RT# 50541: Fix for clear bug. Introduces a speed regression

[dbsrgits/DBM-Deep.git] / Changes

Revision history for DBM::Deep.

1.0019_002 Jan XX XX:00:00 2010 EST
    (This is the second developer release for 1.0020.)
    (This version is compatible with 1.0014)
	- Fixed bug where attempting to store a value tied to something other than
      DBM::Deep would leave the file flocked.
	- Added support for DBD::SQLite
        - Build.PL has been extended to support sqlite vs. mysql
        - Storage::DBI now detects between the two DBDs
    - (RT #51888) Applied POD patch (Thanks, FWIE!)
    - (RT #44981) Added VERSION to ::Array, ::Engine, and ::Hash
    - Removed extraneous slashes from POD links (Thanks ilmari!)
    - (RT #50541) Fixed bug in clear() for hashes in the File backend.
        - This has caused a regression in speed for clear() when clearing
          large hashes using running with the File backend. ->clear() (on my
          machine) now takes ( N / 40 ) ** (1.66) seconds. So, clearing 4000
          keys (as is the test in t/03_bighash.t) would take ~2070 seconds.

1.0019_001 Dec 31 22:00:00 2009 EST
    (This is the first developer release for 1.0020.)
    (This version is compatible with 1.0014)
	- DBM::Deep has been refactored to allow for multiple engines. There are two
      engines built so far:
        - File (the original engine)
        - DBI (an engine based on DBI)
    - The DBI engine has only been tested on MySQL and isn't transactional.
        - InnoDB sucks horribly. When run in a sufficient isolation mode, it
          creates deadlocks. 
    - A custom Build.PL has been written to allow for running tests under
      CPAN.pm against the various engines.
        - This also allows running the long tests under CPAN.pm
    - This has meant a ton of refactoring. Hopefullly, this refactoring will
      allow finding some of the niggly bugs more easily. Those tests have not
      been enabled yet. That's the next developer release.
    - Hopefully, this multi-engine support will allow deprecation of the file
      format in the future.

1.0014 Jun 13 23:15:00 2008 EST
    - (This version is compatible with 1.0013)
    - Fix for RT#36781 (t/44 has an unrequired dependency)
    - lock() has been aliased to lock_exclusive(). There is now
      a lock_shared() method. The :flock constants are no longer
      imported into the DBM::Deep namespace.
      **** THIS IS AN API CHANGE ****

1.0013 Jun 13 23:15:00 2008 EST
    - (This version is compatible with 1.0012)
    - Fix for RT#30144 (Optimization failure on Win32)
    - Fixed a bug in reindex_entry (Thanks, Wulfram Humann!)

1.0012 Jun 09 15:00:00 2008 EST
    - (This version is compatible with 1.0011)
    - Fix for RT#30085 (Remove dependency on XS module)
        - Thank you very much tachyon-II@Perlmonks!!
        - This also resolves RT#35424 (DBM::Deep breaks XML::Twig).
    - Updated the POD with fixes that were made, but still on the TODO list.
    - Bypass for RT#36419 (t/44_upgrade_db.t fails on *BSD)
        - We're just going to skip that for now. It's a number-of-processes-open
        issue. Best is to do a port to another module to do the heavy lifting.

1.0011 May 27 15:00:00 2008 EST
    - (This version is compatible with 1.0010)
    - A test has the wrong plan.

1.0010 May 27 12:00:00 2008 EST
    - (This version is compatible with 1.0009)
    - Fix for RT#35140 (invalid POD links)
    - Fix for RT#34819 (Cannot assign the same value back to the same location)
    - Fix for RT#29957 (Cannot assign the same value back to the same location)
    - Fix for RT#33863 (Cannot shift an arrayref from an array)
    - When something is deleted from a DB, the value is export()ed, allowing it
      to be saved.
      - This exporting is only done if the refcount == 0 after the deletion.

1.0009 Mar 19 12:00:00 2008 EDT
    - (This version is compatible with 1.0008)
    - Internal refactorings to prepare for some optimizations.
        - _fh() has been removed. It was marked as private, so don't complain.
    - Skip a test that was spuriously failing on Win32 (Thanks, Alias!)

1.0008 Mar 09 20:00:00 2008 EDT
    - (This version is compatible with 1.0007)
    - Fixed a number of Win32 issues (Reported by Steven Samelson - thank you!)
        - Much thanks to Nigel Sandever and David Golden for their help
          debugging the issues, particularly with DBM::Deep's usage of
          File::Temp (which removes a number of warnings).
        - Autovivification now works on Win32. It turns out that when a
          process takes a shared flock on a file, it's not allowed to write to
          it under Win32, unlike *nix. This is probably a good catch.
            - Note: The fix is a hack. All locks are now exclusive until a
              better fix is found.

1.0007 Jan 10 00:00:00 2008 EDT
    - (This version is compatible with 1.0006)
    - Applied a patch+failing test submitted by sprout@cpan.org. Thanks!
      - Turns out that the case of 17 keys with the same first character in the
        MD5 hash wasn't being tested for. This was a crashbug.
    - A fix has been made to upgrade_db.pl (RT# 30067)
      - The version determinations were in the wrong order or evaluation. This
        meant that upgrade_db.pl wouldn't work as expected (or at all).
    - Added a minimum Pod::Usage requirement (RT# 29976)
      - It's an optional item in Build.PL
      - utils/upgrade_db.pl now checks for that version, as does the test.

1.0006 Oct 01 23:15:00 2007 EDT
    - (This version is compatible with 1.0005)
    - Removed Clone and replaced it with a hand-written datastructure walker.
      - This greatly reduces the footprint of a large import
      - This bypasses a failure of Clone under Perl 5.9.5
      - Moved t/37_delete_edge_cases.t to t_attic because it wasn't really used
    - import() has a stricter API now. This is a potentially incompatible API
      change. Only HASH and ARRAY refs are now allowed and they must match the type
      of the object being imported into.

1.0005 Oct 01 11:15:00 2007 EDT
    - (This version is compatible with 1.0004)
    - Added proper singleton support. This means that the following now works:
        $db->{foo} = [ 1 .. 3];
        my $x = $db->{foo};
        my $y = $db->{foo};
        is( $x, $y ); # Now passes
      - This means that Data::Dumper now properly reports when $db->{foo} = $db->{bar}

1.0004 Sep 28 12:15:00 2007 EDT
    - (This version is compatible with 1.0003)
    - Fixed the Changes file (wrong version was displayed for 1.0003)
    - Added filter sugar methods to be more API-compatible with other DBMs
      - This was added to support a patch provided to IO::All so it can
        use DBM::Deep as a DBM provider.
    - Implemented _dump_file in order to display the file structure. As a
      result, the following bugs were fixed:
      - Arrays and hashes now clean up after themselves better.
      - Bucketlists now clean up after themselves better.
      - Reindexing properly clears the old bucketlist before freeing it.

1.0003 Sep 24 14:00:00 2007 EDT
    - THIS VERSION IS INCOMPATIBLE WITH FILES FROM ALL OTHER PRIOR VERSIONS.
    - Further fixes for unshift/shift/splice and references (RT# 29583)
    - To fix that, I had to put support for real references in.
      - the 16 and 22 tests are now re-enabled.
      - Yes, this means that real references work. See t/45_references.t

1.0002 Sep 20 22:00:00 2007 EDT
    - (This version is compatible with 1.0001)
    - Expanded _throw_error() so that it provides better information.
      (Thanks brian d foy!)
    - Fixed how shift, unshift, and splice work when there are references
      being moved. It now no longer dies.
    - Added diag in t/17_import.t to note that the failing tests on blead
      are due to Clone being broken, not DBM::Deep. The tests will still
      fail because I don't want users to install something that's broken
      and deal with those bug reports.

1.0001 Mar 12 16:15:00 2007 EDT
    - (This version is compatible with 1.0000)
    - Added a missing dependency on IO::Scalar (RT #25387)
    - Fixed how t/44_upgrade_db.t and utils/upgrade_db.pl worked
      - utils/upgrade_db.pl now uses #!perl, not #!/usr/bin/perl
      - t/44_upgrade_db.t now explicitly calls $^X
        (Thanks merlyn!)

1.0000 Feb 26 22:30:00 2007 EST
    - THIS VERSION IS INCOMPATIBLE WITH FILES FROM ALL OTHER PRIOR VERSIONS.
      - To aid in this form of upgrades, DBM::Deep now checks the file format
        version to make sure that it knows how to read it.
      - db_upgrade.pl was added to utils/. This will -NOT- install onto
        your system. This is deliberate.
      - db_upgrade.pl will not handle developer release file formats. This
        is due to the fact that all developer releases in preparation for a
        given release share the same file version, even though the file
        format may change. This is deliberate.
    - Importing no longer takes place within a transaction
    - The following parameters were added:
      - data_sector_size - this determines the default size of a data sector.
    - Correctly handle opening readonly files

0.99_04 Jan 24 22:30:00 2007 EST
    - Added the missing lib/DBM/Deep.pod file to the MANIFEST
    - Fixed a poorly-designed test that was failing depending on what Clone::Any
      was using.
    - All "use 5.6.0;" lines are now "use 5.006_000;" to avoid warnings about
      unsupported vstrings in 5.9.x

0.99_03 Jan 23 22:30:00 2007 EST
    - THIS VERSION IS INCOMPATIBLE WITH FILES FROM ALL OTHER PRIOR VERSIONS.
      - The fileformat changed completely. I will be writing a converter, but
        it's not there right now. Do NOT expect that this module will
        correctly detect older versions and handle them sanely. Sanity will be
        there for 1.00, but we're not there yet, are we?
    - Converted to use FileHandle::Fmode to handle filehandle status checks
    - Fixed bug with deleting already-deleted items on Win32 (reported by Nigel Sandever)
    - The guts of how transactions work has been rewritten to better handle
      some edgecases. This required a complete rewrite of the engine.
    - Freespace management is now in place. It's not perfect, but it's there.
    - The rewrite of the engine required a rewrite of how first_key/next_key
      was implemented. This should result in significant speed improvements.
    - Self-reference has been removed. This means you cannot do:
          $db->{foo} = { x => 'y' };
          $db->{bar} = $db->{foo};
      I hope to be able to return this functionality by 1.00, but I cannot
      promise anything. To do this properly, it requires refcounting in order
      to correctly handle deletions and transactions. Once you move away from
      a simple tree, everything becomes really hard.

0.99_02 Apr 28 05:00:00 2006 Pacific
    - Added missing file to the MANIFEST

0.99_01 Apr 27 18:00:00 2006 Pacific
    - Added explicit dependency on Perl 5.6.0
      - Digest::MD5 requires 5.6.0
      - Sub::Uplevel (dep of Test::Exception) requires 5.6.0
    - Removed error()/clear_error()
      - All error-handling is done with die()
    - Broke out DBM::Deep's code into DBM::Deep::Engine
    - Tied variables can no longer be assigned to a DBM::Deep object.
      - This includes cross-file assignments.
    - Autovivification now works
      - This is a consequence of the fact that all assignments are tied.
    - set_pack() and set_digest() have been removed.
      - Instead, you will now pass the appropriate values into new()
      - A pack_size parameter has been added to make 64-bit files easier
    - Transactions now work

0.983 Apr 10 20:00:00 2006 Pacific
    - Added patch inspired by Jeff Janes (Thanks!)
      - Autovivification now works correctly
      - The following now works correctly
        my %hash = ( a => 1 );
        $db->{hash} = \%hash;
        $hash{b} = 2;
        cmp_ok( $db->{hash}{b}, '==', 2 );
    - (RT#18530) - DBM::Deep now plays nicely with -l

0.982 Mar 08 11:00:00 2006 Pacific
    - Fixed smoketests that were failing on Win32
    - Added restriction for Perl 5.6.0 or higher.
      - Digest::MD5 and Sub::Uplevel (dep of Test::Exception) require 5.6+

0.981 Mar 06 11:00:00 2006 Pacific
    - (RT#17947) - Fixed test that was failing on older Perls

0.98  Feb 28 11:00:00 2006 Pacific
    - Added in patch by David Cantrell to allow use of DATA filehandle
    - Fixed bug where attempting to export() a structure that used autobless would die
    - Fixed arraytest slowness by localizing $SIG{__DIE__} to prevent Test::Builder's
      $SIG{__DIE__} from being called. (q.v. http://perldoc.perl.org/functions/eval.html)
    - More methods have been made private:
        - root() is now _root()
        - base_offset() is now _base_offset()
        - fh() is now _fh()
        - type() is now _type()
        - precalc_sizes() is now _precalc_sizes()

0.97  Feb 24 10:00:00 2006 Pacific
    - Reorganization of distribution to a more standard layout
    - Migration to Module::Build with EU::MM backwards compatibility
    - Migration of all tests to use Test::More and Test::Exception
    - Added Devel::Cover report to DBM::Deep POD
      - Test coverage improved to 89.6% (and climbing)
    - The following methods have been renamed to reflect their private nature:
        - init() is now _init()
        - open() is now _open()
        - close() is now _close()
        - load_tag() is now _load_tag()
        - index_lookup() is now _index_lookup()
        - add_bucket() is now _add_bucket()
        - get_bucket_value() is now _get_bucket_value()
        - delete_bucket() is now _delete_bucket()
        - bucket_exists() is now _bucket_exists()
        - find_bucket_list() is now _find_bucket_list()
        - traverse_index() is now _traverse_index()
        - get_next_key() is now _get_next_key()
        - copy_node() is now _copy_node()
        - throw_error() is now _throw_error()
    - The various tied classes have been broken out. This means that testing
        "ref( $obj ) eq 'DBM::Deep'" will now fail. The correct test is
        "eval { $obj->isa( 'DBM::Deep' ) }".
    - The various methods like push and delete now have the same return values as
      the standard builtins.
    - TIEARRAY and TIEHASH now check their parameters more thoroughly
    - Negative indices for arrays works as expected, including throwing the appropriate
      errors.
    - RT #16877 is fixed (DBM::Deep broken with Perl 5.9.3+).
    - RT #14893 is fixed (tie() and new() use different parameter lists).
    - A bug with optimize and threading is fixed.
    - autobless has received some attention, resulting in a number of bugs fixed.
    - Removed mode option as it just caused confusion.
    - Removed volatile option as it is pretty useless (use locking instead)
    - Locking now implicitly enables autoflush

0.96  Oct 14 09:55:00 2005 Pacific
    - Fixed build (OS X hidden files killed it)
    - You can now pass in an optional filehandle to the constructor

0.95  Oct 12 13:58:00 2005 Pacific
    - Added optional autobless flag to preserve and restore blessed hashes
    - Fixed bug where 0 could not be fetched using get_next_key
    - Fixed bug where tie() constructor didn't accept a hash ref for args
    - optimize() now preserves user/group/permissions
    - Errors are now FATAL (meaning it calls die()), unless you set debug flag

0.94  Apr 13 19:00:26 2004 Pacific
    - Fixed bug reported by John Cardenas (corruption at key level when
         replace of less data was done on bucket)

0.93  Feb 15 19:53:17 2004 Pacific
    - Fixed optmize() on Win32 where orig file couldn't be overwritten unless
         filehandle was closed first.  This change introduces a potential race
         condition when using locking and optmize() on Win32, but it can be 
         fixed in the future using a soft copy instead of Perl's rename().

0.92  Feb 12 19:10:22 2004 Pacific
    - Fixed bug where passing a reference to a different DBM::Deep object
         would still result in an internal reference.
    - Added export() method for recursively extracting hashes/arrays into
         standard in-memory Perl structures.
    - Added import() method for recursively importing existing Perl hash/
         array structures
    - Fixed bug where optimize() wouldn't work if base level of DB was
         an array instead of a hash.

0.91  Feb 12 02:30:22 2004 Pacific
    - Fixed bug with splice() when length of removed section was 0
    - Updated POD re: circular refs and optimize()
    - Had to jump version numbers to 0.91 because in previous releases
         I used only a single digit after the decimal which was confusing
         the CPAN indexer.

0.10  Feb 11 08:58:35 2004 Pacific
    - Fixed bug where default file mode was CLEARING files (Thanks Rich!)
    - Added experimental support for circular references
    - Fixed bugs in shift(), unshift() and splice() where nested objects
      	in array would be recursively re-stored as basic hashes/arrays
    - Fixed typos in POD docs

0.9   Feb 10 03:25:48 2004 Pacific
    - Added Filters for storing/fetching keys/values
    - Added hook for supplying own hashing algorithm
    - FIxed some typos in POD docs, added new sections

0.8   Feb 8 02:38:22 2004 Pacific
    - Renamed to DBM::Deep for CPAN
    - Added optimize() method for rekindling unused space
    - Now returning hybrid tie()/OO object from new()
    - Basic error handling introduced
    - Added debug mode for printing errors to STDERR
    - Added TYPE_HASH and TYPE_ARRAY constants for "type" param
    - Added clone() method for safe copying of objects
    - Wrote POD documentation
    - Added set_pack() function for manipulating LONG_SIZE / LONG_PACK
    - Added aliases for most tied functions for public use
    - Now setting binmode() on FileHandle for Win32
    - Added 45 unit tests

0.7   Jan 4 11:31:50 2003 UTC
    - Renamed to DeepDB
    - Changed file signature to DPDB (not compatible with older versions)
    - Converted array length to packed long instead of sprintf()ed string

0.6   Dec 31 15:12:03 2002 UTC
    - Some misc optimizations for speed

0.5   Oct 18 08:55:29 2002 UTC
    - support for force_return_next parameter in traverse_index() method for 
      ultra-fast combined key search/removal

0.4   Oct 15 20:07:47 2002 UTC
    - now making sure filehandle is open for all DB calls

0.3   Oct 3 19:04:13 2002 UTC
    - fixed bug that could cause corrupted data when using locking

0.2   Aug 6 16:37:32 2002 UTC
    - Removed base index caching, as it can cause problems when two processes 
      are populating the db at the same time (even with locking)

0.1   Jun 3 08:06:26 2002 UTC
    - initial release