This and other RFCs are available on the web at http://dev.perl.org/rfc/ =head1 TITLE Objects: Revamp tie to support extensibility (Massive tie changes) =head1 VERSION Maintainer: Nathan Wiger <[EMAIL PROTECTED]> Date: 7 Sep 2000 Last Modified: 20 Sep 2000 Mailing List: [EMAIL PROTECTED] Number: 200 Version: 2 Status: Developing Requires: RFC 159 =head1 CHANGES 1. Dropped all references to "indirect object" syntax, since it was misleading, inaccurate, and unimplementable. Instead, added a C<use tie> pragma to arbitrarily override functions. 2. Dropped suggestion to merge TIE* methods 3. Added an UNTIE method and dropped the idea that untie should take all references out of scope. =head1 ABSTRACT C<tie> is really cool. Mostly. It has an amazing amount of power in concept, but suffers from several limitations which this RFC attempts to address. =head1 DESCRIPTION =head2 Overview Many people have expressed problems with tie, including Larry [1]. C<tie> suffers from several limitations: 1. It is non-extensible; you are limited to using functions that have been implemented with tie hooks in them already. 2. Any additional functions require mixed calls to tied and OO interfaces, defeating a chief goal: transparency. 3. It is slow. Very slow, in fact. 4. You can't easily integrate tie and operator overloading. 5. If defining tied and OO interfaces, you must define duplicate functions or use typeglobs. 6. Some parts of the syntax are, well, kludgey This RFC attempts to address all of these points with some changes to syntax and implementation concepts. It interacts with the concept of B<polymorphic objects>, described in B<RFC 159>, to provide a simple and extensible framework. =head2 New Concepts This RFC proposes two key principles that will provide a more general-purpose C<tie> framework: 1. Operator, data, and syntax overloading will be done via the ALLCAPS methods described in B<RFC 159>. 2. All functions can be overloaded via the C<use tie> pragma. In addition, the declaration of a tie statement is suggested to be changed into a standard indirect object function: $object = tie Tie::Class @array_to_tie; The default C<tie>ing would be performed by C<UNIVERSAL::tie>, which would be a new method that properly "blessed" the tied variable and then simply turned around and called the class's C<TIE*> method, similar to how the builtin C<tie> works currently. There are many changes, so let's go through them one at a time and then revisit how they will all tie (ha-ha) together at the end. =head2 Syntax Changes =head3 Drop C<tie> builtin and replace with C<UNIVERSAL::tie> As mentioned above, this allows us to call C<tie> in a simple indirect object form. This eliminates one more special-case function which currently requires that quotes be placed around the class name. This syntax should simply be modified to be called on the object it will be tied to, since C<tie> is after all an object constructor. =head3 Drop C<TIEHANDLE> Thanks to the below syntax, differentiating between filehandles and other scalars is no longer important. It would also be very difficult to make this distinction, since in Perl 6 filehandles are intended to be C<$scalars>. =head3 Continue to do data handling through ALLCAPS methods This will not change. C<STORE> and C<FETCH>, along with other functions described in B<RFC 159> and below, will continue to do data handling. In addition, these methods will be used for operator overloading as well, providing a unified C<tie> and C<use overload> environment. =head3 Add C<UNTIE> method called by C<untie> When called, C<untie> currently suffers the somewhat nasty problem of not being able to automatically destroy inner references. This means if you've mixed OO and C<tie>d calls, you may not be able to destroy your tied object as easily as you like. [2] An C<UNTIE> method should be added which is called when a tied variable is untied. This solves the problem of C<DESTROY> not being called when you think it's going to be. =head3 Ability to C<tie> arbitrary functions Currently, C<tie> suffers from being non-extensible: push @tied_array, $value; sort { $a <=> $b } @tied_array; The first one can be implemented as C<PUSH> by your tied array class, but there is no way that you can transparently offer a custom C<sort> routine. While Perl 5.6 finally has a fairly substantial collection of C<tie> methods, it is easy to imagine that future functions will arise which you want to C<tie>, but which support has not been added for yet. Plus, if you want to support extra methods of your own, you must mix object and tied calls: # Perl 5 $obj = tie %trans, 'Transaction'; $trans{$var} = $value; $obj->lock($var); Unfortunately, this defeats one of the key purposes of C<tie>, which is B<OO transparency>. And, creating a class that supports both OO and tied interfaces is difficult, requiring typeglobs or duplicate handler functions. Instead, this RFC proposes that C<tie>'s operation become much more fundamental and generalizable, through the introduction of a new C<use tie> pragma. This pragma can be used to overload arbitrary functions: package MyData; use tie sort => \&SORT, push => \&MYPUSH, lock => \&lock; sub TIEARRAY { .. .} # called on sort sub SORT { ... } # called on push sub MYPUSH { ... } # called on lock sub lock { ... } When a function is called with the given name from a program that uses the tied array, then that function is automatically overloaded. If a function does not exist in the package's namespace that is using C<tie>, then a function alias is automatically exported. So: tie MyData @data; push @data, $stuff; # $obj->MYPUSH($stuff); @s = sort { ... } @data; # $obj->SORT({...}); lock @data; # $obj->lock; In order for this to be realistic, the tied argument B<must> be the first data argument to the function. As such, these: push @untied, @data; lock $junk, @data; Would not cause C<@data>'s custom methods to be called. Also, a fully qualified function name: CORE::push @data, $stuff; Would also cause C<@data>'s custom C<MYPUSH> method not to be called. In addition, C<tied> methods can be called on individual elements as well. For example: lock $data[0]; # $obj->lock(0); Note that here, the index is passed to the C<lock> function, just like how C<STORE> and C<FETCH> work for arrays. This allows your C<lock> function to handle locking both the whole array and also individual elements. Note that operator and data access is still done by ALLCAPS methods, in fact the same ones described in L<RFC 159>. The reason for this is symmetry: Like polymorphic objects, we can now warp our C<tie>d classes in whatever way we desire. In fact, one could imagine a simple matrix math class: tie My::Matrix @a; @a + @b; # $obj->ADD(@b); $a[0] = 4; # $obj->STORE(0, 4); @a * @b; # $obj->MUL(@b); We also no longer have to care about the differences between filehandles and other scalars: tie My::Handle $FILE; print $FILE @stuff; # $obj->print(@stuff); flush $FILE; # $obj->flush; close $FILE; # $obj->close In each of these examples, function overriding is accomplished by the C<use tie> pragma. =head2 Function Summary This is a summary of all the functions that should be implemented in C<tie> in Perl 6. Any functions not mentioned here should be dropped from the C<tie> interface in Perl 6, instead replaced with the automatic indirect object calling form: General Methods ----------------------------------------------------- TIE Constructor DESTROY Destructor STORE Data storage FETCH Data retrieval Hash-Specific Methods ----------------------------------------------------- FIRSTKEY Get first key during keys/values/each NEXTKEY Iterate through keys/values/each CLEAR Clearing or resetting of hash Array-Specific Methods ----------------------------------------------------- FETCHSIZE scalar @array (basically) STORESIZE Set $#array EXTEND Pre-extend array size CLEAR Clearing or resetting of array Other Methods ----------------------------------------------------- Include all other methods described in RFC 159 That's it. Anything else that you want to override must be specified with the C<use tie> pragma. =head2 Example: Transaction Here is an example of how a tied class may be implemented under this RFC: # A class to do some simple transactional locking # A much more robust version could implement RFC 130 package Transaction; use Carp; use strict; use tie lock => \&lock, unlock => \&unlock, unlock_all => \&unlock_all; # Include tied interface sub TIEHASH { my $self = self; # RFC 152 :-) bless {@_}, $self; } # And also include OO interface per RFC 189 # Note: For both of these, simply allow UNIVERSAL::new and # UNIVERSAL::tie to take care of the actual calls. sub NEW { my $self = self; bless {@_}, $self; } sub RENEW { croak "Fatal: Reblessing transactional hashes not allowed!"; } # Include those functions we want to override # Our internal data functions are in ALLCAPS and most come # from RFC 159 (as well as previous tie implementations) sub STORE { my $self = self; if ($self->{LOCKED}->{$_[0]}) { croak "Fatal: Attempt to modify locked key $_[0]!"; } $self->{DATA}->{$_[0]} = $_[1]; } sub FETCH { my $self = self; return $self->{DATA}->{$_[0]}; } # Hash-specific method sub CLEAR { my $self = self; # Check for any locked values still remaining if (keys %{$self->{LOCKED}}) { croak "Fatal: Attempt to clear hash with locked keys!"; } undef $self->{DATA}; } # Want to override what each() and keys() do # Mostly stolen from Camel-3 p. 383 sub FIRSTKEY { my $self = self; my $temp = keys %{$self->{DATA}}; return scalar each %{$self->{DATA}}; } sub NEXTKEY { my $self = self; return scalar each %{$self->{DATA}}; } # Override addition just for demonstration purposes sub ADD { my $self = self; $self->{DATA}->{$_[0]} += (rand * $_[1]); } # Now add any Perl or custom functions that we want these # objects to be able to handle sub lock { my $self = self; $self->{LOCKED}->{$_[0]} = 1; } sub unlock { my $self = self; carp "Warning: Key $_[0] already unlocked" unless $self->{LOCKED}->{$_[0]}; delete $self->{LOCKED}->{$_[0]}; } sub unlock_all { my $self = self; carp "Notice: All values unlocked" unless $self->{LOCKED}; undef $self->{LOCKED}; } # Warn if we have locked values still sub DESTROY { my $self = self; if (keys %{$self->{LOCKED}}) { carp "Warning: Destroying transaction with locked keys!"; } undef $self->{LOCKED}; undef $self->{DATA}; } # Use our Transaction class package main; use CGI; my $cgi = new CGI; tie Transaction %trans; # Transaction->TIEHASH (thru UNIVERSAL::tie) # Generate our session id # Yes I know this is massively insecure ;-) srand; $trans{session} = rand; # All of these call $obj->STORE($var) $trans{name} = $cgi->param('name'); $trans{email} = $cgi->param('email'); $trans{cc} = $cgi->param('cc'); $trans{amount}= $cgi->param('amount'); # Lock our amount while we're charging the card... lock $trans{cc}; # $obj->lock('cc'); lock $trans{amount}; # $obj->lock('amount'); for ($try = 0; $try < 3; $try++) { # Attempt to charge them next unless charge_card($trans{cc}, $trans{amount}); $trans{chargedate} = localtime; } unlock $trans{cc}; # $obj->unlock('cc'); # Check if we were successful die "Could no charge card $trans{cc}" unless $trans{chargedate} # Increment our session id # ++$trans{session} calls $obj->STORE($obj->ADD('session', 1)) $cgi->param('session') = ++$trans{session}; # Kill our transaction unlock_all %trans; # $obj->unlock_all; Note how we are easily able to add three new methods, C<lock>, C<unlock>, and C<unlock_all>, which are directly translated for us, meaning we don't have to mix OO and tied variable calls. This provides true object transparence. Note also how our overloaded C<ADD> operator is used to increment our session number as well, all transparently to the user. =head1 IMPLEMENTATION Conceptually, implementation is straightforward, but quite different from tie's current form: 1. Drop C<tie> builtin and replace with C<UNIVERSAL::tie>. 2. Drop hardwired internal function translation and instead add the C<use tie> pragma to overload arbitrary functions. 3. Add C<UNTIE> method called by C<untie>. Looking at C<pp_sys.c> it appears this may be in 5.7 already. 4. Drop C<TIEHANDLE> method. I'm in the process of coming up with a "real" implementation section, but I'm so short on time I doubt this will happen by the time this RFC freezes. =head1 MIGRATION To keep complete backwards compatibility, the p52p6 translator could simply add a line like this: use tie push => \&PUSH, pop => \&POP, shift => \&SHIFT ... which would include all of the Perl 5 methods for an array. Similar lines could be added for hashes. No translation would have to occur for scalars, since data methods remain automatically invoked still per C<RFC 159>. Many of the changes in this RFC build on and add power to C<tie>, so do not require translation because they are new. =head1 NOTES [1] http://www.mail-archive.com/perl6-language@perl.org/msg02087.html [2] Camel-3 p. 395 has an excellent description of this problem. =head1 REFERENCES RFC 159: True Polymorphic Objects RFC 152: Replace invocant in @_ with self() builtin RFC 189: Objects : Hierarchical calls to initializers and destructors RFC 130: Transaction-enabled variables for Perl6 Camel-3 Chapter on C<tie>, p363-398 Thanks to Nathan Torkington for his input and support