ok
Direktori : /proc/thread-self/root/usr/local/share/perl5/IO/ |
Current File : //proc/thread-self/root/usr/local/share/perl5/IO/WrapTie.pm |
package IO::WrapTie; use strict; use Exporter; # Inheritance, exporting, and package version: our @ISA = qw(Exporter); our @EXPORT = qw(wraptie); our $VERSION = '2.113'; # Function, exported. sub wraptie { IO::WrapTie::Master->new(@_); } # Class method; BACKWARDS-COMPATIBILITY ONLY! sub new { shift; IO::WrapTie::Master->new(@_); } #------------------------------------------------------------ package # hide from pause IO::WrapTie::Master; #------------------------------------------------------------ use strict; use vars qw($AUTOLOAD); use IO::Handle; # We inherit from IO::Handle to get methods which invoke i/o operators, # like print(), on our tied handle: our @ISA = qw(IO::Handle); #------------------------------ # new SLAVE, TIEARGS... #------------------------------ # Create a new subclass of IO::Handle which... # # (1) Handles i/o OPERATORS because it is tied to an instance of # an i/o-like class, like IO::Scalar. # # (2) Handles i/o METHODS by delegating them to that same tied object!. # # Arguments are the slave class (e.g., IO::Scalar), followed by all # the arguments normally sent into that class's C<TIEHANDLE> method. # In other words, much like the arguments to tie(). :-) # # NOTE: # The thing $x we return must be a BLESSED REF, for ($x->print()). # The underlying symbol must be a FILEHANDLE, for (print $x "foo"). # It has to have a way of getting to the "real" back-end object... # sub new { my $master = shift; my $io = IO::Handle->new; ### create a new handle my $slave = shift; tie *$io, $slave, @_; ### tie: will invoke slave's TIEHANDLE bless $io, $master; ### return a master } #------------------------------ # AUTOLOAD #------------------------------ # Delegate method invocations on the master to the underlying slave. # sub AUTOLOAD { my $method = $AUTOLOAD; $method =~ s/.*:://; my $self = shift; tied(*$self)->$method(\@_); } #------------------------------ # PRELOAD #------------------------------ # Utility. # # Most methods like print(), getline(), etc. which work on the tied object # via Perl's i/o operators (like 'print') are inherited from IO::Handle. # # Other methods, like seek() and sref(), we must delegate ourselves. # AUTOLOAD takes care of these. # # However, it may be necessary to preload delegators into your # own class. PRELOAD will do this. # sub PRELOAD { my $class = shift; foreach (@_) { eval "sub ${class}::$_ { my \$s = shift; tied(*\$s)->$_(\@_) }"; } } # Preload delegators for some standard methods which we can't simply # inherit from IO::Handle... for example, some IO::Handle methods # assume that there is an underlying file descriptor. # PRELOAD IO::WrapTie::Master qw(open opened close read clearerr eof seek tell setpos getpos); #------------------------------------------------------------ package # hide from pause IO::WrapTie::Slave; #------------------------------------------------------------ # Teeny private class providing a new_tie constructor... # # HOW IT ALL WORKS: # # Slaves inherit from this class. # # When you send a new_tie() message to a tie-slave class (like IO::Scalar), # it first determines what class should provide its master, via TIE_MASTER. # In this case, IO::Scalar->TIE_MASTER would return IO::Scalar::Master. # Then, we create a new master (an IO::Scalar::Master) with the same args # sent to new_tie. # # In general, the new() method of the master is inherited directly # from IO::WrapTie::Master. # sub new_tie { my $self = shift; $self->TIE_MASTER->new($self,@_); ### e.g., IO::Scalar::Master->new(@_) } # Default class method for new_tie(). # All your tie-slave class (like IO::Scalar) has to do is override this # method with a method that returns the name of an appropriate "master" # class for tying that slave. # sub TIE_MASTER { 'IO::WrapTie::Master' } #------------------------------ 1; __END__ package IO::WrapTie; ### for doc generator =head1 NAME IO::WrapTie - wrap tieable objects in IO::Handle interface I<This is currently Alpha code, released for comments. Please give me your feedback!> =head1 SYNOPSIS First of all, you'll need tie(), so: require 5.004; I<Function interface (experimental).> Use this with any existing class... use IO::WrapTie; use FooHandle; ### implements TIEHANDLE interface ### Suppose we want a "FooHandle->new(&FOO_RDWR, 2)". ### We can instead say... $FH = wraptie('FooHandle', &FOO_RDWR, 2); ### Now we can use... print $FH "Hello, "; ### traditional operator syntax... $FH->print("world!\n"); ### ...and OO syntax as well! I<OO interface (preferred).> You can inherit from the L<IO::WrapTie/"Slave"> mixin to get a nifty C<new_tie()> constructor... #------------------------------ package FooHandle; ### a class which can TIEHANDLE use IO::WrapTie; @ISA = qw(IO::WrapTie::Slave); ### inherit new_tie() ... #------------------------------ package main; $FH = FooHandle->new_tie(&FOO_RDWR, 2); ### $FH is an IO::WrapTie::Master print $FH "Hello, "; ### traditional operator syntax $FH->print("world!\n"); ### OO syntax See IO::Scalar as an example. It also shows you how to create classes which work both with and without 5.004. =head1 DESCRIPTION Suppose you have a class C<FooHandle>, where... =over 4 =item * C<FooHandle> does not inherit from L<IO::Handle>. That is, it performs file handle-like I/O, but to something other than an underlying file descriptor. Good examples are L<IO::Scalar> (for printing to a string) and L<IO::Lines> (for printing to an array of lines). =item * C<FooHandle> implements the C<TIEHANDLE> interface (see L<perltie>). That is, it provides methods C<TIEHANDLE>, C<GETC>, C<PRINT>, C<PRINTF>, C<READ>, and C<READLINE>. =item * C<FooHandle> implements the traditional OO interface of L<FileHandle> and L<IO::Handle>. i.e., it contains methods like C<getline>, C<read>, C<print>, C<seek>, C<tell>, C<eof>, etc. =back Normally, users of your class would have two options: =over 4 =item * B<Use only OO syntax,> and forsake named I/O operators like C<print>. =item * B<Use with tie,> and forsake treating it as a first-class object (i.e., class-specific methods can only be invoked through the underlying object via C<tied>... giving the object a "split personality"). =back But now with L<IO::WrapTie>, you can say: $WT = wraptie('FooHandle', &FOO_RDWR, 2); $WT->print("Hello, world\n"); ### OO syntax print $WT "Yes!\n"; ### Named operator syntax too! $WT->weird_stuff; ### Other methods! And if you're authoring a class like C<FooHandle>, just have it inherit from C<IO::WrapTie::Slave> and that first line becomes even prettier: $WT = FooHandle->new_tie(&FOO_RDWR, 2); B<The bottom line:> now, almost any class can look and work exactly like an L<IO::Handle> and be used both with OO and non-OO file handle syntax. =head1 HOW IT ALL WORKS =head2 The data structures Consider this example code, using classes in this distribution: use IO::Scalar; use IO::WrapTie; $WT = wraptie('IO::Scalar',\$s); print $WT "Hello, "; $WT->print("world!\n"); In it, the C<wraptie> function creates a data structure as follows: * $WT is a blessed reference to a tied filehandle $WT glob; that glob is tied to the "Slave" object. | * You would do all your i/o with $WT directly. | | | ,---isa--> IO::WrapTie::Master >--isa--> IO::Handle V / .-------------. | | | | * Perl i/o operators work on the tied object, | "Master" | invoking the C<TIEHANDLE> methods. | | * Method invocations are delegated to the tied | | slave. `-------------' | tied(*$WT) | .---isa--> IO::WrapTie::Slave V / .-------------. | | | "Slave" | * Instance of FileHandle-like class which doesn't | | actually use file descriptors, like IO::Scalar. | IO::Scalar | * The slave can be any kind of object. | | * Must implement the C<TIEHANDLE> interface. `-------------' I<NOTE:> just as an L<IO::Handle> is really just a blessed reference to a I<traditional> file handle glob. So also, an C<IO::WrapTie::Master> is really just a blessed reference to a file handle glob I<which has been tied to some "slave" class.> =head2 How C<wraptie> works =over 4 =item 1. The call to function C<wraptie(SLAVECLASS, TIEARGS...)> is passed onto C<IO::WrapTie::Master::new()>. Note that class C<IO::WrapTie::Master> is a subclass of L<IO::Handle>. =item 2. The C<< IO::WrapTie::Master->new >> method creates a new L<IO::Handle> object, re-blessed into class C<IO::WrapTie::Master>. This object is the I<master>, which will be returned from the constructor. At the same time... =item 3. The C<new> method also creates the I<slave>: this is an instance of C<SLAVECLASS> which is created by tying the master's L<IO::Handle> to C<SLAVECLASS> via C<tie>. This call to C<tie> creates the slave in the following manner: =item 4. Class C<SLAVECLASS> is sent the message C<TIEHANDLE>; it will usually delegate this to C<< SLAVECLASS->new(TIEARGS) >>, resulting in a new instance of C<SLAVECLASS> being created and returned. =item 5. Once both master and slave have been created, the master is returned to the caller. =back =head2 How I/O operators work (on the master) Consider using an i/o operator on the master: print $WT "Hello, world!\n"; Since the master C<$WT> is really a C<blessed> reference to a glob, the normal Perl I/O operators like C<print> may be used on it. They will just operate on the symbol part of the glob. Since the glob is tied to the slave, the slave's C<PRINT> method (part of the C<TIEHANDLE> interface) will be automatically invoked. If the slave is an L<IO::Scalar>, that means L<IO::Scalar/"PRINT"> will be invoked, and that method happens to delegate to the C<print> method of the same class. So the I<real> work is ultimately done by L<IO::Scalar/"print">. =head2 How methods work (on the master) Consider using a method on the master: $WT->print("Hello, world!\n"); Since the master C<$WT> is blessed into the class C<IO::WrapTie::Master>, Perl first attempts to find a C<print> method there. Failing that, Perl next attempts to find a C<print> method in the super class, L<IO::Handle>. It just so happens that there I<is> such a method; that method merely invokes the C<print> I/O operator on the self object... and for that, see above! But let's suppose we're dealing with a method which I<isn't> part of L<IO::Handle>... for example: my $sref = $WT->sref; In this case, the intuitive behavior is to have the master delegate the method invocation to the slave (now do you see where the designations come from?). This is indeed what happens: C<IO::WrapTie::Master> contains an C<AUTOLOAD> method which performs the delegation. So: when C<sref> can't be found in L<IO::Handle>, the C<AUTOLOAD> method of C<IO::WrapTie::Master> is invoked, and the standard behavior of delegating the method to the underlying slave (here, an L<IO::Scalar>) is done. Sometimes, to get this to work properly, you may need to create a subclass of C<IO::WrapTie::Master> which is an effective master for I<your> class, and do the delegation there. =head1 NOTES B<Why not simply use the object's OO interface?> Because that means forsaking the use of named operators like C<print>, and you may need to pass the object to a subroutine which will attempt to use those operators: $O = FooHandle->new(&FOO_RDWR, 2); $O->print("Hello, world\n"); ### OO syntax is okay, BUT.... sub nope { print $_[0] "Nope!\n" } X nope($O); ### ERROR!!! (not a glob ref) B<Why not simply use tie()?> Because (1) you have to use C<tied> to invoke methods in the object's public interface (yuck), and (2) you may need to pass the tied symbol to another subroutine which will attempt to treat it in an OO-way... and that will break it: tie *T, 'FooHandle', &FOO_RDWR, 2; print T "Hello, world\n"; ### Operator is okay, BUT... tied(*T)->other_stuff; ### yuck! AND... sub nope { shift->print("Nope!\n") } X nope(\*T); ### ERROR!!! (method "print" on unblessed ref) B<Why a master and slave?> Why not simply write C<FooHandle> to inherit from L<IO::Handle?> I tried this, with an implementation similar to that of L<IO::Socket>. The problem is that I<the whole point is to use this with objects that don't have an underlying file/socket descriptor.>. Subclassing L<IO::Handle> will work fine for the OO stuff, and fine with named operators I<if> you C<tie>... but if you just attempt to say: $IO = FooHandle->new(&FOO_RDWR, 2); print $IO "Hello!\n"; you get a warning from Perl like: Filehandle GEN001 never opened because it's trying to do system-level I/O on an (unopened) file descriptor. To avoid this, you apparently have to C<tie> the handle... which brings us right back to where we started! At least the L<IO::WrapTie> mixin lets us say: $IO = FooHandle->new_tie(&FOO_RDWR, 2); print $IO "Hello!\n"; and so is not I<too> bad. C<:-)> =head1 WARNINGS Remember: this stuff is for doing L<FileHandle>-like I/O on things I<without underlying file descriptors>. If you have an underlying file descriptor, you're better off just inheriting from L<IO::Handle>. B<Be aware that new_tie() always returns an instance of a kind of IO::WrapTie::Master...> it does B<not> return an instance of the I/O class you're tying to! Invoking some methods on the master object causes C<AUTOLOAD> to delegate them to the slave object... so it I<looks> like you're manipulating a C<FooHandle> object directly, but you're not. I have not explored all the ramifications of this use of C<tie>. I<Here there be dragons>. =head1 AUTHOR Eryq (F<eryq@zeegee.com>). President, ZeeGee Software Inc (F<http://www.zeegee.com>). =head1 CONTRIBUTORS Dianne Skoll (F<dfs@roaringpenguin.com>). =head1 COPYRIGHT & LICENSE Copyright (c) 1997 Erik (Eryq) Dorfman, ZeeGee Software, Inc. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself. =cut