Ima::DBI does some of this
stuff automatically, other times it just makes it more convenient.

=item * Encapsulation of DBI's cache system

DBI's automatic handle caching system is relatively new, and some people
aren't aware of its use.  Ima::DBI uses it automatically, so you don't
have to worry about it. (It even makes it a bit more efficient)

=item * Sharing of database and sql information amongst inherited classes

Any SQL statements and connections created by a class are available to
its children via normal method inheritance.

=item * Guarantees one connection per program.

One program, one database connection (per database user).  One program,
one prepared statement handle (per statement, per database user).
That's what Ima::DBI enforces.  Extremely handy in persistant environments
(servers, daemons, mod_perl, FastCGI, etc...)

=item * Encourages use of bind parameters and columns

Bind parameters are safer and more efficient than embedding the column
information straight into the SQL statement.  Bind columns are more
efficient than normal fetching.  Ima::DBI pretty much requires the usage
of the former, and eases the use of the latter.

=back

=head2 Why shouldn't I use this thing.

=over 4

=item * It's all about OO

Although it is possible to use Ima::DBI as a stand-alone module as
part of a function-oriented design, its generally not to be used
unless integrated into an object-oriented design.

=item * Overkill for small programs

=item * Overkill for programs with only one or two SQL statements

Its up to you whether the trouble of setting up a class and jumping
through the necessary Ima::DBI hoops is worth it for small programs.
To me, it takes just as much time to set up an Ima::DBI subclass as it
would to access DBI without it... but then again I wrote the module.
YMMV.

=item * Overkill for programs that only use their SQL statements once

Ima::DBI's caching might prove to be an unecessary performance hog if
you never use the same SQL statement twice.  Not sure, I haven't
looked into it.

=back


=head1 USAGE

The basic steps to "DBIing" a class are:

=over 4

=item 1

Inherit from Ima::DBI

=item 2

Set up and name all your database connections via set_db()

=item 3

Set up and name all your SQL statements via set_sql()

=item 4

Use sql_* to retrieve your statement handles ($sth) as needed and db_*
to retreive database handles ($dbh).


=back

Have a look at L<EXAMPLE> below.

=head1 TAINTING

Ima::DBI, by default, uses DBI's Taint flag on all connections.

This means that Ima::DBI methods do not accept tainted data, and that all
data fetched from the database will be tainted.  This may be different
from the DBI behavior you're used to.  See L<DBI/Taint> for details.

=head1 Class Methods

=head2 set_db

    Foo->set_db($db_name, $data_source, $user, $password);
    Foo->set_db($db_name, $data_source, $user, $password, \%attr);

This method is used in place of DBI->connect to create your database
handles. It sets up a new DBI database handle associated to $db_name.
All other arguments are passed through to DBI->connect_cached.

A new method is created for each db you setup.  This new method is called
"db_$db_name"... so, for example, Foo->set_db("foo", ...) will create
a method called "db_foo()". (Spaces in $db_name will be translated into
underscores: '_')

%attr is combined with a set of defaults (RaiseError => 1, AutoCommit
=> 0, PrintError => 0, Taint => 1).  This is a better default IMHO,
however it does give databases without transactions (such as MySQL when
used with the default MyISAM table type) a hard time.  Be sure to turn
AutoCommit back on if your database does not support transactions.

The actual database handle creation (and thus the database connection)
is held off until a prepare is attempted with this handle.

=cut

sub _croak { my $self = shift; require Carp; Carp::croak(@_) }

sub set_db {
	my $class = shift;
	my $db_name = shift or $class->_croak("Need a db name");
	$db_name =~ s/\s/_/g;

	my $data_source = shift or $class->_croak("Need a data source");
	my $user     = shift || "";
	my $password = shift || "";
	my $attr     = shift || {};
	ref $attr eq 'HASH' or $class->_croak("$attr must be a hash reference");
	$attr = $class->_add_default_attributes($attr);

	$class->_remember_handle($db_name);
	no strict 'refs';
	*{ $class . "::db_$db_name" } =
		$class->_mk_db_closure($data_source, $user, $password, $attr);

	return 1;
}

sub _add_default_attributes {
	my ($class, $user_attr) = @_;
	my %attr = $class->_default_attributes;
	@attr{ keys %$user_attr } = values %$user_attr;
	return \%attr;
}

sub _default_attributes {
	(
		RaiseError => 1,
		AutoCommit => 0,
		PrintError => 0,
		Taint      => 1,
		RootClass  => "DBIx::ContextualFetch"
	);
}

sub _remember_handle {
	my ($class, $db) = @_;
	my $handles = $class->__Database_Names || [];
	push @$handles, $db;
	$class->__Database_Names($handles);
}

sub _mk_db_closure {
	my ($class, $dsn, $user, $pass, $attr) = @_;
        $attr ||= {};

	my $dbh;
	my $process_id = $$;
	return sub {
		# set the PID in a private cache key to prevent us
		# from sharing one with the parent after fork.  This
		# is better than disconnecting the existing $dbh since
		# the parent may still need the connection open.  Note
		# that forking code also needs to set InactiveDestroy
		# on all open handles in the child or the connection
		# will be broken during DESTROY.
		$attr->{private_cache_key_pid} = $$;

                # reopen if this is a new process or if the connection
                # is bad
		if ($process_id != $$ or
                    not ($dbh && $dbh->FETCH('Active') && $dbh->ping)) {
                    $dbh = DBI->connect_cached($dsn, $user, $pass, $attr);
                    $process_id = $$;
		}
		return $dbh;
	};

}

=head2 set_sql

    Foo->set_sql($sql_name, $statement, $db_name);
    Foo->set_sql($sql_name, $statement, $db_name, $cache);

This method is used in place of DBI->prepare to create your statement
handles. It sets up a new statement handle associated to $sql_name using
the database connection associated with $db_name.  $statement is passed
through to either DBI->prepare or DBI->prepare_cached (depending on
$cache) to create the statement handle.

If $cache is true or isn't given, then prepare_cached() will be used to
prepare the statement handle and it will be cached.  If $cache is false
then a normal prepare() will be used and the statement handle will be
recompiled on every sql_*() call.  If you have a statement which changes
a lot or is used very infrequently you might not want it cached.

A new method is created for each statement you set up.  This new method
is "sql_$sql_name"... so, as with set_db(), Foo->set_sql("bar", ...,
"foo"); will create a method called "sql_bar()" which uses the database
connection from "db_foo()". Again, spaces in $sql_name will be translated
into underscores ('_').

The actual statement handle creation is held off until sql_* is first
called on this name.

=cut

sub set_sql {
	my ($class, $sql_name, $statement, $db_name, $cache) = @_;
	$cache = 1 unless defined $cache;

	# ------------------------- sql_* closure ----------------------- #
	my $db_meth = $db_name;
	$db_meth =~ s/\s/_/g;
	$db_meth = "db_$db_meth";

	(my $sql_meth = $sql_name) =~ s/\s/_/g;
	$sql_meth = "sql_$sql_meth";

	# Remember the name of this handle for the class.
	my $handles = $class->__Statement_Names || [];
	push @$handles, $sql_name;
	$class->__Statement_Names($handles);

	no strict 'refs';
	*{ $class . "::$sql_meth" } =
		$class->_mk_sql_closure($sql_name, $statement, $db_meth, $cache);

	return 1;
}

sub _mk_sql_closure {
	my ($class, $sql_name, $statement, $db_meth, $cache) = @_;

	return sub {
		my $class = shift;
		my $dbh   = $class->$db_meth();

		# Everything must pass through sprintf, even if @_ is empty.
		# This is to do proper '%%' translation.
		my $sql = $class->transform_sql($statement => @_);
		return $cache
			? $dbh->prepare_cached($sql)
			: $dbh->prepare($sql);
	};
}

=head2 transform_sql

To make up for the limitations of bind parameters, $statement can contain
sprintf() style formatting (ie. %s and such) to allow dynamically
generated SQL statements (so to get a real percent sign, use '%%').

The translation of the SQL happens in transform_sql(), which can be
overridden to do more complex transformations. See L<Class::DBI> for an
example.

=cut

sub transform_sql {
	my ($class, $sql, @args) = @_;
	return sprintf $sql, @args;
}

=head2 db_names / db_handles

  my @database_names   = Foo->db_names;
  my @database_handles = Foo->db_handles;
  my @database_handles = Foo->db_handles(@db_names);

Returns a list of the database handles set up for this class using
set_db().  This includes all inherited handles.

db_names() simply returns the name of the handle, from which it is
possible to access it by converting it to a method name and calling
that db method...

    my @db_names = Foo->db_names;
    my $db_meth = 'db_'.$db_names[0];
    my $dbh = $foo->$db_meth;

Icky, eh?  Fortunately, db_handles() does this for you and returns a
list of database handles in the same order as db_names().  B<Use this
sparingly> as it will connect you to the database if you weren't
already connected.

If given @db_names, db_handles() will return only the handles for
those connections.

These both work as either class or object methods.

=cut

sub db_names { @{ $_[0]->__Database_Names || [] } }

sub db_handles {
	my ($self, @db_names) = @_;
	@db_names = $self->db_names unless @db_names;
	return map $self->$_(), map "db_$_", @db_names;
}

=head2 sql_names

  my @statement_names   = Foo->sql_names;

Similar to db_names() this returns the names of all SQL statements set
up for this class using set_sql(), inherited or otherwise.

There is no corresponding sql_handles() because we can't know what
arguments to pass in.

=cut

sub sql_names { @{ $_[0]->__Statement_Names || [] } }

=head1 Object Methods

=head2 db_*

    $dbh = $obj->db_*;

This is how you directly access a database handle you set up with set_db.

The actual particular method name is derived from what you told set_db.

db_* will handle all the issues of making sure you're already
connected to the database.

=head2 sql_*

    $sth = $obj->sql_*;
    $sth = $obj->sql_*(@sql_pieces);

sql_*() is a catch-all name for the methods you set up with set_sql().
For instance, if you did:

    Foo->set_sql('GetAllFoo', 'Select * From Foo', 'SomeDb');

you'd run that statement with sql_GetAllFoo().

sql_* will handle all the issues of making sure the database is
already connected, and the statement handle is prepared.  It returns a
prepared statement handle for you to use.  (You're expected to
execute() it)

If sql_*() is given a list of @sql_pieces it will use them to fill in
your statement, assuming you have sprintf() formatting tags in your
statement.  For example:

    Foo->set_sql('GetTable', 'Select * From %s', 'Things');

    # Assuming we have created an object... this will prepare the
    # statement 'Select * From Bar'
    $sth = $obj->sql_Search('Bar');

Be B<very careful> with what you feed this function.  It cannot
do any quoting or escaping for you, so it is totally up to you
to take care of that.  Fortunately if you have tainting on you
will be spared the worst.

It is recommended you only use this in cases where bind parameters
will not work.

=head2 DBIwarn

    $obj->DBIwarn($what, $doing);

Produces a useful error for exceptions with DBI.

B<I'm not particularly happy with this interface>

Most useful like this:

    eval {
        $self->sql_Something->execute($self->{ID}, @stuff);
    };
    if($@) {
        $self->DBIwarn($self->{ID}, 'Something');
                return;
    }


=cut

sub DBIwarn {
	my ($self, $thing, $doing) = @_;
	my $errstr = "Failure while doing '$doing' with '$thing'\n";
	$errstr .= $@ if $@;

	require Carp;
	Carp::carp $errstr;

	return 1;
}

=head1 Modified database handle methods

Ima::DBI makes some of the methods available to your object that are
normally only available via the database handle.  In addition, it
spices up the API a bit.

=head2 commit

    $rc = $obj->commit;
    $rc = $obj->commit(@db_names);

Derived from $dbh->commit() and basically does the same thing.

If called with no arguments, it causes commit() to be called on all
database handles associated with $obj.  Otherwise it commits all
database handles whose names are listed in @db_names.

Alternatively, you may like to do:  $rc = $obj->db_Name->commit;

If all the commits succeeded it returns true, false otherwise.

=cut

sub commit {
	my ($self, @db_names) = @_;
	return grep(!$_, map $_->commit, $self->db_handles(@db_names)) ? 0 : 1;
}

=head2 rollback

    $rc = $obj->rollback;
    $rc = $obj->rollback(@db_names);

Derived from $dbh->rollback, this acts just like Ima::DBI->commit,
except that it calls rollback().

Alternatively, you may like to do:  $rc = $obj->db_Name->rollback;

If all the rollbacks succeeded it returns true, false otherwise.

=cut

sub rollback {
	my ($self, @db_names) = @_;
	return grep(!$_, map $_->rollback, $self->db_handles(@db_names)) ? 0 : 1;
}

=head1 EXAMPLE

    package Foo;
    use base qw(Ima::DBI);

    # Set up database connections (but don't connect yet)
    Foo->set_db('Users', 'dbi:Oracle:Foo', 'admin', 'passwd');
    Foo->set_db('Customers', 'dbi:Oracle:Foo', 'Staff', 'passwd');

    # Set up SQL statements to be used through out the program.
    Foo->set_sql('FindUser', <<"SQL", 'Users');
        SELECT  *
        FROM    Users
        WHERE   Name LIKE ?
    SQL

    Foo->set_sql('ChangeLanguage', <<"SQL", 'Customers');
        UPDATE  Customers
        SET     Language = ?
        WHERE   Country = ?
    SQL

    # rest of the class as usual.

    package main;

    $obj = Foo->new;

    eval {
        # Does connect & prepare
        my $sth = $obj->sql_FindUser;
        # bind_params, execute & bind_columns
        $sth->execute(['Likmi%'], [\($name)]);
        while( $sth->fetch ) {
            print $name;
        }

        # Uses cached database and statement handles
        $sth = $obj->sql_FindUser;
        # bind_params & execute.
        $sth->execute('%Hock');
        @names = $sth->fetchall;

        # connects, prepares
        $rows_altered = $obj->sql_ChangeLanguage->execute(qw(es_MX mx));
    };
    unless ($@) {
        # Everything went okay, commit the changes to the customers.
        $obj->commit('Customers');
    }
    else {
        $obj->rollback('Customers');
        warn "DBI failure:  $@";
    }

=head1 USE WITH MOD_PERL, FASTCGI, ETC.

To help with use in forking environments, Ima::DBI database handles keep
track of the PID of the process they were openend under.  If they notice
a change (because you forked a new process), a new handle will be opened
in the new process.  This prevents a common problem seen in environments
like mod_perl where people would open a handle in the parent process and
then run into trouble when they try to use it from a child process.

Because Ima::DBI handles keeping database connections persistent and
prevents problems with handles openend before forking, it is not
necessary to use Apache::DBI when using Ima::DBI.  However, there is
one feature of Apache::DBI which you will need in a mod_perl or FastCGI
environment, and that's the automatic rollback it does at the end of each
request.  This rollback provides safety from transactions left hanging
when some perl code dies -- a serious problem which could grind your
database to a halt with stale locks.

To replace this feature on your own under mod_perl, you can add something
like this in a handler at any phase of the request:

   $r->push_handlers(PerlCleanupHandler => sub {
       MyImaDBI->rollback();
   });

Here C<MyImaDBI> is your subclass of Ima::DBI.  You could also make this
into an actual module and set the PerlCleanupHandler from your httpd.conf.
A similar approach should work in any long-running environment which has
a hook for running some code at the end of each request.

=head1 TODO, Caveat, BUGS, etc....

=over 4

=item I seriously doubt that it's thread safe.

You can bet cupcackes to sno-cones that much havoc will be wrought if
Ima::DBI is used in a threaded Perl.

=item Should make use of private_* handle method to store information

=item The docs stink.

The docs were originally written when I didn't have a good handle on
the module and how it will be used in practical cases.  I need to
rewrite the docs from the ground up.

=item Need to add debugging hooks.

The thing which immediately comes to mind is a Verbose flag to print
out SQL statements as they are made as well as mention when database
connections are made, etc...

=back

=head1 MAINTAINERS

Tony Bowden <tony@tmtm.com> and Perrin Harkins <perrin@elem.com>

=head1 ORIGINAL AUTHOR

Michael G Schwern <schwern@pobox.com>

=head1 LICENSE

This module is free software.  You may distribute under the same terms
as Perl itself.  IT COMES WITHOUT WARRANTY OF ANY KIND.

=head1 THANKS MUCHLY

Tim Bunce, for enduring many DBI questions and adding Taint,
prepare_cached and connect_cached methods to DBI, simplifying this
greatly!

Arena Networks, for effectively paying for Mike to write most of this
module.

=head1 SEE ALSO

L<DBI>.

You may also choose to check out L<Class::DBI> which hides most of this
from view.

=cut

return 1001001;