Fri Jun 7 23:55:41 BST 2002 Nicholas Clark
Version 2.04
Bug fix from Radu Greab <[email protected]> (plus regression test)
to fix a recently introduced bug detected by Dave Rolsky.
Bug was that for a non threaded build, the class information was
being lost at freeze time on the first object with a STORABLE_freeze
hook. Consequentially the object was not blessed at all when thawed.
(The presence (or lack) of STORABLE_thaw was irrelevant; this was
a store-time data lost bug, caused by failure to initialize internal
context)
The bug was introduced as development perl change 16442 (on
2002/05/07), so has been present since 2.00.
Patches to introduce more regression tests to reduce the chance of
a reoccurance of this sort of goof are always welcome.
Thu May 30 20:31:08 BST 2002 Nicholas Clark <[email protected]>
Version 2.03 Header changes on 5.6.x on Unix where IV is long long
5.6.x introduced the ability to have IVs as long long. However,
Configure still defined BYTEORDER based on the size of a long.
Storable uses the BYTEORDER value as part of the header, but
doesn't explicity store sizeof(IV) anywhere in the header.
Hence on 5.6.x built with IV as long long on a platform that
uses Configure (ie most things except VMS and Windows) headers
are identical for the different IV sizes, despite the files
containing some fields based on sizeof(IV)
5.8.0 is consistent; all platforms have BYTEORDER in config.h
based on sizeof(IV) rather than sizeof(long). This means that
the value of BYTEORDER will change from (say) 4321 to 87654321
between 5.6.1 and 5.8.0 built with the same options to Configure
on the same machine. This means that the Storable header will
differ, and the two versions will wrongly thing that they are
incompatible.
For the benefit of long term consistency, Storable now
implements the 5.8.0 BYTEORDER policy on 5.6.x. This means that
2.03 onwards default to be incompatible with 2.02 and earlier
(ie the large 1.0.x installed base) on the same 5.6.x perl.
To allow interworking, a new variable
$Storable::interwork_56_64bit is introduced. It defaults to
false. Set it to true to read and write old format files. Don't
use it unless you have existing stored data written with 5.6.x
that you couldn't otherwise read, or you need to interwork with
a machine running older Storable on a 5.6.x with long long IVs
(i.e., you probably don't need to use it).
Sat May 25 22:38:39 BST 2002 Nicholas Clark <[email protected]>
Version 2.02
Rewrite Storable.xs so that the file header structure for write_magic
is built at compile time, and check_magic attempts to the header in
blocks rather than byte per byte. These changes make the compiled
extension 2.25% smaller, but are not significant enough to give a
noticeable speed up.
Thu May 23 22:50:41 BST 2002 Nicholas Clark <[email protected]>
Version 2.01
- New regression tests integer.t
- Add code to safely store large unsigned integers.
- Change code not to attempt to store large integers (ie > 32 bits)
in network order as 32 bits.
*Never* underestimate the value of a pathological test suite carefully
crafted with maximum malice before writing a line of real code. It
prevents crafty bugs from stowing away in your released code.
It's much less embarrassing to find them before you ship.
(Well, never underestimate it if you ever want to work for me)
Fri May 17 22:48:59 BST 2002 Nicholas Clark <[email protected]>
Version 2.0, binary format 2.5 (but writes format 2.4 on pre 5.7.3)
The perl5 porters have decided to make sure that Storable still
builds on pre-5.8 perls, and make the 5.8 version available on CPAN.
The VERSION is now 2.0, and it passes all tests on 5.005_03, 5.6.1
and 5.6.1 with threads. On 5.6.0 t/downgrade.t fails tests 34 and 37,
due to a bug in 5.6.0 - upgrade to 5.6.1.
Jarkko and I have collated the list of changes the perl5 porters have
from the perl5 Changes file:
- data features of upcoming perl 5.8.0 are supported: Unicode hash
keys (Unicode hash values have been supported since Storable 1.0.1)
and "restricted hashes" (readonly hashes and hash entries)
- a newer version of perl can now be used to serialize data which is
not supported in earlier perls: Storable will attempt to do the
right thing for as long as possible, croaking only when safe data
conversion simply isn't possible. Alternatively earlier perls can
opt to have a lossy downgrade data instead of croaking
- when built with perls pre 5.7.3 this Storable writes out files
with binary format 2.4, the same format as Storable 1.0.8 onwards.
This should mean that this Storable will inter-operate seamlessly
with any Storable 1.0.8 or newer on perls pre 5.7.3
- dclone() now works with empty string scalar objects
- retrieving of large hashes is now more efficient
- more routines autosplit out of the main module, so Storable should
load slightly more quickly
- better documentation
- the internal context objects are now freed explicitly, rather than
relying on thread or process exit
- bugs fixed in debugging trace code affecting builds made with 64 bit
IVs
- code tidy-ups to allow clean compiles with more warning options
turned on avoid problems with $@ getting corrupted on 5.005_03 if
Carp wasn't already loaded
- added &show_file_magic, so you can add to /etc/magic and teach
Unix's file command about Storable files
We plan to keep Storable on CPAN in sync with the Perl core, so
if you encounter bugs or other problems building or using Storable,
please let us know at [email protected]
Patches welcome!
Sat Dec 1 14:37:54 MET 2001 Raphael Manfredi <[email protected]>
This is the LAST maintenance release of the Storable module.
Indeed, Storable is now part of perl 5.8, and will be maintained
as part of Perl. The CPAN module will remain available there
for people running pre-5.8 perls.
Avoid requiring Fcntl upfront, useful to embedded runtimes.
Use an eval {} for testing, instead of making Storable.pm
simply fail its compilation in the BEGIN block.
store_fd() will now correctly autoflush file if needed.
Tue Aug 28 23:53:20 MEST 2001 Raphael Manfredi <[email protected]>
Fixed truncation race with lock_retrieve() in lock_store().
The file has to be truncated only once the exclusive lock is held.
Removed spurious debugging messages in .xs file.
Sun Jul 1 13:27:32 MEST 2001 Raphael Manfredi <[email protected]>
Systematically use "=over 4" for POD linters.
Apparently, POD linters are much stricter than would
otherwise be needed, but that's OK.
Fixed memory corruption on croaks during thaw(). Thanks
to Claudio Garcia for reproducing this bug and providing the
code to exercise it. Added test cases for this bug, adapted
from Claudio's code.
Made code compile cleanly with -Wall (from Jarkko Hietaniemi).
Changed tagnum and classnum from I32 to IV in context. Also
from Jarkko.
Thu Mar 15 01:22:32 MET 2001 Raphael Manfredi <[email protected]>
Last version was wrongly compiling with assertions on, due
to an edit glitch. That did not cause any problem (apart from
a slight performance loss) excepted on Win* platforms, where the
assertion code does not compile.
Sat Feb 17 13:37:37 MET 2001 Raphael Manfredi <[email protected]>
Version 1.0.10.
Forgot to increase version number at previous patch (there were
two of them, which is why we jump from 1.0.8 to 1.0.10).
Sat Feb 17 13:35:00 MET 2001 Raphael Manfredi <[email protected]>
Version 1.0.8, binary format 2.4.
Fixed incorrect error message.
Now bless objects ASAP at retrieve time, which is meant to fix
two bugs:
* Indirect references to overloaded object were not able to
restore overloading if the object was not blessed yet,
which was possible since blessing occurred only after the
recursive retrieval.
* Storable hooks asking for serialization of blessed ref could
get un-blessed refs at retrieval time, for the very same
reason.
The fix implemented here was suggested by Nick Ing-Simmons.
Added support for blessed ref to tied structures. This is the
cause for the binary format change.
Added EBCDIC version of the compatibility test with 0.6.11,
from Peter Prymmer
Added tests for the new features, and to make sure the bugs they
are meant to fix are indeed fixed.
Wed Jan 3 10:43:18 MET 2001 Raphael Manfredi <[email protected]>
Removed spurious 'clean' entry in Makefile.PL.
Added CAN_FLOCK to determine whether we can flock() or not,
by inspecting Perl's configuration parameters, as determined
by Configure.
Trace offending package when overloading cannot be restored
on a scalar.
Made context cleanup safer to avoid dup freeing, mostly in the
presence of repeated exceptions during store/retrieve (which can
cause memory leaks anyway, so it's just additional safety, not a
definite fix).
Sun Nov 5 18:23:48 MET 2000 Raphael Manfredi <[email protected]>
Version 1.0.6.
Fixed severe "object lost" bug for STORABLE_freeze returns,
when refs to lexicals, taken within the hook, were to be
serialized by Storable. Enhanced the t/recurse.t test to
stress hook a little more with refs to lexicals.
Thu Oct 26 19:14:38 MEST 2000 Raphael Manfredi <[email protected]>
Version 1.0.5.
Documented that store() and retrieve() can return undef.
That is, the error reporting is not always made via exceptions,
as the paragraph on error reporting was implying.
Auto requires module of blessed ref when STORABLE_thaw misses.
When the Storable engine looks for the STORABLE_thaw hook and
does not find it, it now tries to require the package into which
the blessed reference is.
Just check $^O, in t/lock.t: there's no need to pull the whole
Config module for that.
Mon Oct 23 20:03:49 MEST 2000 Raphael Manfredi <[email protected]>
Version 1.0.4.
Protected calls to flock() for DOS platform: apparently, the
flock/fcnlt emulation is reported to be broken on that
platform.
Added logcarp emulation if they don't have Log::Agent, since
we now use it to carp when lock_store/lock_retrieve is used
on DOS.
Fri Sep 29 21:52:29 MEST 2000 Raphael Manfredi <[email protected]>
Version 1.0.3.
Avoid using "tainted" and "dirty" since Perl remaps them via
cpp (i.e. #define). This is deeply harmful when threading
is enabled. This concerned both the context structure and
local variable and argument names. Brrr..., scary!
Thu Sep 28 23:46:39 MEST 2000 Raphael Manfredi <[email protected]>
Version 1.0.2.
Fixed spelling in README.
Added lock_store, lock_nstore, and lock_retrieve (advisory locking)
after a proposal from Erik Haugan <[email protected]>.
Perls before 5.004_04 lack newSVpvn, added remapping in XS.
Fixed stupid typo in the t/utf8.t test.
Sun Sep 17 18:51:10 MEST 2000 Raphael Manfredi <[email protected]>
Version 1.0.1, binary format 2.3.
Documented that doubles are stored stringified by nstore().
Added Salvador Ortiz Garcia in CREDITS section, He identified
a bug in the store hooks and proposed the right fix: the class
id was allocated too soon. His bug case was also added to
the regression test suite.
Now only taint retrieved data when source was tainted. A bug
discovered by Marc Lehmann.
Added support for UTF-8 strings, a contribution of Marc Lehmann.
This is normally only activated in post-5.6 perls.
Thu Aug 31 23:06:06 MEST 2000 Raphael Manfredi <[email protected]>
First official release Storable 1.0, for inclusion in perl 5.7.0.
The license scheme is now compatible with Perl's.
Thu Aug 24 01:02:02 MEST 2000 Raphael Manfredi <[email protected]>
ANSI-fied most of the code, preparing for Perl core integration.
The next version of Storable will be 0.8, and will be integrated
into the Perl core (development branch).
Dispatch tables were moved upfront to relieve some compilers,
especially on AIX and Windows platforms.
Merged 64-bit fixes from perl5-porters.
Mon Aug 14 09:22:04 MEST 2000 Raphael Manfredi <[email protected]>
Added a refcnt dec in retrieve_tied_key(): sv_magic() increases
the refcnt on the mg_ptr as well.
Removed spurious dependency to Devel::Peek, which was used for
testing only in t/tied_items.t. Thanks to Conrad Heiney
<[email protected]> for spotting it first.
Sun Aug 13 22:12:59 MEST 2000 Raphael Manfredi <[email protected]>
Marc Lehmann kindly contributed code to add overloading support
and to handle references to tied variables.
Rewrote leading blurb about compatibility to make it clearer what
"backward compatibility" is about: when I say 0.7 is backward
compatible with 0.6, it means the revision 0.7 can read files
produced by 0.6.
Mention new Clone(3) extension in SEE ALSO.
Was wrongly optimizing for "undef" values in hashes by not
fully recursing: as a result, tied "undef" values were incorrectly
serialized.
Sun Jul 30 12:59:17 MEST 2000 Raphael Manfredi <[email protected]>
First revision of Storable 0.7.
The serializing format is new, known as version 2.0. It is fully
backward compatible with 0.6. Earlier formats are deprecated and
have not even been tested: next version will drop pre-0.6 format.
Changes since 0.6@11:
- Moved interface to the "beta" status. Some tiny parts are still
subject to change, but nothing important enough to warrant an "alpha"
status any longer.
- Slightly reduced the size of the Storable image by factorizing
object class names and removing final object storage notification due
to a redesign of the blessed object storing.
- Classes can now redefine how they wish their instances to be serialized
and/or deep cloned. Serializing hooks are written in Perl code.
- The engine is now fully re-entrant.
Sun Apr 2 23:47:50 MEST 2000 Raphael Manfredi <[email protected]>
Added provision to detect more recent binary formats, since
the new upcoming Storable-0.7 will use a different format.
In order to prevent attempting the de-serialization of newer
formats by older versions, I'm adding this now to the 0.6 series.
I'm expecting this revision to be the last of the 0.6 series.
Unless it does not work with perl 5.6, which I don't use yet,
and therefore against which I cannot test.
Wed Mar 29 19:55:21 MEST 2000 Raphael Manfredi <[email protected]>
Added note about format incompatibilities with old versions
(i.e. pre 0.5@9 formats, which cannot be understood as there
was no versionning information in the file by then).
Protect all $@ variables when eval {} used, to avoid corrupting
it when store/retrieve is called within an exception handler.
Mistakenly included "patchlevel.h" instead of <patchlevel.h>,
preventing Perl's patchlevel from being included, which is
needed starting from 5.6.
Tue May 12 09:15:15 METDST 1998 Raphael Manfredi <[email protected]>
Fixed shared "undef" bug in hashes, which did not remain shared
through store/retrieve.
Thu Feb 10 19:48:16 MET 2000 Raphael Manfredi <[email protected]>
added last_op_in_netorder() predicate
documented last_op_in_netorder()
added tests for the new last_op_in_netorder() predicate
Wed Oct 20 19:07:36 MEST 1999 Raphael Manfredi <[email protected]>
Forgot to update VERSION
Tue Oct 19 21:25:02 MEST 1999 Raphael Manfredi <[email protected]>
Added mention of japanese translation for the manual page.
Fixed typo in macro that made threaded code not compilable,
especially on Win32 platforms.
Changed detection of older perls (pre-5.005) by testing PATCHLEVEL
directly instead of relying on internal symbols.
Tue Sep 14 22:13:28 MEST 1999 Raphael Manfredi <[email protected]>
Integrated "thread-safe" patch from Murray Nesbitt.
Note that this may not be very efficient for threaded code,
see comment in the code.
Try to avoid compilation warning on 64-bit CPUs. Can't test it,
since I don't have access to such machines.
Mon Jul 12 14:37:19 METDST 1999 Raphael Manfredi <[email protected]>
changed my e-mail to pobox.
mentionned it is not thread-safe.
updated version number.
uses new internal PL_* naming convention.
Fri Jul 3 13:38:16 METDST 1998 Raphael Manfredi <[email protected]>
Updated benchmark figures due to recent optimizations done in
store(): tagnums are now stored as-is in the hash table, so
no surrounding SV is created. And the "shared keys" mode for
hash table was turned off.
Fixed backward compatibility (wrt 0.5@9) for retrieval of
blessed refs. That old version did something wrong, but the
bugfix prevented correct retrieval of the old format.
Mon Jun 22 11:00:48 METDST 1998 Raphael Manfredi <[email protected]>
Changed benchmark figures.
Adjust refcnt of tied objects after calling sv_magic() to avoid
memory leaks. Contributed by Jeff Gresham.
Fri Jun 12 11:50:04 METDST 1998 Raphael Manfredi <[email protected]>
Added workaround for persistent LVALUE-ness in perl5.004. All
scalars tagged as being an lvalue are handled as if they were
not an lvalue at all. Added test for that LVALUE bug workaround.
Now handles Perl immortal scalars explicitely, by storing &sv_yes
as such, explicitely.
Retrieval of non-immortal undef cannot be shared. Previous
version was over-optimizing by not creating a separate SV for
all undefined scalars seen.
Thu Jun 4 17:21:51 METDST 1998 Raphael Manfredi <[email protected]>
Baseline for Storable-0.6@0.
This version introduces a binary incompatibility in the generated
binary image, which is more compact than older ones by approximatively
15%, depending on the exact degree of sharing in your structures.
The good news is that your older images can still be retrieved with
this version, i.e. backward compatibility is preserved. This version
of Storable can only generate new binaries however.
Another good news is that the retrieval of data structure is
significantly quicker than before, because a Perl array is used
instead of a hash table to keep track of retrieved objects, and
also because the image being smaller, less I/O function calls are
made.
Tue May 12 09:15:15 METDST 1998 Raphael Manfredi <[email protected]>
Version number now got from Storable.pm directly.
Fixed overzealous sv_type() optimization, which would make
Storable fail when faced with an "upgraded" SV to the PVIV
or PVNV kind containing a reference.
Thu Apr 30 15:11:30 METDST 1998 Raphael Manfredi <[email protected]>
Extended the SYNOPSIS section to give quick overview of the
routines and their signature.
Optimized sv_type() to avoid flags checking when not needed, i.e.
when their type makes it impossible for them to be refs or tied.
This slightly increases throughput by a few percents when refs
and tied variables are marginal occurrences in your data.
Stubs for XS now use OutputStream and InputStream file types to
make it work when the given file is actually a socket. Perl
makes a distinction for sockets in its internal I/O structures
by having both a read and a write structure, whereas plain files
share the same one.
Tue Jun 3 09:41:33 METDST 1997 Raphael Manfredi <[email protected]>
Thanks to a contribution from Benjamin A. Holzman, Storable is now
able to correctly serialize tied SVs, i.e. tied arrays, hashes
and scalars.
Thu Apr 9 18:07:51 METDST 1998 Raphael Manfredi <[email protected]>
I said SvPOK() had changed to SvPOKp(), but that was a lie...
Wed Apr 8 13:14:29 METDST 1998 Raphael Manfredi <[email protected]>
Wrote sizeof(SV *) instead of sizeof(I32) when portable, which
in effect mangled the object tags and prevented portability
accross 32/64 bit architectures!
Wed Mar 25 14:57:02 MET 1998 Raphael Manfredi <[email protected]>
Added code example for store_fd() and retrieve_fd() in the
man page, to emphasize that file descriptors must be passed as
globs, not as plain strings.
Cannot use SV addresses as tag when using nstore() on LP64. This
was the cause of problems when creating a storable image on an
LP64 machine and retrieving it on an ILP32 system, which is
exactly what nstore() is meant for...
However, we continue to use SV addresses as tags for plain store(),
because benchamarking shows that it saves up to 8% of the store
time, and store() is meant to be fast at the expense of lack
of portability.
This means there will be approximately an 8% degradation of
performance for nstore(), but it's now working as expected.
That cost may vary on your machine of course, since it is
solely caused by the memory allocation overhead used to create
unique SV tags for each distinct stored SV.
Tue Jan 20 09:21:53 MET 1998 Raphael Manfredi <[email protected]>
Don't use any '_' in version number.
Tue Jan 13 17:51:50 MET 1998 Raphael Manfredi <[email protected]>
Updated version number.
added binmode() calls for systems where it matters.
Be sure to pass globs, not plain file strings, to C routines,
so that Storable can be used under the Perl debugger.
Wed Nov 5 10:53:22 MET 1997 Raphael Manfredi <[email protected]>
Fix memory leaks on seen hash table and returned SV refs.
Storable did not work properly when tainting enabled.
Fixed "Allocation too large" messages in freeze/thaw and added.
proper regression test in t/freeze.t.
Tue Jun 3 09:41:33 METDST 1997 Raphael Manfredi <[email protected]>
Updated version number
Added freeze/thaw interface and dclone.
Fri May 16 10:45:47 METDST 1997 Raphael Manfredi <[email protected]>
Forgot that AutoLoader does not export its own AUTOLOAD.
I could use
use AutoLoader 'AUTOLOAD';
but that would not be backward compatible. So the export is
done by hand...
Tue Mar 25 11:21:32 MET 1997 Raphael Manfredi <[email protected]>
Empty scalar strings are now "defined" at retrieval time.
New test to ensure an empty string is defined when retrieved.
Thu Feb 27 16:32:44 MET 1997 Raphael Manfredi <[email protected]>
Updated version number
Declare VERSION as being used
Fixed a typo in the PerlIO_putc remapping.
PerlIO_read and perlIO_write inverted size/nb_items.
(only relevant for pre-perl5.004 versions)
Thu Feb 27 15:58:31 MET 1997 Raphael Manfredi <[email protected]>
Updated version number
Added VERSION identification
Allow build with perl5.003, which is ante perlIO time
Mon Jan 13 17:53:18 MET 1997 Raphael Manfredi <[email protected]>
Random code fixes.
Wed Jan 22 15:19:56 MET 1997 Raphael Manfredi <[email protected]>
Updated version number in Makefile.PL.
Added "thanks to" section to README.
Documented new forgive_me variable.
Made 64-bit clean.
Added forgive_me support to allow store() of data structures
containing non-storable items like CODE refs.
|
