Ima::DBI man page on Fedora

Man page or keyword search:  
man Server   31170 pages
apropos Keyword Search (all sections)
Output format
Fedora logo
[printable version]

Ima::DBI(3)	      User Contributed Perl Documentation	   Ima::DBI(3)

NAME
       Ima::DBI - Database connection caching and organization

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);

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.

   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...

       ·   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.

       ·   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.

       ·   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.

       ·   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
	   "Taint" in DBI for details.

       ·   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 "Taint" in DBI for details.

       ·   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.

       ·   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)

       ·   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.

       ·   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...)

       ·   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.

   Why shouldn't I use this thing.
       ·   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.

       ·   Overkill for small programs

       ·   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.

       ·   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.

USAGE
       The basic steps to "DBIing" a class are:

       1.  Inherit from Ima::DBI

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

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

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

       Have a look at EXAMPLE below.

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 "Taint" in DBI for
       details.

Class Methods
   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.

   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.

   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 Class::DBI for an
       example.

   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().  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.

   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.

Object Methods
   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.

   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 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.

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

       Produces a useful error for exceptions with DBI.

       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;
	   }

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.

   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.

   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.

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:  $@";
	   }

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 "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.

TODO, Caveat, BUGS, etc....
       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.

       Should make use of private_* handle method to store information
       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.

       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...

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

ORIGINAL AUTHOR
       Michael G Schwern <schwern@pobox.com>

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

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.

SEE ALSO
       DBI.

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

perl v5.14.1			  2007-06-10			   Ima::DBI(3)
[top]

List of man pages available for Fedora

Copyright (c) for man pages and the logo by the respective OS vendor.

For those who want to learn more, the polarhome community provides shell access and support.

[legal] [privacy] [GNU] [policy] [cookies] [netiquette] [sponsors] [FAQ]
Tweet
Polarhome, production since 1999.
Member of Polarhome portal.
Based on Fawad Halim's script.
....................................................................
Vote for polarhome
Free Shell Accounts :: the biggest list on the net