ICU term lists are handled for search.

[idzebra-moved-to-github.git] / ONEWS

$Id: ONEWS,v 1.1 2007-01-22 11:02:12 mike Exp $

Archived news for pre-2.0.0 Zebra.  The Zebra source code was forked
after release 1.3.16: development on the 1.3.x branch continuing as
documented in the post-1.3.16 entries in this file.  In the mean time,
the other branch was initially known as the 1.4.x branch and
eventually became the 2.0.x branch.  Release news subsequent to 2.0.0
is in the NEWS file.  During the period between the fork at 1.3.16 and
the formal release of 2.0.0, some but not all of the subsequent 1.3.x
changes were also made in the 1.4.x (= 2.0.x) branch, so some but not
all of the post-1.3.16 changes listed in this file also apply to
2.0.0.  One day we'll annotate which do and don't apply.



Update code to use new YAZ log functions/defines.

--- 1.3.48 2006/12/22

Updated to use YLOG_-defines for YAZ, rather than the obsolte LOG-defines.

--- 1.3.46 2006/10/31

Fixed several compilation warnings. (gcc 4.1.2, -O3 -g -Wall)

Fixed bug #710: Duplicate keys for CDATA in xelm/melm rules.

--- 1.3.44 2006/10/23

Fixed bug #529: multiple simultaneous updates by extended services trashes
register files.

--- 1.3.42 2006/10/16

Fixed bug #47: Commit needs to check for roll-back.

Fixed bug #672: Trailing characters in password are ignored

Added extra presence check for tcl.h, because some systems have
tclConfig.sh installed even though Tcl C headers are missing.

--- 1.3.40 2006/10/04

Optimized melm performance.

Do not use sync(2) during commit (but rely on sync'd individual files).

--- 1.3.38 2006/08/14

Updated Debian package to use sym links for zebraidx, zebrasrv.

Fixed bug in isamb (isam:b) regarding ISAM tree splitting.

Fixed problem with file locking during commit phase. This error was
present in 1.3.36.

--- 1.3.36 2006/06/07

Fixed bug #602: Any extra terms appended to a wordlist with a hitcount
of 0, gets incorrect term count

Fixed bug #597: Support null missing key for sort.

--- 1.3.34 2006/02/23

Fixed bug #465: dup fields in ISO2709 retrieval.

Another fix for X-Path indexing. The previous fix, unfortunately, indexed
too much. Bug #460.
  
--- 1.3.32 2005/11/08

Fix X-Path attribute indexing. Bug #431.

--- 1.3.30 2005/10/27

Fixed bug #415: Strange truncation behavior.

Added 'melm' directive to absyn format to simplify config files
for MARC-style databases. See tab/marc21.abs for an example.

Added support for special slement set _sysno_ which returns a
record ID for a record packed as a SUTRS record.

--- 1.3.28 2005/05/30

Documented authentication facility in Zebra. Added zebra.cfg directive
'passwd.c' which specifies user accounts file with encrypted passwords. The
directive 'passwd' specifies user accounts file with clear-text passwords.
The previous version of Zebra used plain/clear text depending on
configuration automatically. That caused upgrade trouble. Bug #356.

--- 1.3.26 2005/05/24

Depend on YAZ 2.0.18 or later in configure.

Fixed crash that could occur if ES update transaction failed.

Configure enables the use of the crypt API - if available.

Fixed bug #304: Fuzzy search regExpr-2 did not use proper error distance
of 1.

Fixed bug #305: Scan now handles negative preferred-position-in-response.

Implemented the 'equivalent' directive for .chr-files.

--- 1.3.24 2005/02/09

For configure, support threading again. It was removed by mistake
for 1.3.18-1.3.22.

Fixed bug #262: spaces in control fields in MARC returned.

Fixed bug #259: Second indicator lost in MARC records

--- 1.3.22 2005/01/23

Fixed bug #253: Setting group.database not honored

Fixed bug #252: Sort does not work.

Fixed bug #248: hit counts in combinatoric (and) searches in specific ..

--- 1.3.20 2005/01/17

Fixed bug #245: Time for getting records changes a lot based on record
position.

Fixed bug #235: weird x-path results.

Avoid crash in ISAMB when isamb_pp_open gets ISAMB_POS = 0. In
this case EOF (no entries) is signalled. It fixes problem with
terms being deleted. See bug #109.

Fixed bug #169: Phrase term counts does not work. The bug exists in
Zebra 1.3.16-1.3.18.

Added mechanism to ignore leading articles when doing full-field indexing,
based on the character map files. See the manual for further discussion.

--- 1.3.18 2004/08/20

Fixed bug in record management. Releasing blocks could result in
partial read.

Fixed bug in isam:b. A tree split could result in a lost item.

--- 1.3.17 2004/08/17

Add IDZebra.i to dist so that Perl extension builds again.

--- 1.3.16 2004/08/16

Added facility to make attibutes in grs.regx and grs.tcl filter using the
data command with argument -attribute <name> . The content of data is
the value of the attribute. This command should be used inside a
begin element , end element section.

Update zebra.nsi to NSIS 2.

Added a new 'cut' directive to charmaps (.chr files) which specifies that
only characters after the cutting char should be indexed.

Update Perl internals so that it matches the current Zebra API.
The recordGroup structure is no longer available. A group of resources
can still be referenced by setting groupName=>..  in various methods.

Maximum number of records to be sorted in a result set can be
specified by setting "sortmax". Default is 1000.

Allow use of string use attributes for regular attribute sets. The
name matches the name given in the attribute set file. All strings
starting with / are considered X-Path as usual.

Fixed bug in grs.regx. filter . 'end element' could pop off top tag
element for XML tree. It may only pop off if -record is given.

Added grs.danbib filter - for Danish Bibliographic Centre.

Rename CHANGELOG to NEWS.

For text filter, return only header if elementSetName=H . elementSetName=R
returns contents only. Other elementSetName returns both header+content.

Added test for charmap and rusmarc.

Added feature charmaps (.chr) so that characters may be specified in
\LXXXX HEX notation.

Fixed problem with encoding directive for charmap(.chr) files.

Allow Remote insert/delete/replace/update with record, recordIdNumber
(sysno) and/or recordIdOpaque(user supplied record Id). If both
IDs are omitted internal record ID match is assumed (recordId: - in
zebra cfg).

--- 1.3.15 2004/01/15

Fix bug. X-Path attribute expressions with spaces in them now works.

Fix base address for MARC output.

--- 1.3.14 2003/11/29

Fix bug with shadow and result set handling.

Implement MARCXML to ISO2709 conversion.

--- 1.3.13 2003/09/26

Add missing examples for Windows install.

Fix bug in regx filter to make it "greedy" again. This bug appeared
in version 1.3.12.

Fix a few tests.

--- 1.3.12 2003/09/08

Fix XML error handling. Stop XML parse immediately if XML parse error
occur (i.e.  produce one error only).

Zebra ignores "unsupported use attribute" for individual databases
when search multiple databases (unless all databases fail).

New filter grs.marcxml which works like grs.marc but produces MARCXML.

Added support for database deletion. It is possible to create/drop
a database from zebraidx utility. Note: only for isam:b.

Write zebrasrv.pid to lockdir.

Bug fix: result sets were not recovered correctly. Had to
add ODR handle for zebra_search_RPN in order to make it work.

Fixed a bug in regx filters that didn't do anchors (^) correctly.

Fixed a bug in searches with X-Path searches sometimes giving "extra"
hits.

Zebra server checks for zebrasrv.pid and refuses to start if it is already
locked by another (running) zebrasrv.

Fixed a bug with text being chunked in pieces for the grs.xml filter.

--- 1.3.11 2003/04/25

xelm code updates. xelm works regardless state of 'xpath enable/disable'
Avoid -L/usr/lib since that is already default library path.

Allow multiple updates within one transaction.

Fixed a bug with >2GB files (overflow in integer expression).

--- 1.3.10 2003/04/01

Fix linker error for Perl module.

Fix bug in and operation which in some cases could result in "extra"
hits. Bug was introduced in 1.3.5.

Fix bug in handling of schema conversion when producing numeric tags.

--- 1.3.9 2003/03/27

Zvrank updates. 

Add missing files doc/zvrank.txt and doc/marc_indexing.xml.

--- 1.3.8 2003/03/26

Zvrank: an experimental ranking algorithm. See doc/zvrank.txt and
source in index/zvrank.c. Enable this by using rank: zvrank in zebra.cfg.
Contributed by Johannes Leveling <Johannes.Leveling at fernuni-hagen.de>

livrank: another experimental ranking algorithm. Source in livcode.c.
Enable this by using rank: livrank in zebra.cfg and use -DLIV_CODE=1
for CFLAGS.
Contributed by Pete Mallinson, University of Liverpool. 

Advanced MARC indexing. See doc/marc_indexing.xml
 Oleg Kolobov <oleg at lib.tpu.ru>

Perl API updates and fixes. 
 Peter Popovics <pop at technomat.hu>

Fixed 'zebraidx delete'.

Implemented 'zebraidx clean'.

64-bit offsets for register files on WIN32 (no 2 GB limit).

Fixed a few memory leaks WRT sorting.

--- 1.3.7 2003/02/27

Fixed error handling : error code was not properly returned.

Support Truncation 104 (CCL).

--- 1.3.6 2003/02/25

Added missing source files for perl extension.

--- 1.3.5 2003/02/23

Implemented xelm directive.

Updated for newer version of YAZ (introduction of string schema).

Directory examples/zthes now part of distribution (was missing
in previous release).

New .abs directive, systag, that control where to put retrieval
information. The directive takes two arguments: system tag, element name.
System tag is one of : rank, sysno, size.

--- 1.3.4 2002/11/26

Perl Filter and Perl API. By Peter Popovics.

For zebra.cfg, if no profilePath is specified, directory
 (prefix)/share/idzebra/tab
is used.

Zebra Examples in examples . Zebra tests in test.

Bug fix: sort index was not properly modified on 
record updates/deletes.

Fix handling of character entities for sgml filter.

Move data1 to Zebra (used to be part of YAZ).

--- 1.3.3 2002/10/05

Fix character encoding of scan response terms.

Fix character decoding of scan request terms.

Fix ESpec handling (requires YAZ 1.9.1)

Fix searches for complete fields.

--- 1.3.2 2002/09/09

When name zebra is used in a filename or directory 'idzebra' is used
instead to avoid confusion with GNU zebra (routing software).

Zebra server stops with a fatal error if config file cannot be read.

New config setting, followLinks, that controls whether update of files
should follow symbolic. Set it to 1 (for enable) or 0 (to disable).
By default symbolic links are followed.

Fix MARC transfer . MARC fields had wrong data for multiple fields.

XML record reader moved from YAZ to Zebra, to make YAZ less 
dependant on external libraries.

Zebra uses yaz_iconv which is mini iconv library supporting UTF-8,
UCS4, ISO-8859-1. This means that Zebra does UNICODE even
on systems that doesn't offer iconv.

XML record reader supports external system entities.

--- 1.3.1 2002/08/20

New .abs-directive "xpath" that takes one argument: "enable"
or "disable" to enable and disable XPath -indexing. If no "xpath"
direcive is found in .abs-file , XPath-indexing is disabled to ensure
backwards compatibility. For missing .abs-files XPath-indexing is
enabled so that such records are searchable.

Zebra warns about missing .abs-file only once (for each type).

Fixed a bug in file update where already-inserted files could
be treated as "new".

--- 1.3.0 2002/08/05

Zebra license changed to GNU GPL.

XPath-like queries used when RPN string attributes are used, eg.
   @attr 1=/portal/title sometitle
   @attr 1=/portal/title[@xml:lang=da] danishtitle
   @attr 1=/portal/title/@xml:lang da
   @attr 1=//title sometitle

Zebra uses UTF-8 internally:
1) New setting "encoding" for zebra.cfg that specifies encoding for
OCTET terms in queries and record encoding for most transfer syntaxes
(except those that use International Strings, such as GRS-1).
2) The encoding of International strings is UTF-8 by default. It
may be changed by character set negotiation. If character set
negotiation is in effect and if records are selected for conversion
these'll be converted to the selected character set - thus overriding
the encoding setting in zebra.cfg.
3) New directive "encoding" in .abs-files. This specifies the external
character encoding for files indexed by zebra. However, if records
themselves have an XML header that specifies and encoding that'll be used
instead.

XML filter (-t grs.xml).

Multiple registers. New setting in resource 'root' that holds base
directory for register(s). A group a databases may be put in separate
register in directory root/reg by using db name 'reg/db1' ... 'reg/dbN'.

--- 1.1.1 2002/03/21

Fixes for Digital Unix

Implemented hits per term using USR:SearchResult-1.

New Zebra API. Locking system re-implemented.

--- 1.1.stable 2002/02/20

Rank weight can be controlled with attribute type 9. Default
value is 34. Recommended values between 1-36.

--- 1.1 2001/10/25

Updated for YAZ version 1.8.

Added support for termsets - a result set of terms matching
a given query. For @attr 8=<set> creates termset named <set>.

Added support for raw retrieval. Element Set Name R forces the
text filter which returns the record in its original form.

Added numerical sort - triggered by structure=numeric (4=109).

Remote record import using Z39.50 Extended Services and Segments.

Fixed bug where updating a database with user-defined attributes
could corrupt the register (bad storeKeys).

Multi-threaded version.

Fixed bug regarding proximity.

Documentation updates.

Fixed bug in record retrieval module that occured on 64-bit OSF 
architectures.

--- 1.0.1 2000/2/10

Fixed bug in makefile for WIN32.

Fixed bug in configure script - used bash-specific features.

--- 1.0 1999/12/10

Added support for multiple records in one file for filter grs.sgml.

Changed record index structure. New layout is incompatible with
previous releases. Added setting "recordcompression" to control
compression of records. Possible values are "none" (no
compression) and bzip2 (compression using libbz2).

Added XML transfer syntax support for retrieval of structured records.
Schema in CompSpec is recognised in retrieval of structured records.

Changed Tcl record filter so that it attemps to read  <filt>.tflt. If
that fails, the filter reads the file <filt>.flt (regx style filter).

Implemented new Tcl record filter -  use grs.tcl.<filter> to enable it.
Zebra's configure script automatically attempts to locate Tcl. For
manual Tcl configuration use option --with-tclconfig=<path> to specify
where Tcl's library files are located.

Implemented "compression" of Dictionary and ISAM system. Dictionary
format HAS changed.

Added "tagsysno" directive to zebra.cfg to control under which tag the
system ID is placed. Use tagsysno: 0 to disable Zebra's system number
entirely.

Added "tagrank" as above.

Changed file naming scheme for register files from <name>.mf.<no> to
<name>-<no>.mf.

Implemented "position"-flag for register type (as defined in
default.idx). When set to zero no position (or seqence number) is
saved in register for each word occurrence, thus saving some register
space.

Implemented database mapping. Using mapdb one can specify a database
to be mapped to one or more physical databases. Usage:
mapdb <fromdb> <todb> ..

Added SOIF-filter. Thanks to Peter Valkenburg.

For the regx-filter "end element -record" may trigger a mark-of-record
if outer level is reached.

Tag sets may be typed in the reference to it. From the .abs-file the
"tagset" directive takes a third optional integer type for the tag set
referenced. From a .tag-file the "include" directive takes a third
optional type as well. The old "type" directive in the tag set itself
is still recognized but acts as the default type for the tag set.
 
Zebra supports the specification of arbitrary attributes sets, schemas
and tag sets, because of the change in YAZ' OID management system.

Fixed bug in Sort that caused it NOT to use character mapping as it
should.

Zebra now uses GNU configure to generate Makefile(s).

Added un-optimised support for left and left/right truncation attributes.

Added support for relational operators on text when using RPN queries.

Added support for sort specifications in RPN queries. Type 7 specifies
'sort' where value 1=ascending, value 2=descending. The use attribute
specifies the field criteria as usual.  The term specifies priority
where 0=first, 1=second, ...

Changed the way use attributes are specified in the recordId
specification.

Maximum number of databases in one Zebra register increased.

New setting, databasePath, which specifies that first directory during
update traversal is the database name (instead of a fixed one).

New setting, explainDatabase, which specifies that databases are
EXPLAIN aware.

Modified Zebra so that it works with ASN.1 compiled code for YAZ.

Implemented EXPLAIN database maintenance. Zebra automatically
generate - and update CategoryList, TargetInfo, DatabaseInfo,
AttributeSetInfo and AttributeDetails records at this stage. The
records may be transferred as GRS-1, SUTRS or Explain.

Fixed register spec so that colon isn't treated as size separator
unless followed by [0-9+-] in order to allow DOS drive specifications.

Fixed two bugs in ISAMC system.

Changed the way Zebra keeps its maintenance information about attribute
sets, available attributes, etc.. Records in "SGML" notation using an
EXPLAIN schema is now used when appropriate.

Bug fix: Index didn't handle update/insert/delete of the same record
(i.e. same recordId) in one run (one invocation of zebraidx). Only the
first occurence of a record is considered.

Most searches now return correct number of hits.

New modular ranking system. Interested programmers are encouraged to
inspect rank1.c and improve the algorithm.

Bug fix: Lock files weren't removed as they should on NT.

Implemented Z39.50 Sort. Zebra's sort handler uses use attributes to
specify a "sort register". Refer to the gils sample records which refer
to index type "s" which is specified as "sort" in the default.idx file.
Each sort criteria can either be Ascending or Descending and at most
three sort elements can be specified.  

Bug fix: Character mapping didn't work for text files.

--- 1.0b1 1998/1/29

Simple ranked searches now return correct number of hits.

The test option (-s) only makes a read-lock on the index as well
as using read-only operations anywhere.

Moved towards generic character mapping. Configuration file default.idx
specifies character map files for register types w, p, u, etc.

Implemented "begin variant" for the sgml.regx - filter.

Fixed a few memory leaks.

Added support for C++, headers uses extern "C" for public definitions.

Bug fix: The show records facility (-s) only displayed information for
the first record in a file (and not for every record in the file).

Added option "-f <n>" to limit the logging of record operations. After
<n> records has been processed no logging is performed (unless errors
occur).

Bug fix: the compressed ISAM system didn't handle update operations
correctly.

Added setting, "maxResultSetSize", to hold the number of records to 
save in a result set.

Bug fix: Complete phrase did't work for search operations.

Bug fix: temporary result sets weren't deleted.

Reduced disk space for saved keys (storeKeys = 1).

Added optional, physical ANY (key replication)

Implemented proximity operator in search.

Bug fix: the path name buffers used by file match traversal routines
have been extended to support long file names.

New C(ompressed) ISAM system. To enable it, specify "isam: c" in the
configuration file. The resulting register without "storeKeys" is about
half the size, and the memory used by zebraidx during phase 2 (merge) is
reduced to a minimum.

Reworked the way Regexp-2 queries with error tolerance are handled and
specified. The documentation has been updated accordingly.

Bug fix: Zebrasrv didn't search correctly when queries contained masking
characters. This bug was introduced in 1.0a8.

Zebrasrv now tag records with the proper database name.

New settings, memMax and keyTmpDir.

Changed name of setting lockDir (previously called lockPath) and
setTmpDir (previously called tempSetPath).

Generalized and changed record type specifications. In short, there are:
       text                plain SUTRS
       grs.sgml            structured, "SGML-like" syntax
       grs.regx.<filter>   structured, Regular expression filter
       grs.marc.<abs>      Reads *MARC records in the ISO2709 format. <abs>
                           is the name of an abstract syntax file.                           
Bug fix: Result sets weren't sorted in operations involving boolean
operations with "ranked" operands.

--- 1.0a8 1996/6/6

Added national character-handling subsystem.

Various fixes.

Small modifications to input filters and profiles.

Added support for SOIF syntax (with private OID).

--- 1.0a7 1996/5/16

Fixed buffer-size problem in indexing.

Added compression to temporary files for updating.

Added phrase registers.

Added dynamic mapping of search attribute to multiple termlists (ANY).

Scan support in multiple databases/registers.

Configuration settings are case-insensitive and single dash (-)
characters are ignored in comparisons.

The index processing ignores empty files - warning given.

New option to zebraidx (-V) displays version information.

--- 1.0a6 1996/2/24

Fixed problem in file-update system.

Fixed problem in shadow system; register was sometimes corrupted after
a commit operation.

--- 1.0a5 1996/2/10

Fixed problems in the ISAM subsystem. Caused difficulties when updating
existing registers.

Fixed small problem in SUTRS-filter. A newline was sometimes inserted before
the rank and record number.

Fixed bug in the isam subsystem - caused a malfunction when accessing
words which occurred more than 10000 times.

Distribution should now include YAZ (Z39.50 protocol stack) to simplify
installation.

Server can now run under inetd. Use option -i, and -w <directory> to
set working directory to desired location.

New zebraidx command: clean - removes temporary shadow files.

Fixed bug in ISAM system. Occurred rarely during register updates.

Logging during index merge phase is improved. The remaining running
time is estimated.

Temporary files generated by zebraidx are removed after each run.

Bug fix: Dictionary didn't handle 8-bit characters correctly; was obvious
when doing scan operations in dictionaries with European characters.

--- 1.0a4 1996/01/11

A whole slew of updates, to make the first publicized release. Get the doc
and check it out.

--- 1.0a3 1995/12/06

Memory-problems in ISAM fixed. More blocktypes added to the default setup
to increase performance on larger databases.

Various minor changes in data management system.

--- 1.0a2 1995/12/05

A couple of portability-problems resolved.

Changed some malloc() to xmalloc().

--- 1.0a1 1995/11/28

First release.