1 ;# $Id: Storable.pm,v 1.0 2000/09/01 19:40:41 ram Exp $
3 ;# Copyright (c) 1995-2000, Raphael Manfredi
5 ;# You may redistribute only under the same terms as Perl 5, as specified
6 ;# in the README file that comes with the distribution.
8 ;# $Log: Storable.pm,v $
9 ;# Revision 1.0 2000/09/01 19:40:41 ram
10 ;# Baseline for first official release.
15 package Storable; @ISA = qw(Exporter DynaLoader);
17 @EXPORT = qw(store retrieve);
19 nstore store_fd nstore_fd fd_retrieve
23 lock_store lock_nstore lock_retrieve
27 use vars qw($forgive_me $VERSION);
30 *AUTOLOAD = \&AutoLoader::AUTOLOAD; # Grrr...
33 # Use of Log::Agent is optional
36 eval "use Log::Agent";
38 unless (defined @Log::Agent::EXPORT) {
52 # They might miss :flock in Fcntl
57 if (exists $Fcntl::EXPORT_TAGS{'flock'}) {
58 Fcntl->import(':flock');
70 sub retrieve_fd { &fd_retrieve } # Backward compatibility
79 # Store target object hierarchy, identified by a reference to its root.
80 # The stored object tree may later be retrieved to memory via retrieve.
81 # Returns undef if an I/O error occurred, in which case the file is
85 return _store(\&pstore, @_, 0);
91 # Same as store, but in network order.
94 return _store(\&net_pstore, @_, 0);
100 # Same as store, but flock the file first (advisory locking).
103 return _store(\&pstore, @_, 1);
109 # Same as nstore, but flock the file first (advisory locking).
112 return _store(\&net_pstore, @_, 1);
115 # Internal store to file routine
119 my ($file, $use_locking) = @_;
120 logcroak "not a reference" unless ref($self);
121 logcroak "too many arguments" unless @_ == 2; # No @foo in arglist
123 open(FILE, ">$file") || logcroak "can't create $file: $!";
124 binmode FILE; # Archaic systems...
127 logcarp "Storable::lock_store: fcntl/flock emulation broken on $^O";
130 flock(FILE, LOCK_EX) ||
131 logcroak "can't get exclusive lock on $file: $!";
133 # Unlocking will happen when FILE is closed
135 my $da = $@; # Don't mess if called from exception handler
137 # Call C routine nstore or pstore, depending on network order
138 eval { $ret = &$xsptr(*FILE, $self) };
139 close(FILE) or $ret = undef;
140 unlink($file) or warn "Can't unlink $file: $!\n" if $@ || !defined $ret;
141 logcroak $@ if $@ =~ s/\.?\n$/,/;
143 return $ret ? $ret : undef;
149 # Same as store, but perform on an already opened file descriptor instead.
150 # Returns undef if an I/O error occurred.
153 return _store_fd(\&pstore, @_);
159 # Same as store_fd, but in network order.
162 my ($self, $file) = @_;
163 return _store_fd(\&net_pstore, @_);
166 # Internal store routine on opened file descriptor
171 logcroak "not a reference" unless ref($self);
172 logcroak "too many arguments" unless @_ == 1; # No @foo in arglist
173 my $fd = fileno($file);
174 logcroak "not a valid file descriptor" unless defined $fd;
175 my $da = $@; # Don't mess if called from exception handler
177 # Call C routine nstore or pstore, depending on network order
178 eval { $ret = &$xsptr($file, $self) };
179 logcroak $@ if $@ =~ s/\.?\n$/,/;
181 return $ret ? $ret : undef;
187 # Store oject and its hierarchy in memory and return a scalar
188 # containing the result.
191 _freeze(\&mstore, @_);
197 # Same as freeze but in network order.
200 _freeze(\&net_mstore, @_);
203 # Internal freeze routine
207 logcroak "not a reference" unless ref($self);
208 logcroak "too many arguments" unless @_ == 0; # No @foo in arglist
209 my $da = $@; # Don't mess if called from exception handler
211 # Call C routine mstore or net_mstore, depending on network order
212 eval { $ret = &$xsptr($self) };
213 logcroak $@ if $@ =~ s/\.?\n$/,/;
215 return $ret ? $ret : undef;
221 # Retrieve object hierarchy from disk, returning a reference to the root
222 # object of that tree.
231 # Same as retrieve, but with advisory locking.
237 # Internal retrieve routine
239 my ($file, $use_locking) = @_;
241 open(FILE, $file) || logcroak "can't open $file: $!";
242 binmode FILE; # Archaic systems...
244 my $da = $@; # Could be from exception handler
247 logcarp "Storable::lock_store: fcntl/flock emulation broken on $^O";
250 flock(FILE, LOCK_SH) || logcroak "can't get shared lock on $file: $!";
251 # Unlocking will happen when FILE is closed
253 eval { $self = pretrieve(*FILE) }; # Call C routine
255 logcroak $@ if $@ =~ s/\.?\n$/,/;
263 # Same as retrieve, but perform from an already opened file descriptor instead.
267 my $fd = fileno($file);
268 logcroak "not a valid file descriptor" unless defined $fd;
270 my $da = $@; # Could be from exception handler
271 eval { $self = pretrieve($file) }; # Call C routine
272 logcroak $@ if $@ =~ s/\.?\n$/,/;
280 # Recreate objects in memory from an existing frozen image created
281 # by freeze. If the frozen image passed is undef, return undef.
285 return undef unless defined $frozen;
287 my $da = $@; # Could be from exception handler
288 eval { $self = mretrieve($frozen) }; # Call C routine
289 logcroak $@ if $@ =~ s/\.?\n$/,/;
296 Storable - persistency for perl data structures
301 store \%table, 'file';
302 $hashref = retrieve('file');
304 use Storable qw(nstore store_fd nstore_fd freeze thaw dclone);
307 nstore \%table, 'file';
308 $hashref = retrieve('file'); # There is NO nretrieve()
310 # Storing to and retrieving from an already opened file
311 store_fd \@array, \*STDOUT;
312 nstore_fd \%table, \*STDOUT;
313 $aryref = fd_retrieve(\*SOCKET);
314 $hashref = fd_retrieve(\*SOCKET);
316 # Serializing to memory
317 $serialized = freeze \%table;
318 %table_clone = %{ thaw($serialized) };
320 # Deep (recursive) cloning
321 $cloneref = dclone($ref);
324 use Storable qw(lock_store lock_nstore lock_retrieve)
325 lock_store \%table, 'file';
326 lock_nstore \%table, 'file';
327 $hashref = lock_retrieve('file');
331 The Storable package brings persistency to your perl data structures
332 containing SCALAR, ARRAY, HASH or REF objects, i.e. anything that can be
333 convenientely stored to disk and retrieved at a later time.
335 It can be used in the regular procedural way by calling C<store> with
336 a reference to the object to be stored, along with the file name where
337 the image should be written.
338 The routine returns C<undef> for I/O problems or other internal error,
339 a true value otherwise. Serious errors are propagated as a C<die> exception.
341 To retrieve data stored to disk, use C<retrieve> with a file name,
342 and the objects stored into that file are recreated into memory for you,
343 a I<reference> to the root object being returned. In case an I/O error
344 occurs while reading, C<undef> is returned instead. Other serious
345 errors are propagated via C<die>.
347 Since storage is performed recursively, you might want to stuff references
348 to objects that share a lot of common data into a single array or hash
349 table, and then store that object. That way, when you retrieve back the
350 whole thing, the objects will continue to share what they originally shared.
352 At the cost of a slight header overhead, you may store to an already
353 opened file descriptor using the C<store_fd> routine, and retrieve
354 from a file via C<fd_retrieve>. Those names aren't imported by default,
355 so you will have to do that explicitely if you need those routines.
356 The file descriptor you supply must be already opened, for read
357 if you're going to retrieve and for write if you wish to store.
359 store_fd(\%table, *STDOUT) || die "can't store to stdout\n";
360 $hashref = fd_retrieve(*STDIN);
362 You can also store data in network order to allow easy sharing across
363 multiple platforms, or when storing on a socket known to be remotely
364 connected. The routines to call have an initial C<n> prefix for I<network>,
365 as in C<nstore> and C<nstore_fd>. At retrieval time, your data will be
366 correctly restored so you don't have to know whether you're restoring
367 from native or network ordered data. Double values are stored stringified
368 to ensure portability as well, at the slight risk of loosing some precision
369 in the last decimals.
371 When using C<fd_retrieve>, objects are retrieved in sequence, one
372 object (i.e. one recursive tree) per associated C<store_fd>.
374 If you're more from the object-oriented camp, you can inherit from
375 Storable and directly store your objects by invoking C<store> as
376 a method. The fact that the root of the to-be-stored tree is a
377 blessed reference (i.e. an object) is special-cased so that the
378 retrieve does not provide a reference to that object but rather the
379 blessed object reference itself. (Otherwise, you'd get a reference
380 to that blessed object).
384 The Storable engine can also store data into a Perl scalar instead, to
385 later retrieve them. This is mainly used to freeze a complex structure in
386 some safe compact memory place (where it can possibly be sent to another
387 process via some IPC, since freezing the structure also serializes it in
388 effect). Later on, and maybe somewhere else, you can thaw the Perl scalar
389 out and recreate the original complex structure in memory.
391 Surprisingly, the routines to be called are named C<freeze> and C<thaw>.
392 If you wish to send out the frozen scalar to another machine, use
393 C<nfreeze> instead to get a portable image.
395 Note that freezing an object structure and immediately thawing it
396 actually achieves a deep cloning of that structure:
398 dclone(.) = thaw(freeze(.))
400 Storable provides you with a C<dclone> interface which does not create
401 that intermediary scalar but instead freezes the structure in some
402 internal memory space and then immediatly thaws it out.
404 =head1 ADVISORY LOCKING
406 The C<lock_store> and C<lock_nstore> routine are equivalent to C<store>
407 and C<nstore>, only they get an exclusive lock on the file before
408 writing. Likewise, C<lock_retrieve> performs as C<retrieve>, but also
409 gets a shared lock on the file before reading.
411 Like with any advisory locking scheme, the protection only works if
412 you systematically use C<lock_store> and C<lock_retrieve>. If one
413 side of your application uses C<store> whilst the other uses C<lock_retrieve>,
414 you will get no protection at all.
416 The internal advisory locking is implemented using Perl's flock() routine.
417 If your system does not support any form of flock(), or if you share
418 your files across NFS, you might wish to use other forms of locking by
419 using modules like LockFile::Simple which lock a file using a filesystem
420 entry, instead of locking the file descriptor.
424 The heart of Storable is written in C for decent speed. Extra low-level
425 optimization have been made when manipulating perl internals, to
426 sacrifice encapsulation for the benefit of a greater speed.
428 =head1 CANONICAL REPRESENTATION
430 Normally Storable stores elements of hashes in the order they are
431 stored internally by Perl, i.e. pseudo-randomly. If you set
432 C<$Storable::canonical> to some C<TRUE> value, Storable will store
433 hashes with the elements sorted by their key. This allows you to
434 compare data structures by comparing their frozen representations (or
435 even the compressed frozen representations), which can be useful for
436 creating lookup tables for complicated queries.
438 Canonical order does not imply network order, those are two orthogonal
441 =head1 ERROR REPORTING
443 Storable uses the "exception" paradigm, in that it does not try to workaround
444 failures: if something bad happens, an exception is generated from the
445 caller's perspective (see L<Carp> and C<croak()>). Use eval {} to trap
448 When Storable croaks, it tries to report the error via the C<logcroak()>
449 routine from the C<Log::Agent> package, if it is available.
455 Any class may define hooks that will be called during the serialization
456 and deserialization process on objects that are instances of that class.
457 Those hooks can redefine the way serialization is performed (and therefore,
458 how the symetrical deserialization should be conducted).
460 Since we said earlier:
462 dclone(.) = thaw(freeze(.))
464 everything we say about hooks should also hold for deep cloning. However,
465 hooks get to know whether the operation is a mere serialization, or a cloning.
467 Therefore, when serializing hooks are involved,
469 dclone(.) <> thaw(freeze(.))
471 Well, you could keep them in sync, but there's no guarantee it will always
472 hold on classes somebody else wrote. Besides, there is little to gain in
473 doing so: a serializing hook could only keep one attribute of an object,
474 which is probably not what should happen during a deep cloning of that
477 Here is the hooking interface:
481 =item C<STORABLE_freeze> I<obj>, I<cloning>
483 The serializing hook, called on the object during serialization. It can be
484 inherited, or defined in the class itself, like any other method.
486 Arguments: I<obj> is the object to serialize, I<cloning> is a flag indicating
487 whether we're in a dclone() or a regular serialization via store() or freeze().
489 Returned value: A LIST C<($serialized, $ref1, $ref2, ...)> where $serialized
490 is the serialized form to be used, and the optional $ref1, $ref2, etc... are
491 extra references that you wish to let the Storable engine serialize.
493 At deserialization time, you will be given back the same LIST, but all the
494 extra references will be pointing into the deserialized structure.
496 The B<first time> the hook is hit in a serialization flow, you may have it
497 return an empty list. That will signal the Storable engine to further
498 discard that hook for this class and to therefore revert to the default
499 serialization of the underlying Perl data. The hook will again be normally
500 processed in the next serialization.
502 Unless you know better, serializing hook should always say:
504 sub STORABLE_freeze {
505 my ($self, $cloning) = @_;
506 return if $cloning; # Regular default serialization
510 in order to keep reasonable dclone() semantics.
512 =item C<STORABLE_thaw> I<obj>, I<cloning>, I<serialized>, ...
514 The deserializing hook called on the object during deserialization.
515 But wait. If we're deserializing, there's no object yet... right?
517 Wrong: the Storable engine creates an empty one for you. If you know Eiffel,
518 you can view C<STORABLE_thaw> as an alternate creation routine.
520 This means the hook can be inherited like any other method, and that
521 I<obj> is your blessed reference for this particular instance.
523 The other arguments should look familiar if you know C<STORABLE_freeze>:
524 I<cloning> is true when we're part of a deep clone operation, I<serialized>
525 is the serialized string you returned to the engine in C<STORABLE_freeze>,
526 and there may be an optional list of references, in the same order you gave
527 them at serialization time, pointing to the deserialized objects (which
528 have been processed courtesy of the Storable engine).
530 It is up to you to use these information to populate I<obj> the way you want.
532 Returned value: none.
538 Predicates are not exportable. They must be called by explicitely prefixing
539 them with the Storable package name.
543 =item C<Storable::last_op_in_netorder>
545 The C<Storable::last_op_in_netorder()> predicate will tell you whether
546 network order was used in the last store or retrieve operation. If you
547 don't know how to use this, just forget about it.
549 =item C<Storable::is_storing>
551 Returns true if within a store operation (via STORABLE_freeze hook).
553 =item C<Storable::is_retrieving>
555 Returns true if within a retrieve operation, (via STORABLE_thaw hook).
561 With hooks comes the ability to recurse back to the Storable engine. Indeed,
562 hooks are regular Perl code, and Storable is convenient when it comes to
563 serialize and deserialize things, so why not use it to handle the
564 serialization string?
566 There are a few things you need to know however:
572 You can create endless loops if the things you serialize via freeze()
573 (for instance) point back to the object we're trying to serialize in the hook.
577 Shared references among objects will not stay shared: if we're serializing
578 the list of object [A, C] where both object A and C refer to the SAME object
579 B, and if there is a serializing hook in A that says freeze(B), then when
580 deserializing, we'll get [A', C'] where A' refers to B', but C' refers to D,
581 a deep clone of B'. The topology was not preserved.
585 That's why C<STORABLE_freeze> lets you provide a list of references
586 to serialize. The engine guarantees that those will be serialized in the
587 same context as the other objects, and therefore that shared objects will
590 In the above [A, C] example, the C<STORABLE_freeze> hook could return:
592 ("something", $self->{B})
594 and the B part would be serialized by the engine. In C<STORABLE_thaw>, you
595 would get back the reference to the B' object, deserialized for you.
597 Therefore, recursion should normally be avoided, but is nonetheless supported.
601 There is a new Clone module available on CPAN which implements deep cloning
602 natively, i.e. without freezing to memory and thawing the result. It is
603 aimed to replace Storable's dclone() some day. However, it does not currently
604 support Storable hooks to redefine the way deep cloning is performed.
608 Here are some code samples showing a possible usage of Storable:
610 use Storable qw(store retrieve freeze thaw dclone);
612 %color = ('Blue' => 0.1, 'Red' => 0.8, 'Black' => 0, 'White' => 1);
614 store(\%color, '/tmp/colors') or die "Can't store %a in /tmp/colors!\n";
616 $colref = retrieve('/tmp/colors');
617 die "Unable to retrieve from /tmp/colors!\n" unless defined $colref;
618 printf "Blue is still %lf\n", $colref->{'Blue'};
620 $colref2 = dclone(\%color);
622 $str = freeze(\%color);
623 printf "Serialization of %%color is %d bytes long.\n", length($str);
624 $colref3 = thaw($str);
626 which prints (on my machine):
628 Blue is still 0.100000
629 Serialization of %color is 102 bytes long.
633 If you're using references as keys within your hash tables, you're bound
634 to disapointment when retrieving your data. Indeed, Perl stringifies
635 references used as hash table keys. If you later wish to access the
636 items via another reference stringification (i.e. using the same
637 reference that was used for the key originally to record the value into
638 the hash table), it will work because both references stringify to the
641 It won't work across a C<store> and C<retrieve> operations however, because
642 the addresses in the retrieved objects, which are part of the stringified
643 references, will probably differ from the original addresses. The
644 topology of your structure is preserved, but not hidden semantics
647 On platforms where it matters, be sure to call C<binmode()> on the
648 descriptors that you pass to Storable functions.
650 Storing data canonically that contains large hashes can be
651 significantly slower than storing the same data normally, as
652 temprorary arrays to hold the keys for each hash have to be allocated,
653 populated, sorted and freed. Some tests have shown a halving of the
654 speed of storing -- the exact penalty will depend on the complexity of
655 your data. There is no slowdown on retrieval.
659 You can't store GLOB, CODE, FORMLINE, etc... If you can define
660 semantics for those operations, feel free to enhance Storable so that
661 it can deal with them.
663 The store functions will C<croak> if they run into such references
664 unless you set C<$Storable::forgive_me> to some C<TRUE> value. In that
665 case, the fatal message is turned in a warning and some
666 meaningless string is stored instead.
668 Setting C<$Storable::canonical> may not yield frozen strings that
669 compare equal due to possible stringification of numbers. When the
670 string version of a scalar exists, it is the form stored, therefore
671 if you happen to use your numbers as strings between two freezing
672 operations on the same data structures, you will get different
675 When storing doubles in network order, their value is stored as text.
676 However, you should also not expect non-numeric floating-point values
677 such as infinity and "not a number" to pass successfully through a
678 nstore()/retrieve() pair.
680 As Storable neither knows nor cares about character sets (although it
681 does know that characters may be more than eight bits wide), any difference
682 in the interpretation of character codes between a host and a target
683 system is your problem. In particular, if host and target use different
684 code points to represent the characters used in the text representation
685 of floating-point numbers, you will not be able be able to exchange
686 floating-point data, even with nstore().
690 Thank you to (in chronological order):
692 Jarkko Hietaniemi <jhi@iki.fi>
693 Ulrich Pfeifer <pfeifer@charly.informatik.uni-dortmund.de>
694 Benjamin A. Holzman <bah@ecnvantage.com>
695 Andrew Ford <A.Ford@ford-mason.co.uk>
696 Gisle Aas <gisle@aas.no>
697 Jeff Gresham <gresham_jeffrey@jpmorgan.com>
698 Murray Nesbitt <murray@activestate.com>
699 Marc Lehmann <pcg@opengroup.org>
700 Justin Banks <justinb@wamnet.com>
701 Jarkko Hietaniemi <jhi@iki.fi> (AGAIN, as perl 5.7.0 Pumpkin!)
702 Salvador Ortiz Garcia <sog@msg.com.mx>
703 Dominic Dunlop <domo@computer.org>
704 Erik Haugan <erik@solbors.no>
706 for their bug reports, suggestions and contributions.
708 Benjamin Holzman contributed the tied variable support, Andrew Ford
709 contributed the canonical order for hashes, and Gisle Aas fixed
710 a few misunderstandings of mine regarding the Perl internals,
711 and optimized the emission of "tags" in the output streams by
712 simply counting the objects instead of tagging them (leading to
713 a binary incompatibility for the Storable image starting at version
714 0.6--older images are of course still properly understood).
715 Murray Nesbitt made Storable thread-safe. Marc Lehmann added overloading
716 and reference to tied items support.
720 There is a Japanese translation of this man page available at
721 http://member.nifty.ne.jp/hippo2000/perltips/storable.htm ,
722 courtesy of Kawai, Takanori <kawai@nippon-rad.co.jp>.
726 Raphael Manfredi F<E<lt>Raphael_Manfredi@pobox.comE<gt>>