Sun Feb 28 23:56:41 PST 2016
Version 2.0.3
-- include a patch from Petr Velan to deal with include files related to
     util.h and pwd.h
-- make interpretation of internally generated column names of the form
   __xx in src/part.cpp
-- lock accesses to values in sharedInt32 and sharedInt64 when atomic
   operations are not available
-- rename select clause keyword distinct to countdistinct
-- address the problem with query evaluation cost reported by Petr Velan
   on 05/07/2015
-- add guards to called to strcmp to prevent nil pointer from being used
-- allocate sufficient space before calling ibis::fileManger::doRead to
   avoid recursive invocation of mutex locks (i.e., avoid a deadlock
   situation)
-- per suggestion from Jon Strabala, change all string read/write
   functions from unbuffered versions (read/write) to buffered versions
   (fread/fwrite)
-- replace std::binder1st and std::binder2nd with std::bind1st and
   std::bind2nd
-- rename ibis::util::cleanDatasets to be ibis::util::emptyCache


Mon Mar 23 23:14:31 PDT 2015
Version 2.0.2
-- address thread safty issue with bitvector::do_cnt
-- modify dictionary impelementation to allow users to specify arbitrary
   non-negative values (reserve -1 for NULL values)
-- install header files in subdirectory fastbit (requiring #include
   <fastbit/ibis.h> instead of #include <ibis.h>)
-- accept a patch from Chen HANG to improve the performance of reading a
   small number of string values in the query results
-- add support for ibis::BIT as a data type
-- accept a patch from Justin Swnhart to fix problem in tafel::writeData
-- fix problem with bitvecotr::do_cnt zeroing out the incorrect data
   elements
-- accept a patch from Divya Chandrasekar in sorting string values in
   colValues.cpp
-- fix the problem of part::countPages returning 0 when all data are
   located on the same page


Mon Dec 15 01:26:31 PST 2014
Version 2.0.1
-- fixing problem with html documentation
-- clear the output when ibis.cpp encounters an error (suggested by Jon Strabala)
-- add test cases to exercise the option of using external dictionary
-- fixing problem with indexing a single value in iapi.cpp
-- update the compilation instructions in INSTALL to reflect the requirement of C++11 features by FastBit


Fri Nov 21 21:38:00 PST 2014
Version 2.0.0
-- remove the prefix 'ibis' from the version number based on a
   recommendation from lscott@kiemaadvisors.com 
-- accepted a patch from Sean McNamara to fix a problem with testing C++ versions
-- add entry in runconfig to support edison.nersc.gov
-- accepted a patch from Zhongbo Tian on handling string-valued data
-- zero out the memory allocated through ibis::array_t<T>::reserve
-- change the default binned index option to be 2-digit precision bins
-- modify index generation code so that they will run without col object
   if possible
-- correct a set of problems identified by Zhongbo Tian in getColumnAsTypes
-- accepted a patch from Zhongbo Tian that fixes the order of bound
   check in ibis::mensa::cursor
-- add iapi for operations on in-memory data with the help in design and
   debugging from Junmin Gu
-- address thread-safety issue report by Dominique Prunier
-- address data type recognition problem reported by Li Jun
-- address index file locking problem reported by Petr Velan
-- address pointer access problem reported by Jerome Soumagne
-- add support for treating integers as time_t with extensive help from
   Jon Strabala, who suggested the idea, provided sample code for
   implementation, provided test cases, and helped with debugging
-- Received a new license on the software


Thu Mar 13 21:31:45 PDT 2014
Version ibis1.3.9
-- bug fix in ibis::bitvector
-- update ibis::tafel::write and friends
-- update quickstart.html to fix inconsistency reported by Neils Johnson
-- add function addConditions to allow query conditions to be modified
-- streamline the evaluation of discrete range conditions, adjust the
   scoring option to prefer sorted values
-- use linear interpolation to accelerate binary searches
-- replace auto_ptr with unique_ptr
-- change ibis::dictionary to use hash map, and write the dictionary with
   a new version number
-- merge the strings in a dictionary before writing the values to a file
   to accelerate the writing
-- correctly detect select clause without aggregation operatioins as not
   separable as the inline documentation promised (a problem reported by
   Sean McNamara)


Fri Oct 18 22:54:36 PDT 2013
Version ibis1.3.8
-- add macro FASTBIT_ESTIMATION_IGNORE_COST
-- allow relative path to be used in specifying data directory
-- rename function distinct to countd (meaning count distinct)
-- updated configure script for ORNL machines
-- add function EXISTS to test the presence of the named column
-- fix a bad test in array_t<T>::deepCopy based on a test case provided
   by Wang JiLong


Sat Jun 15 14:55:28 PDT 2013
Version ibis1.3.7
-- add function fastbit_get_result_row_ids to C API and JNI
-- add guard to deal with 0-length bit fills in bitwise logical
   operations per suggestion of Srikanth Sundarrajan
-- add functions ibis::query::getCandiateVector and getCandidateRows per
   request of Nan Zhou
-- address problem with logical OR operations in selection conditions (a
   problem reported by Steven Enns)
-- address problems in handling text columns reported by Bailu Ding and
   Mark Hansen
-- address problem with memory leaks based on the test case from Amihay
   Genon


Wed Mar 27 08:44:25 PDT 2013
Version ibis1.3.6
-- update index building functions to work with in-memory data
-- convert internal integer representation back to string in output
-- add ibis::table::dropPartition as the opposite of addPartition


Thu Jan  3 21:05:47 PST 2013
Version ibis1.3.5
-- in ardea.cpp, do not write the data if a data directory is not
   specified by the caller - this allows the all queries to be performed
   on in-memory data
-- replace the use of /dev/random to /dev/urandom to avoid blocking,
   change the default option to initialize the random number generator
   to use the current time (to address a unexpected delay in starting up
   of FastBit code raised by Preeti Gupta)
-- change the protected construction function of index classes ibis::bin
   and ibis::relic to be public functions to avoid the need of
   inheriting from them in client code
-- Patrik Nisen fixed incorrect names in ibis::bord::cursor::fillRow
-- change filter class to take previously saved query results at the
   request of Patrik Nisen


Thu Nov 15 17:01:37 PST 2012
Version ibis1.3.4
-- change the templated constructor of qIntHod and qUIntHod to use fixed
   data types (to address a problem noticed by Andrew Olson)
-- checking the member variable actual to avoid accessing a nil pointer
   in case of unmanaged array
-- restructure index creation process to address a problem with indexing
   categorical values (reported by Gaurav Agrawal)
-- streamline the use of sorted values (following a suggestion from
   Andrew Olson)
-- update in-memory data partitions to better handle strings
-- update macros and in-line comments to conform to more strict ANSI
   standard (a problem reported by Hari Krishnan)
-- add ibis::bitvector::operator< to make it usable as keys in std::map
-- fix problem with incorrect coordiantes in the output lines from
   ibis::meshQuery::getHitsAsLines


Tue Oct 16 08:43:28 PDT 2012
Version ibis1.3.3
-- index direkts neglected to take into account of null mask (a problem
   reported by Michael Beauregard)
-- consolidate the handling of deletion in array_t to function
   freeMemory, streamline the test of current size to address a problem
   with push_back reported by Petr Velan and Michael Beauregard
-- use the null masks when constructing bundles (was assuming all values
   to be valid)
-- allow dictionaries of meta tags to be merged like other type of
   categorical values, allow meta tags in tables to be reordered (other
   types of string-valued columns are still not allowed)
-- allow column::getDoubles to convert all other numerical types to
   double.  This was promised in the documentation, but actually missed
   the case for 8-byte integers (a problem reported by Gaurav Agarwal).
-- update the testing on array bounds in ibis::bylt to address the new
   problems revealed after the modification to random number generators
   used in src/part.cpp


Tue Sep  4 16:36:53 PDT 2012
Version ibis1.3.2
-- add group_concat as a new group-by opertors in the select clauses
   (requested by Michael Beauregard)
-- modify handling of array_t in functions reserve, resize and push_back
   to allow a shallow copy to be used when the modification does not
   expand beyond the existing storage capacity (a feature necessary to
   support a more efficient way to combine serialized bitvectors in
   FastQuery code)
-- more consistent handling of null masks in query processing and
   reordering of data partitions, modify the built-in test suite to
   exercise the null handling
-- finalize the class ibis::slice after renaming the previous version to
   be ibis::skive, ibis::slice is now closer to the original definition
   of bit-sliced index


Wed Aug 15 21:58:00 PDT 2012
Version ibis1.3.1
-- impelement ibis::util::read and ibis::util::write to get around the
   limitations with POSIX read and write functions in handling large
   buffers
-- add UDT as a new data type for future expansion
-- support for BLOB with ibis::opaque class
-- process categorical values as integers in memory and on output,
   convert query conditions on categorical values to be integer
   comparisons
-- add query expression with the operator "CONTAINS" to allow for a more
   consistent handling of keyword searches (also known as full-text
   search)
-- reorganize -output option of ibis.cpp to add new flavors for
   -output-as-binary so that the resulting output can be used direct by
   FastBit (total three flavors: -output, -output-with-header,
   -output-as-binary)


Mon Apr  2 20:07:36 PDT 2012
Version ibis1.3.0
-- adopt case-sensitive comparisons for strings in a dictionary at the
   suggestion of Dominique Prunier
-- clean up left-over problems with using ibis::direkte in
   ibis::category
-- Andrew Olson adds dumpJSON
-- add function sort dictionaries and merge dictioaries
-- enable array_t::deepCopy to copy an empty array (problem reported by
   Teryl Taylor)
-- Complete the functions to allow string values to be used as sorting
   keys in group by operations
-- replace most of the call to ibis::bitvector::cnt with
   ibis::bitvector::sloppyCount
-- inprove the performance of ibis::category::patternSearch by using
   ibis::index::sumBins
-- add more test case for exercise special characters in dictionary
-- add support for "NOT NULL" in where clause
-- distinguish explicitly quoted empty string and unquoted empty string
   in the CSV input files


Sun Mar 11 22:58:38 PDT 2012
Version ibis1.2.9
-- faster string handling through internal use of integer representation
   (with contribution from Dominique Prunier and Alexandre Maurel)
-- change the log message to use stderr/std::clog
-- make sure the result from a join can be passed to another query
   (added test cases in ibis.cpp)
-- add win/xMinGW.mak for cross compiling Windows binary on ubuntu
   (mostly from Dominique Prunier)
-- change ibis::category to use a slender index class ibis::direkte
   instead of ibis::relic
-- add option to perform pattern match on words in a dictionary
   (Dominique Prunier) and allow for case-sensitive operations to make
   use of the binary search algorithm


Fri Jan 20 16:43:32 PST 2012
Version ibis1.2.8
-- change ibis::part::buildIndexes to rebuild indexes if the existing
   ones are out of date
-- Dominique Prunier added function to interate over bundles, fixed
   initailization issues
-- add option to force the index building procedure to invoke an extra
   round of compression calls
-- complete the code that makes uses of .int files for categorical
   values
-- change ibis::part::buildIndexes to build all auxiliary data files to
   avoid the need of write access to data files during query processing
-- enable control of whether to sort RIDs when creating bundles, sorting
   RIDS could be as expensive sorting to create the bundles
-- implement aggregation (group by) operations COUNT, COUNT-DISTINCT,
   MIN, and MAX for string values


Mon Jan  9 18:04:47 PST 2012
Version ibis1.2.7
-- Tomas Rybka implemented algorithm to perform binary merge to reduce
   the number of copies made during the merging of partial results
-- update configure script to better interact with older versions of
   make
-- update lock handling in array_t::nosharing (suggested by Tomas Rybka)
   and in part::readRIDs (suggested by Dominique Prunier)


Mon Dec 12 21:43:53 PST 2011
Version ibis1.2.6
-- clean up the functions to merge partial aggregation results with the
   help of Petr Velan <thorgrin @ gmail.com>
-- clean up the code for compilation on MS Windows with Visual Studio
   and MinGW compiler


Wed Dec  7 14:41:05 PST 2011
Version ibis1.2.5
-- accepted a patch from Jon Strabala (Quantum Systems Integrators) to
   support is_zero, is_nonzero, is_eql, is_gte, is_lte in the where
   clauses
-- add function to support aggregations by parts
-- rewrite aggregation expressions of AVG, VAR and STDDEV to use SUM and
   COUNT to allow them to be computed by parts
-- support DESC and ASC for each column mentined in the order-by clauses
   as suggested by Jon Strabala
-- accepted a patch from Robert Wong to allow query processing to
   continue in case of errors
-- clean up the code to remove most warning messages from GCC and pgCC
-- fix bugs in jNatural and jRange


Fri May 13 16:47:06 PDT 2011
Version ibis1.2.4
-- overwrite output file with an empty file when there is no output at
     the request of Jon Strabla
-- refactor ibis::index::create to allow for better control of I/O
     options
-- refactor the constructor of ibis::keywords to make more consistent
     use of the internal parser for text
-- enable the user to choose whether to use quiet NaN or maximum values
     as the default NULL values
-- fix a problem with shallow copy constructors of ibis::array_t to
     address the problem with sorting functions (a problem reported by
     Alexandre Maurel)
-- fix a problem with ordering of select clause reported by Jon Strabala
-- add runconfig and bin/fastbit-config


Wed Mar 30 13:14:35 PDT 2011
Version ibis1.2.3
-- clean up the implementation of low-precision binning option by
   refactoring ibis::bin::granule (in response to questioins raised by
   Jinoh Kim)
-- reducing memory usage in handling selected values by allowing more
   shallow copying with ibis::array_t
-- adjust the cost estimation of ibis::bin::estimateCost to take into
   account of anticipated candidate checking
-- change ibis::bitvector::setBit to decompress the bitvector when
   attempting to change a bit in the middle of a compressed one (faster
   operation, but takes more memory)
-- add util::refHolder to avoid copying objects in scope guards
-- add more error handling statements to bundle.cpp
-- add configure option --without-getpwuid to avoid invoking getpwuid on
   compute node linux machines
-- update operations with meta tags of data partitions by improving the
   operations of category object with a single string value (to address
   problem reported by Andrew Olson)
-- fix problem with ibis::fuzz::readCoarse that computes the incorrect
   number of elements for offset32/offset64 (to address a problem
   reported by Jon Strabala)
-- expand java/milky.java to traverse subdirectories and report
   processing time in each subdirectory


Wed Dec 22 14:31:36 PST 2010
Version ibis1.2.2
-- add functions to output query lines from ibis::meshQuery and label
   the query lines, add test functions for crosschecking the two
   different sets of connected component labeling functions
-- add functions to perform connected component labeling on the blocks
   produced by ibis::meshQuery::getHitsAsBlocks
-- update the visual studio project files to remove unnecessary
   definitions to make them work better with newer visual studio
   versions
-- add a rudimentary parser to extract keywords for keyword indexes
   as ibis::text::tokenizer class hierarchy
-- allow embedded spaces in keywords of a keyword index to address a use
   case from Justo Ruiz Ferrer, add jrf.cpp in tests directory to
   replicate the test
-- add function to update metadata information after modification of
   datasets by other processes (at request of Chavdar Kopoev)
-- address file map issues on windows machines raised by Chavdar Kopoev
-- clean up print issues raised by Anderson Chaves

Wed Nov  3 18:21:35 PDT 2010
Version ibis1.2.1
-- address the locking issues reported by Andreas Streichardt
-- update the code for Compute Node Linux (mostly related to the usage
   of getpwuid)
-- add using clauses to address the complaints produced by PGI compilers
-- remove references to commons logging objects to reduce the
   dependencies of Java interface (prompted by a test case from Anderson
   Chaves)
-- add members to the colValues class hierarchy to handle shorter
   integers (to reduce the memory usage, prompted by a use case from Ben
   Williams)
-- clean up the regression tests


Fri Sep 17 13:35:06 PDT 2010
Version ibis1.2.0
-- reserve more space in the dictionary to avoid incrementing the
   dictionary sizes in small chunks
-- fixing bitvector::operator- problem (by Peter Tardif)
-- update the memory management in ibis::bundle to avoid clobbering
   other data structures (to address a problem reported by Andreas
   Streichardt)
-- update the error handling in the select operation following the
   suggestion from Jan Steemann
-- update makefiles for windows environment
-- fix problems with JNI code (reported by otrack50@hotmail.com)
-- fix problems with user specified log files (reported by Karen
   Schomburg)


Thu Jul 15 20:20:18 PDT 2010
Version ibis1.1.9
-- fix JNI memory leak (Zhengbing Li and Xun Hu)
-- add test cases for very small datasets
-- implement a new dictionary format for categorical values
-- add reordering functions to C API by Georgiy Zhytar
-- add integer version number, make the constructors of storage use its
   arguments more consistently
-- add a two-argument version of scope guard
-- implement the range join class, add testing in tests/Makefile.am
-- wrap most of the raw pointers with std::auto_ptr
-- add code to accept 64-bit integers with suffices LL and ULL (only on
   the right-hand side of a equality or a list of IN operator)
-- add testing code for query the retrieves a larger number of rows or
   require a lot of intermediate memory (to address the memory leak
   issues required by Jan Krivanek)
-- fix the ordering of columns returned by columnNames and columnTypes
   (a problem reported by Jan Steemann)


Fri May  7 14:02:15 PDT 2010
Version ibis1.1.8
-- modify ibis::part::barrel to work with in-memory data partitions (to
   address a problem reported by Andreas Streichardt)
-- update ibis::bord::column to work with string values
-- refactor string search functions to accomodate queries on in-memory
   data
-- add back ibis::part::mutexLock to provide support for HDF5_FastQuery
   (an issue raised by Prabhat)
-- implement support for the limit clause with two arguments (suggested
   by Valeria Lorenzetti)
-- fix problem with unrolled loops in string comparison functions
   spotted by Bernd Jaenichen
-- added code from Bernd Jaenichen to test discrete range queries on
   in-memory data partitions
-- consolidate the functions for evaluating discrete range queries


Wed Apr 14 13:14:28 PDT 2010
Version ibis1.1.7
-- address the problem with trailing directory separator reported by
   Valeria Lorenzetti
-- rename old join expression to deprecatedJoin
-- add special case to handle constant expressions in the where clauses
-- add parser for the from clauses
-- add a skeleton of query class named ibis::quaere
-- add checks to address reference counting problems reported by Zeid
   Derhally
-- remove the verboseness level argument from ibis::init, add a new
   function named ibis::util::setVerboseLevel to address the problem
   with 'make check'
-- correct the values returned by ibis::tabele::cursor (a patch by
   Andreas Streichardt)
-- fix the bug related to 'count(*)' reported by Andreas Streichardt
-- fix incorrect variable names in fileManager::getFile reported by
   Teryl Taylor
-- add fastbit_get_qualified_strings in the C API at the request of Joze
   Nazario
-- add util::readUInt to handle the test case suggested by Arian Baer
-- expand the implementation of ibis::bord::column to make it fully
   compatible with ibis::column (to support queries on in-memory data
   partitions produced from another query, a feature suggested by both
   Arian Baer and Teryl Taylor)
-- accept a patch from Shiran Pasternak to allow ant jobs to access the
   dynamic library


Fri Jan 29 12:10:21 PST 2010
Version ibis1.1.6
-- minor changes to avoid some problems with PGI compilers
-- allow the input of null values from csv files at the request of
   Andreas Streichardt
-- remove the default space delimiters when reading CSV inputs to
   address an issue raised by Elaheh Pourabbas
-- address an indexing building problem involving categorical values
   reported by Elaheh Pourabbas
-- add tests/scripts/hcompare.pl to allow small variations in the
   histograms produced in the tests
-- exclude the null values in the select clause in ibis::countQuery,
   ibis::query already does this, to address an issue raised by Andreas
   Streichardt
-- add aggregation function median, update the implementation of
   counting distinct values to avoid copying data
-- correct the handling of signs in front of numbers to correct an issue
   reported by Andreas Strechardt
-- streamline the handling of constant expressions
-- fix bugs in group-by operations reported by Valeria Lorenzetti
-- fix a bug in the constructor of ibis::liga (reported by Joel Dolisy)
-- clear the cache before performing the reordering function as the
   suggestion of Joel Dolisy

Thu Dec 17 14:14:28 PST 2009
Version ibis1.1.5
-- fix compilation problem with gcc 4.4 (idirekte.cpp neglected to
   include typeinfo (reported by Luca Deri)
-- change the test in array_t<>::nosharing to leave the content managed
   by the fileManager as read-only (to address an issue raised by
   Alexandre Maurel)
-- update query parsing function in examples/ibis.cpp to correctly
   handle order by clauses (to address problem reported by Jan Hapke)
-- change the associativities of unary operators from left to right to
   address the parser problem reported by Andrew Olson
-- treat the quoted strings as string literal (and therefore remove
   the possibility of treating it as a column name) to fix a problem
   reported by Alexandre Maurel
-- all string values to be used for group by operations
-- add contrib directory to include the file fbmerge.cpp contributed by
   Luca Deri and Valeria Lorenzetti
-- add tests and data contributed by Alexandre Maurel to check for the
   problem with parsing string valued query conditions
-- add support for operator LIKE for comparing strings
-- allow aggregation functions (in the select clauses) to appear before
   the group-by columns
-- allow count(*) to be explicitly used in a select clause

Tue Nov 10 14:17:41 PST 2009
Version ibis1.1.4
-- add code to switch bitmap offset between 4-byte and 8-byte integers.
   This allows the index files to grow beyond 2GB, but the number of
   rows in a data partition is still limited to 2 billion.
-- apply patch from Jan Steemann to add new aggregation functions
-- apply patch from Luca Deri to correct bundle computation
-- fix functions segement in colValues.cpp after messing up the patch
   from Alexandre Maurel
-- fixes bugs discovered during regression testing

Thu Sep 10 15:56:08 PDT 2009
Version ibis1.1.3
-- add function ibis::tablex::readSQLDump to read sql dump files, update
   ibis::tablex::readCSV to output the extract data as it reads
-- add column type ibis::BLOB (implementation as ibis::blob in
   category.h and category.cpp, a feature request by Matthias Vallentin)
-- add ibis::util::guard class as an implementation of ScopeGuard
   (Andrei Alexandrescu and Petru Marginean, Generic: Change the Way You
   Write Exception-Safe Code -- Forever, <http://www.ddj.com/cpp/184403758>)
-- incorporate a bug fix from Alexandre Maurel to address problems in
   function segment in ibis::colValues hierarchy (Alexandre's original
   fix was for ibis::colFloats only)
-- add function ibis::table::select that works with a list of data
   partitions (to address a need raised by Andrew Olson)
-- modify bitwise logical operations in bitvector.cpp to ensure the
   lower level functions only operation on bit vectors representing the
   same number of bits (to address a set of issues reported by David
   Reynolds)
-- update ibis::util::coarsen to make output more robust (make 1e-14 and
   10e-15 be represented by exactly the same binary representation) 
-- fix bug in idbak2.cpp line 998 that causes crashes

Wed Aug  5 09:32:18 PDT 2009
Version ibis1.1.2
-- delay the reading of dictionary and index for categorical values to
   speed up the start up
-- add ibis::colStrings to enable group by operations involve string
   valued sorting keys (a feature requested by Andrew Olson)
-- query::doEvaluate, query::doScan, countQuery::doEvaluate and
   countQuery::doScan handle empty query conditions incorrectly (a bug
   reported by Jochen Schlosser)
-- allow categorical values to be used as unsigned integers for
   histogram functions
-- adaptive binning functions computed an internal scaling factor
   incorrectly and causes the bin number to be out of bounds (the cause
   of error reported by Andrew Olson)
-- change ibis::bak2::grain to use three bitvectors to store whether the
   incoming value >, =, < than the target low-precision value.  This
   allows for the binned index to resolve all possible range conditions
   involving low-precision query boundaries
-- add support for "count(*)" in select clause, allow aliases to be used
   in the select clauses and in aggregation functions
-- bug fixes: check for nil point in bundle.cpp, cast to char and
   unsigned char (which are interpreted as 8-bit integers) to int before
   printing, make sure the value of k in top-k and bottom-k are not 0

Mon Jul  6 15:32:26 PDT 2009
Version ibis1.1.1
-- put array_t in namespace ibis (requested by Matthias Vallentin)
-- check the state of the partition before truncating the data files (to
   address a problem reported by Jochen Schlosser)
-- in text::startPositioins test for the value fo currentDataDir()
   before using it in strcmp (to avoid a crash reported by Jochen
   Schlosser)
-- in bitvector::subset, avoid accessing a void pointer (to address a
   crash reported by Jochen Schlosser)
-- in constructor of ibis::bord::part (and ibis::mensa::doSelect) avoid
   allocating space to the same column name multiple times (to address a
   problem reported by Shane Adams)
-- allow purgeInactive to work with string valued columns
-- add ibis::liga to provide ibis::table API for arbitrary collection of
   ibis::part
-- add code to address query boundaries out side of valid ranges of
   column values (a problem noted by Jochen Schlosser when using
   negative values with unsigned integer columns)

Fri Jun 12 17:10:16 PDT 2009
Version ibis1.1.0
-- automate the line buffer management in function tafel::readCSV to
   address a problem reported by Nicolae Mihalache <mache at abcpages
   dot com>
-- update ibis::tablex and ibis::tafel to work with a genomic
   application, make it easier to derive from and extend
-- add column::appendValues and appendStrings to directly work with
   in-memory data (previous one has to prepare a data partition and then
   append it to another one)
-- consistently assign the state of data partition in all constructors
   of ibis::part to address an issue raised by Jochen Schlosser
-- fix issues in update column masks when repeated adding new data
   (reported by Jochen Schlosser)
-- fix a memory leak (losing the value of activeDir) in the constructor
   ibis::part (reported by Jochen Schlosser)
-- change tests to use sampled histograms in more cases in function
   ibis::index::mapValues (a temporary work-around for issues with very
   large std::map)
-- use a mutex lock to protect the statements that access the value of
   ibis::column::idx in constructor of ibis::column::indexLock and in
   function ibis::loadIndex (to prevent use of ibis::column::idx while
   the index is being computed.  In the previous implementation, there
   could be a long delay between when idx is given a valid memory
   address and the completion of the index construction, which can cause
   a partially constructed index to be used by another thread)
-- update some log messages

Tue May 12 10:07:42 PDT 2009
Version ibis1.0.9
-- refine all estimateCost functions (wrong cost estimates have led to a
   wrong choice for evaluation strategy, 1st reported by Prabhat and
   Allen Sanderson)
-- Change template ibis::column::getRawData to a virtual function named
   ibis::column::getValuesArray to make it easier to override it by derived
   classes (to addresss a problem first reported by Allen Sanderson)
-- use the library function nextafter inside incrDouble and decrDouble
   when available
-- refine the code in templates ibis::part::doScan and
   ibis::part::doCount to reduce the amount of computations
-- add macro WITHOUT_FASTBIT_CONFIG_H as the opposite of HAVE_CONFIG_H
   so that the user code does not need to define HAVE_CONFIG_H when
   configure script can be run (this change assumes the FastBit
   config.h file is present and removes the need for client code to 
   define HAVE_CONFIG_H when including FastBit headers, a problem first
   encountered by Hank Childs)
-- integrate part::testRangeOperators as part of selfTest to exercise
   all combinations of logical operators in a range expression
-- clean up ibin.cpp, irange.cpp, imesa.cpp, icentre.cpp, ixambit.cpp,
   ixpack.cpp, ixpale.cpp, and ixzone.cpp to address errors identified
   through 'make full-check'

Fri Apr 24 14:45:27 PDT 2009
Version ibis1.0.8
-- clean up implementation of ibis::tablex in ibis::tafel
-- add support for all elementary data types to functions
   ibis::query::getQualifiedTTTs and ibis::part::selectTTTs
-- use references (instead of pointer) when retrieving a single values
   through ibis::table::cursor
-- clean up documentation about a number of functions
-- update the detection rule on where to locate FlexLexer.h in configure
   script
-- use explicit type casting in the print statements of examples/tcapi.c
   to avoid problem with print the wrong values (first saw on a Mac with
   gcc 4.0.1)

Wed Apr 15 16:42:04 PDT 2009
Version ibis1.0.7
-- fix a bug in ibis::qExpr::simplfy that returns a deleted query
   expression (a bug reported by Hank Childs and Prabhat)
-- rename the file for external Row IDentifiers from "rids" to "-rids"
   to prevent name collision
-- change getRIDs to return implicit/internal RIDs if external RIDs are
   not provided (a feature requested by Jochen Schlosser)
-- add functions to ibis::tablex interface: nRows, nColumns, capacity,
   and reserveSpace
-- add macro FASTBIT_SYNC_WRITE to control whether or not to wait for
   write operations

Tue Apr  7 21:44:08 PDT 2009
Version ibis1.0.6
-- implement functions to work with sorted data, add flags to indicate a
   data column is sorted (a feature requested by Prabhat and Oliver
   Ruebel)
-- add functions to computed weighted histograms (a feature requested by
   Oliver Ruebel)
-- add ibis::countQuery as a simplified version of ibis::query
-- extend ibis.cpp to exercise the new functions
-- clean up some of the printing functions
-- update error handling functions
-- correct a bug in ibis::fileManager::buffer that records the wrong
   size.  This causes FastBit to mismanage memory.
-- change the definition of ibis::util::counter to use macro
   HAVE_GCC_ATOMIC32 (to address a problem reported by Andrey Kolchanov)
-- correct usage problems with olddesc in ibis::tafel::write

Mon Mar  9 17:03:43 PDT 2009
Version ibis1.0.5
-- add FastBitStringReader, FastBitStringWriter and their assocaited
   unite test function contributed by Andrey Kolchanov
-- add new test cases 'check-marksdb' to 'make check'.  These tests make
   use of data and queries contributed by David B. Marks
-- fix a bug in the computation of the starting positions in function
   ibis::text::startPositions (a problem reported by Andrey Kolchanov)
-- sort incoming values using an array instead of std::set at the
   sugestion of Alesandro Capucci
-- update configuration scripts
-- add an option to specify the maximum number of rows to expect to
   ibis::tafel::readCSV to reduce the number of dynamic memory
   allocations (a suggestion from Norbert Heusser)
-- add code to produce sorted list of values for any column with
   fixed-sized values
-- make it possible to use aliases defined in select clause inside where
   clauses (to address a case raised by David Marks)

Fri Feb  6 12:00:26 PST 2009
Version ibis1.0.4
-- add example program rara.cpp as a minimalistic client code
-- refine the macros for controlling the use of 64-bit atomic operations
   on windows platforms
-- add arguement to ibis::part::loadIndex to allow all bitmaps to be
   read into memory in one-shot
-- change the semantics of fastbit_get_qualified_longs to return the
   starting positions of strings when it is called with the name of a
   string-valued column (to address a request from Andrey Kolchanov)
-- split src/parth3d.cpp into three files (one for each function that
   requires massive number of instances of template function) and limit
   the number of template instances.  The is to reduce the compilation
   time under gcc.  The full slew of instances can still be generated if
   the macro FASTBIT_EXPAND_ALL_TYPES is defined.
-- update the parser to accept functions with a single number as the
   argument (a bug reported by David Marks)
-- add calls to free up memory after encountering failure to retrieve
   file content (to address a problem encountered by Alessandro Capucci)

Tue Jan 20 21:14:57 PST 2009
Version ibis1.0.3
-- add ibis::text::selectLongs to return the starting positions of the
   string valued columns (to enable reading of strings for java
   based applications, a hack in response to Andrey Kolchanov's
   question)
-- implement ibis::util::sharedInt32 and ibis::util::sharedInt64 using
   GCC's atomic operations and Visual Studio's atomic operations.  This
   should improve the speed of operations and make reference  counting
   and tracking of memory usage more reliable on multi-threaded
   applications
-- add more test cases to ibis::part::selfTests to exercise query
   evluation function in multi-thread mode
-- add another set of binning functions, get1DBins, get2DBins, and
   get3DBins that return std::vector<ibis::bitvector*>.  These functions
   will return null pointers for empty bins, which can reduce memory
   usage and execution time in cases there are many empty bins

Mon Jan  5 12:32:07 PST 2009
Version ibis1.0.2
-- add another set of binning functions, get1DBins, get2DBins, and
   get3DBins, that take (begin, end, stride) to specify the bin
   boundaries instead choosing bin boundaries by the functions
   themselves.  A feature requested by Oliver and Probhat.
-- use the new version of binning functions to debug the existing
   versions
-- reduce the memory usage of the two versions of ibis::util::intersect
   by explicitly compressing the results and copying the compressed ones
-- make sure a bitvector has its own storage when it is being modified

Thu Dec 18 10:47:05 PST 2008
Version ibis1.0.1
-- correct errors in two-level indexes that prevented them from actually
   generating the coarse level
-- update README file in various directories
-- update Makefiles for MinGW and Visual Studio
-- add option -output-with-header as a variant of -output as requested
   by David Marks

Tue Nov 25 18:27:56 PST 2008
Version ibis1.0.0
-- modify ibis.cpp to accommodate new output produced by arithmetic
   expressions in select clauses
-- add warning messages as suggested by Niklas Bulitta
-- address errors involving short integers (reported by Niklas) and
   other types of integer variables when creating unbinned (precise)
   bitmap indexes
-- update test scripts to catch the errors discovered by Niklas

Mon Nov 10 21:26:14 PST 2008
Version b0.9.9
-- add adaptive binning for 3D histograms
-- add histogram functions that output bitvectors instead of counts
-- add print3DDistribution to examples/ibis.cpp and a test case in
   check-ibis
-- add function tableSelect to ibis.cpp to exercise querying through
   ibis::table select function (which accepts arithmetic expressions in
   the select clauses)

Thu Oct 23 16:42:31 PDT 2008
Version b0.9.8
-- add arithmetic expression to the select clauses accepted by function
   ibis::table::select
-- update java example (correct issues raised by Niklas Bulitta)
-- rewrite the parser for query expressions to use the C++ options of
   bison and flex (predicate* changed to where*)
-- add parsing code to handle arithmetic expression in select clauses
-- reorganizes ibis::qExpr class hierarchy and move arithmetic
   expressions under namespace ibis::math
-- address an error report by David Kantowitz about
   ibis::column::getString
-- minor changes to print statement and code organizations

Fri Sep 19 22:38:43 PDT 2008
Version b0.9.7
-- fix an error in mensa.cpp (involving the implementation of
   ibis::mensa::cursor) spotted by David Kantowitz.  Fix two problems 
   involving the same member variable elsewhere in the same file.
-- change _USRDLL to CXX_USE_DLL in the project files to be consistent
   with conditional macros in source code.  The DLL libraries can now be
   compiled and run correctly.
-- update visual studio project files for visual studio 2008
-- increase the cap on the number of bins in adaptive2DBins to make sure
   Prabhat's tests run without surprises

Thu Aug 21 09:58:44 PDT 2008
Version b0.9.6
-- implement new adaptive binning functions for computing histograms of
   1D and 2D that avoids binary searches.  The new functions are not
   only faster, but because they use more fine bins internally the
   output bins are closer to be equal-weight than before
-- change the mensa::cursor class to read data one block at a time so
   that ardea.cpp can print a few row of a large table without running
   out of memory
-- clean up the constructor of qDiscreteRange to eliminate duplicates
   correctly (an error reported by Prabhat)

Thu Jul 31 10:04:21 PDT 2008
Version b0.9.5
-- adjust definition of exact-width integers in const.h to avoid
   potential conflict with VisIt (suggested by Gunther Weber)
-- adjust scope of histogram in some constructors of two level indexes
-- add option -x to thula.cpp to export content of tables
-- add heap sort to sorting functions of array_t and ibis::util

Fri Jun  6 15:24:18 PDT 2008
Version b0.9.4
-- add ibis::colValues::nosharing to address a seg fault when selecting
   all values in a partition (reported by Ross Bates)
-- add utilidor.h, utilidor.cpp, join.h, joinin.h, and joinin.cpp to
   provide radix sorting and in-memory natural join (requested by
   Matthias Vallentin)
-- fix bug in ibis::part::writeTDC related to column shape statement
   (reported by Xufei Qian)
-- update testing code to match the change of option -j to -y for
   ibis.cpp (-j is now for join)

Wed May 28 07:38:11 PDT 2008
Version b0.9.3
-- update the range-equality encoding (ibis::bylt and ibis::pack) and
   interval-equality encoding (ibis::fuge and ibis::fuzz) to use the new
   default values for the numbers of coarse bins based on the neutral
   space-time trade-off heuristics
-- add option -f to examples/thula.cpp (requested by Jon Strabala)
-- change ibis::util::userName to primarily use getpwuid (instead of
   getlogin) following the suggestion of Ross Bates

Wed Apr 16 08:28:46 PDT 2008
Version b0.9.2
-- remove the statements that deletes index::str in the function clear
   of various derived classes, logical error, the file manager should call
   the actual delete
-- remove the use of variable ibis::bitvector::run::nWords when it is
   not a fill
-- correct typos in functions that perform computation on
   qDiscreteRanges, an error first reported by Ross Bates

Wed Apr  9 07:28:24 PDT 2008
Version b0.9.1
-- fix bugs found under PathScale and PGI compilers (mostly from working
   with Prabhat)
-- change starting positions file (.sp) to use int64_t explicitly
   instead of long, this eliminates the annoying problem of duplicate
   template instantiation of array_t
-- add ibis::tablex::clearData to remove the current data of a tablex
   object (in response to a request by Matthias)

Tue Apr  1 08:11:09 PDT 2008
Version b0.9.0
-- fix bugs involving start and end values of getxDDistribution, first
   observed by Prabhat
-- change decisions regarding the option of using priority queue in
   ibis::index::sumBits to reduces it use, to fix a performance
   irregularity observed by Prabhat
-- fix bug identified while computing histogram with very small number
   of distinct values
-- clean up the printing statements in fileManager.cpp
-- allow multiple print statements to examples/ibis.cpp

Mon Mar 24 11:15:41 PDT 2008
Version b0.8.9
-- update column.cpp to include <typeinfo> (from Matthias of ICSI)
-- attempt to improve the speed of computing histograms by sorting input
   data
-- add assignment of nrows in ibis::bin::construct that takes an array
   as input, a problem found by Prabhat
-- correct html errors in indexSpec.html
-- update decisions in column::selectValues to use the function read
   when it anticipate that a small number of seeks will be needed.  This
   should improve the speed of reading a small number of records.
-- add an optional argument to unconditional version of
   get2DDistribution to allow the caller to choose whether to use
   indexes or data (requested by Prabhat)
-- update test cases 7 and 8 of check-ibis to make them more robust
-- add ibis::util::timer class

Sun Mar 16 23:21:42 PDT 2008
Version b0.8.8
-- fix bugs in ibis::bitvector::flip and ibis::bitvector64::flip, a
   problem encountered by Prabhat
-- correct tests 7 and 8 of check-ibis in tests/Makefile.am 
-- add test cases involving arithmetic expressions in
   tests/scripts/query-count.list and src/part.cpp (ibis::part::quickTest)
-- fix bugs reported by Mark Sordo (about JNI) and Matthias VAllentin
   (about stdlib.h and PTHREAD_RWLOCK_INITIALIZER macro)

Wed Mar 12 10:20:16 PDT 2008
Version b0.8.7
-- add try-catch blocks to various interval encoded indexes to free
   memory in case of exception
-- add a new variant of get2DDistribution requested by Oliver and Prabhat
-- add a new count function to remove the need to invoke the bitwise
   logical AND operations while computing histograms using indexes

Sun Mar  9 07:52:50 PDT 2008
Version b0.8.6
-- add ibis::part::get1DDistribution and ibis::part::get2DDistribution
   that accept user specified number of bins (Requested by Oliver Ruebel
   and Prabhat)
-- fix bugs in two-level indexes that cause dead-locks and wrong answers

Tue Mar  4 08:43:42 PST 2008
Version b0.8.5
-- update doxygen documents to work around __declspec directives
-- fix bug reported by Ross Bates <rbates@gmail.com> regarding printing
   categorical values (key), add test cases to check for this error in
   'make check' and 'make quick-check'
-- fix bugs in histograming function ibis::part::count2DBins and
   count3DBins, update the related documentation (bugs reported by Prabhat)
-- in reponse to a question from Devin McNamara, modified readCSV to
   allow arbitrary user-specified delimiters
-- add a set of tests cases that works with very limited memory allowed
   for file manager (invoked through 'make full-check'; inspired by
   Jochen Schlosser).  This has revealed a number of serious problems as
   follows:
-- add ibis::column::softWriteLock to prevent deadlock during tight
   memory situations,
-- add checks for nil pointer of idx before using it,
-- add function in array_t to tell users that it is attached to a file
   or in-memory, this is needed so that ibis::bitvector::compress can
   avoid modifying the ones in files (modifying such bitvector causes
   segmentation faults)
-- add destructors to ibis::bak::grain and ibis::bak2::grain.  If the
   constructor of ibis::bak or ibis::bak2 fails, previous
   implementation of grain do not automatically release the bitvectors
   allocated
-- replace function ibis::util::getBuffer with class ibis::util::buffer
   to ensure the memory allocated is always freed

Wed Feb 20 11:26:11 PST 2008
Version b0.8.4
-- replace ibis::part::doCompare and friends with templated functions
-- add checks for the values returned by lseek and other I/O functions
   to fix a segmentation fault observed by Jochen Schlosser of
   University of Hamburg
-- fix bugs in histogramming functions found by Prabhat: unchecked
   pointers returned by selectInts and friends, uncheck empty arrays,
   swapped string arguments
-- add test cases for checking the histogramming functions in directory
   tests
-- add target check-all to the top-level makefile
-- separate the index generation from their uses in tests (as part of
   check-all) to exercise the index reconstruction functions, fix bug
   revealed by this separation (ibis::fade and ibis::bin)

Wed Feb  6 09:47:21 PST 2008
Version b0.8.3
-- fix a number of bugs reported by Jochen Schlosser of University of
   Hamburg: (1) seg fault due to minimum actually being negative even
   though the metadata indicate it is to be zero or positive, (2) when
   index ibis::direkte is used to answer a query, it erroneously casts
   negative integers numbers to unsigned values and produces incorrect
   answers
-- fix performance a problem reported by Prabhat: a long list in the
   form of "A in (...)" is translated into a massive set of equality
   conditions joined with OR.  Evaluate such a query condition as
   qDiscreteRange.
-- fix bugs in assigning bin boundaries for log-scaled bin on integer
   values and low cardinality columns

Wed Jan 16 12:37:24 PST 2008
Version b0.8.2
-- fix a number of bugs reported by Jochen Schlosser of University of
   Hamburg: ibis::column::evaluateRange calls ibis::part::doScan
   unnecessarily, a number of functions in ibis::fileManager::storage
   class does not check for nil pointers before using them, ibis::part
   has a number of funtions that neglected to free memory returned by
   ibis::column::dataFileName
-- change exception handler to not remove index files unless the index
   files are really at fault
-- add return values to query evaluation function so that a negative
   number will be returned if it fails for any reason, previous version
   neglects to return the error condition to the user

Sun Dec 23 19:33:34 PST 2007
Version b0.8.1
-- add win/MinGW.mak for compiling code under MinGW/MSYS
-- fix ibis::util::makeDir problem with execute bit in directory permission

Tue Dec 11 15:36:04 PST 2007
Version b0.8.0
-- add DLLSPEC macro to invoke dllexport for building DLL on MS Windows
-- update configuration script to better handle java stuff
 
Wed Oct 24 17:18:51 PDT 2007
Version a0.7.9
-- add functions to C API to allow for input of data records to be
   managed by FastBit
-- add functions to JNI interface for input of data records
-- fix a bug in ibis::tafel::write that leaves tdesc nil

Thu Oct 18 18:53:45 PDT 2007
Version a0.7.8
-- add Jave Native Interface for some querying functions and a test program
-- fix errors found on an intel-base Mac
-- fix errors in the ardea.cpp (when run without any arguments)
-- removed -DDEBUG from configuration option --enable-debug

Tue Oct  9 16:52:55 PDT 2007
Version a0.7.7 includes a couple of bug fixes and some minor clean up
-- incorporates a bug fix from Matthias Vallentin to correct a wrong
   function call in ibis::part::reactivate (src/parti.cpp:1261)
-- add test cases for ibis::part::reactivate in ibis.cpp and
   tests/Makefile.am
-- add checks for array_t::m_begin to ensure it is valid before use
-- renamed all functions named dataTable() to partition() to match the
   name of the classes involved

Sun Oct  7 17:21:16 PDT 2007
Version a0.7.6 introduces a new logging mechanism
-- the new class ibis::util::logger is introduced to allow the error
   message to be redirected to a file.  To specify the name for this
   file, use either the function ibis::util::getLofFile or specify
   logfile in a RC file.

Mon Sep 17 17:09:48 PDT 2007
Version a0.7.5 introduces two new features.
-- allow out-dated indexes to be used during query processing.  Since
   the time needed to full update the indexes are typically longer than
   the time need to answer the queries using the raw data, this option
   improves the query response time for the queries that come right
   after the arrival of new rows.
-- allow for rows to be marked as inactive in a data partition.  These
   inactive rows may eventually be re-activated or permenantly purged
   from a data partition.

Fri Jul 27 21:53:57 PDT 2007
-- First public release, version a0.7.


*NOTE* Versions not mentioned here contain bug fixes only.  A more
       detailed revision history is available in file ChangeLog in the
       source code distribution.  The lastest version of source code is
       available at <http://sdm.lbl.gov/fastbit/src/>.