Revision history for XML::Comma modules:

1.998  2007/04/12 r.1653

  - various indexing improvements

  - first crack at def extension

  - use Imager instead of GD for CSI.def

  - all comma-*.pl accept an optional -module argument

1.997  2007/02/04 r.1563

  - fix embarrasing bug in

1.996  2007/02/02 r.1557

  - build out of CPAN should work better now

  - require 5.6.1

  - silence a spurious warning from t/comma_standard_image.def

1.995  2007/01/30 r.1541

  - performance: use Proc::Exists instead of Proc::ProcessTable

  - add get_leaf_nodes() and full_field_texts()

  - removal of VirtualDoc and $it->as_array()

1.994  2007/04/30 r.1068

  - another small fix thanks to andreas - use $^X, not perl to call
    a perl script so we get the "right" one

1.993  2007/04/30 r.1064

  - oops, 1.992 was broken

1.992  2007/04/30 r.1063

  - don't die from a Makefile.PL - pointed out by

1.991  2007/04/29 r.1061

  - more install process streamlining:

    - change fallback db to mysql for ease of install

    - don't invoke if we're going to die
      because of missing dependencies, it wastes the user's time

1.99  2007/04/22 r.1043

  - permissions cleanup in svn

  - a few enhancements (still mysql only, unfortunately)

1.98  2007/04/21 r.1028

  - add, introspection tests to MANIFEST

1.97  2007/04/21 r.1026

  - thanks to kwin's help, find the root of the postgres/textsearches
    issues. for now, we have a horrible hackish workaround (force the
    developer not to use defer_textsearches => 1 with rebild)

  - allow a required element to have a value of 0. this is
    important for required booleans, else validate errors out

  - add ->def() in and docs that inherit fromwhence

  - add introspection tests, fuller API expected in 2.1

  - add usage of timestamp macro on element, only useful with
    forthcoming introspection code.

  - add {Store|Index}->def_name, $store->associated_indices()

  - begin work on allowing postgres to be configured from Makefile.PL

1.96  2007/02/26 r.949

  - auto-instantiation shortcut methods for storage iterators

  - in storage iterators, allow validate => [0,1] argument on read_doc,
    next_read, and prev_read

  - bugfix: if a storage iterator has no elements, return a false value
    from _iterator_has_stuff()

1.95  2007/02/23 r.938

  - introduction of VirtualDoc and $it->as_array()

1.94  2007/02/22 r.927

  - more SQL (particularly Pg) bug fixes - ski

  - use Test::More everywhere in t/ - ski

  - introduce $XML::Comma::Log::warn_only, which, when set to
    a true value, dumps Log::* messages to STDERR instead of

1.93  2007/02/13 r.910

  - fix Pg support - khk

  - fix some SQL bugs encountered in testing - ski

1.90  2007/02/01 r.889

  - note: branches/XML-Comma-DEVEL becomes trunk/XML-Comma again

  - Changed inheritance structure of SQL::* modules.

  - Allow operator overloading on storage iterators, ie:
      while(++$sit) { my $doc = $sit->read_doc(); ... };

  - Fix longstanding bug with storage iterators grabbing docs from the
    wrong directory when a store spanned multiple top level directories

  - add date_8 to unix_time macro

  - add no_mtime optional argument to $doc->store()

  - allow slashes in derived_file <location>s

  - added better support for 'binary table' collection type, deprecated
    'many tables' and 'stringified' collection types.

  - add support for lazy doc evaluation from storage iterators, ie:
      my $it = XML::Comma::Def->d->get_index("i")->iterator();
      $it->indexed_field_one(); #fast
      $it->indexed_field_two(); #fast
      $it->unindexed_field();   #slow, but works

  - add Timestamped_random locations for pseudo-random document
    name keyed of system time.

  - fix index_only storage to allow derive_from 

  - add -module flags to,

  - add $iterator->select_count() for efficiently determining the number
    of elements in indexing iterators (mysql support only)

  - fix race conditions in next_sequential_id pointed out by Bill

  - fix tmp blob copy/move bug

1.25  2007/01/31 r.869

  - Performance, correctness/security fix in XML::Comma::Log

  - new(file => ...), new(block => ...), read(), and retrieve()
    now validate by default, so you can't get an invalid doc.
    override via validate => [0,1] argument or for system-wide
    default, set validate_new in

  - added include_for_hash, which is like ignore_for_hash but
    assumes the default is to exclude, not include an element
    from the hash. handy for defs under active development, or
    defs that must be "future-proofed"

  - bugfix in with aborted transfers

1.21  2004/01/12

  - Added "mysql_local_infile=1" to the default mysql DSN.  Kindly 
    suggested by Brian Szymanski when the FAQ workarounds didn't work

1.20  2003/10/13

  - Fixed a textsearch bug (reported by Eric Folley) that allowed 
    stale textsearch_temp_tables to stick around through 
    iterator_refresh calls which caused the not so helpful "error 124"
    in some versions of mysql.  The original fix introduced a new bug
    (reported by Chuq Yang) that caused iterator() calls which used a 
    textsearch that contained no matches to return an iterator full
    of all the documents in the index queried (Dug).

  - Fixed Index_Only store to throw a store error if its indexing
    operation dies.

1.19  2003/05/22

  - Added doc_id to error messages in basic doc methods

  - Made the internal calls to def-level escape_code and unescape_code
    subs pass any set() or get() args, so that escape and unescape
    routines can vary depending on application code arguments

  - Fixed a read_only bug that made newly-created elements of
    read-only docs NOT read-only

  - Fixed textsearch bug wherein textsearch elements with the same
    name in different indexes/defs became mixed up

  - Added XML::Comma::Def-><name> shortcut method

  - Added Indexing iterator method iterator_dispatch, decoupling the autoload 
    from the the method/element fetch.

  - Moved pnotes handling up an (OO) layer into AbstractElement, so
    all parts of a doc or def can now have their own pnotes objects

  - Proc::ProcessTable is now optional, used if available to try and
    determine whether a (local) process that holds doc locks is still

  - Comma/Pkg/Mason/ParResolver and ParComponent added to core

  - Removed some stopwords from Comma/Pkg/Mason/Preprocessor_En that
    never should have been there in the first place (like "internet").

  - Added STDERR output to error notices from $index->rebuild()

  - New utility bin/

  - New Comma/Storage/Output module (that, confusingly, only does
    input): MailMessageReader
  - Made errors thrown during parsing conform to Log->err style, and
    added $extra_text field to Error constructor. Now errors thrown
    during def method parsing report the method name and have a more
    specific heirarchy field.

  - Removed date_8.macro's Date::Calc dependency (rewriting the
    validate_hook in the process) and added a number of tests (Dug).

  - Changed default Comma/Configuration file to use system directories
    relative to the current directory under TEST/ for all test
    scripts. Tried and failed to add a SQLite database driver so that
    we can install (for example from CPAN) even on a vanilla system
    and hae everything work without needing to configure an
    RDBMS. SQLite presents various issues that will be difficult to
    resolve without a large amount of rewritage.

1.18  2003/03/18

  - Fixed Sequential_dir and Sequential_file to pad with the 0th digit
    rather than with '0'

  - Added new bin/ utility

  - Improved some SimpleC error messages

  - Removed about a dozen stopwords from the english stopwords list

  - Added Storage::Util method to docs

  - Fixed a bug in escape/unescape code that was producing strange character 

  - Added experimental "collection field" features

  - Added a pnotes functionality to elements -- a hash-keyed store for
    any data that the element object would like to salt away for
    itself. not shared in a class-like manner as def_pnotes
    is. Changed the generic level def_pnotes lookup method to
    be called def_pnotes rather than pnotes, to avoid confusion now
    that both instance- and class-like pnotes calls exist.

  - Made Proc::ProcessTable optional. Still need to write docs (and
    perhaps a Makefile.PL notice) saying that its useful to have.

1.17  2003/03/04

  - Fixed missing file in MANIFEST

  - ParResolver is working; added Comma/Pkg/test.par

1.10  2002/12/06

  - Fixed bug in textsearch code that prevented processes that hadn't
    already "indexed" to "retrieve" based on textsearch stuff

  - Added 'digits' argument to Sequential_file and
    Sequential_dir. Modified FileUtil::current_sequential_id to use
    the lockfile rather than listing the files in the directory (which
    fixes the odd behavior Sherrard was complaining about that arises
    if you drop unexpected stuff into directories). New dependency:

  - Made some changes to the core to support the new Index_Only
    storage location module. This is the first location module that is
    truly decoupled from the filesystem. There are some performance
    and feature improvements still to be made, but it seems quite

  - Added new <escapes> syntax to allow Def-defined escape/unescape

  - Moved Configuration stuff into Comma/ -- a
    separate file. This makes more sense from an architectural point
    of view, and could make it easier to run more than one Comma
    instance on a given machine, in the future.

  - Added a "sys_directory" configuration variable; changed the
    SimpleC parser to do its build in the sys directory. Changed to create the comma_root, document_root, sys_directory
    and tmp_directory dirs on startup, if they don't already exist.

  - Rewrote BlobElement framework so that BlobElements act like other
    elements: they are not written to permanent storage until a
    store() is called, and there is no extra store() fanciness
    required to make sure that the pointers are in sync. Under the
    covers, blob elements are saved in files under the tmp_directory
    until the doc they belong to is store()d.

  - The content of an <erase_where_clause> now gets eval'ed before
    being passed to the database *if* it begins with a '{'
    character. This is useful for doing perl-ish calculations before
    having to drop down into SQL.

    now uses a much simpler syntax and a future-flexible approach,
    inheriting from

  - Added #include foo and #include {foo} <args list> to parsers

  - Added PAR resolving functionality to DefManager

1.09  2002/11/27 12:36:38

  - Upgraded Pg-specific code for Postgres 7.3. Comma now REQUIRES Pg
    7.3 -- which is fine, because Comma made poor assumptions pre-7.3
    that didn't work so good. QUESTION: why is postgres so chatty, and
    what can be done to keep it from printing out warnings when
    indexes are created, and to stop errors that are caught with eval
    from going to the log.

1.08  2002/07/26 12:40:45

  - Bug fixes or code cleanup in: method_names, _make_collection_spec,
    Index->DESTROY, Element->method() and friends, boolean macro,
    various tests, date_8 (code contributed by Chuck), macro section
    in Bootstrap Def, bcollection tables, Index/Clean circular
    references, Doc->store(), set(), decl_pos in various Location
    modules, FileUtile make_directory permissions setting.

  - DEPRECATED HTTP_Upload. The plan is to support HTTP_Transfer,
    which is a cleaner architecture, instead. Be that as it may, fixed
    some headers in HTTP_Upload_mod_perl module. Tests for
    HTTP_Transfer are in t/transfer.manual.

  - New date_8 macro methods "today" and "diff_today"

  - Makefile.PL now respects existing installs -- prints out
    message saying that it's not overwriting your if that's
    the case. Also, moved VERSION into, so that upgrades
    show the right version even if doesn't change.

  - Changed enum_choices() in enum macro to guarantee a specific
    return order.

  - Added integer macro.

  - DEPRECATED the XML::Comma->lock_singlet() interface. It's close to
    impossible to use this as intended, across differing database
    architectures. Better to rip it out (advisorily, anyway, by
    deprecating) and replace it with something better in the
    future. In the meantime, added a MySQL-specific "disposable locks"
    module in XML::Comma::Pkg::MySQL_Simple_Lock.

  - Beefed up SQL::DBH_User _connect() routine (which is inherited by
    all core code that needs db access). Sherrard wrote code to retry
    and log failed attempts.

  - TIB contributed new Location module SequentialCheck_file

  - Fixed Blob class such that element_delete() deletes the backing

  - Big fixes to doc-level locking code. Old code was not
    multi-machine safe. New code is, as long as the two (or more)
    machines share a single database server. It is possible (though
    unlikely -- requires bad perl interpreter exit) for a lock record
    to hang around after process goes away. There is no probable fix
    for this, but some utility scripts to print out info about comma
    locks are on the todo list.

  - Added a third passed argument to pre_ and post_store_hook code: a
    hashref pointing to the original store() args. This allows hooks
    to do arbritrarily complex iff'ing based on user-level args (which
    are unconstrained, so this is a very open-ended mechanism).

  - NON-BACKWARDS-COMPATIBLE CHANGE: Added a check for defined-ness
    for the element names passed into $el->elements(). This might
    break sloppy (or buggy) old code, but seemed like very much the
    right thing to do. There's no good reason to be able to pass
    non-existent element names to the elements() method.

  - NON-BACKWARDS-COMPATIBLE CHANGE: Changed the third argument passed
    to set_hook and set_from_file_hook to be a hash_ref, rather than
    passing the whole hash. This might break some existing code
    (although I don't think much code uses this feature, yet), but
    it's the Right Thing (tm).

    Comma::Util::XML_basic_escape() method. This method used to try to
    identify '&' characters that were part of an entity tag, and NOT
    escape them. Because the logic was not symmetrical with
    XML_basic_unescape(), confusing results resulted. New escape
    method escapes all occurrences of '<', '>', and '&' in a string --
    making it entirely symmetrical with the unescape code. Two new
    methods were added, in case folks need the old functionality:
    XML_smart_escape() and XML_bare_amp_escape().

  - Added French Textsearch Preprocessor. Algorithm based on snowball
    code from

  - Added 'erase_where_clause' capabilities to collection table

  - Error reporting on errors passed through DBI is much improved. All
    do() statements in SQL/ where changed to prepare(), etc.

  - Partial rewrite of Preprocessor code structure. Most importantly,
    maximum word length is now tied to a global defined in the
    Preprocessor abstract/template class, and the multi-preproc
    switching code accomodates an "attribute" argument so the Iterator
    can stem keyword arguments correctly. UPGRADE INFORMATION: if you
    are upgrading from an earlier version of Comma, it is best to drop
    all indexes that include <textsearch> fields, and rebuild them
    after the upgrade.

1.01  2002/04/10 12:05:32

  - Changed Comma_Standard_Image.def to use hex (rather than b64)
    digests, to eliminate the +/= characters that could cause problems
    in urls.

  - Added doc_id() method to Storage::Iterator. Returns the doc_id of
    the currently-pointed-at doc.

  - Improved $index->rebuild() code. Fixed bug in stop_rebuild_hook
    handler, added parallelization ('workers=>n' arg). Added sig
    handler to catch Ctrl-C and try to clean up, so as to not leave db
    tables in a "marked" state.

  - Added Methodable::method_names() method so that Def and Index
    objects can return the names of the methods they define.
  - Added $def->index_names() and $def->store_names() methods.

  - Fixed Index::Iterator bug found by TIB that was triggered by
    trying to access columns from an empty iterator.

  - Rewrote Index plurals code. Added a new type of storage
    ('bcollection table'), and integrated 'sort' and 'collection'
    types together. Rewrote SQL code generation and all t/indexing.t
    tests. Added more boolean/notting and partial match capabilities
    into the new collections framework.

  - Added some cleanup code to Index->DESTROY to resolve circular
    references that were causing occasional problems with global

1.00  2002/01/28 03:46:15
  - boolean macro bugfix

  - new features in HTTP_Transfer

  - faster $index->count() SQL
  - added iterator_select_returnval() method to Index Iterator class

  - new boolean- (and grouping-) capable collection and sort
    specs. implemented using Parse::RecDescent -- core code now
    requires Parse::RecDescent to be available

  - added validate() method to Element, NestedElement and BlobElement
    classes. validate_structure() is now DEPRECATED. storage write
    methods all call validate() and refuse to store Docs that fail to
    pass all validity checks. it's now possible to define
    validate_hooks for blob elements.