path: root/akregator/src/mk4storage/metakit/CHANGES

2004-01-26    ############################################    MK 2.4.9.3

    Maintenance release, consolidates past 9 months of changes/tweaks.

    Fixed test for "__name__" in python/metakit.py to run a quick test.

2004-01-22    Fixed refcount problem with temp rows in Mk4tcl

    This was a long-standing bug: "mk::row create" did not work right
    because the tracking of temporary rows was completely messed up.
    Added test case for Tcl (mk6.8), fixes FB14, BTS#78, and BTS#29.

2004-01-21    Documentation updates

    Updated copyright notices to 2004, and udated to Doxygen 1.3.5 for
    a new set of C++ documentation.  In anticipation of next release.

2004-01-20    Don't trip over duplicate property names

    Added code in c4_Field constructor to avoid crashing when there is a
    duplicate property name in the format description string.  Duplicate
    names are now ignored (there is no good way to report errors at this
    point).  Avoids an even bigger problem: conflicting property types.

    Added test s49 to detect this case (FB20, reported by Brian Kelley).

2004-01-18    Fixed rare but very serious subview resizing bug

    Fixed rare but serious file damaging bug, when resizing a comitted
    subview to empty, committing, and then resizing back to containing
    data.  Added new s48 test to force this bug to the surface.

    Fortunately this usage pattern *never* happens in blocked views!
    Fixes are at the end of c4_FormatV::Insert and c4_FormatV::Remove.

2004-01-16    Gracefully deal with bad property type specifiers

    When GetAs is a called with a bad description, such as for example
    "myview[name:A]", the release code would crash on a null pointer
    dereference at some point.  Changed so the code now treats any type
    it does not know about as "I" (works a bit better than using "S").
   
    The debug build still hits an assertion, as before.  Added s47 test.

2004-01-03    Fixed typo in PyView.cpp

    Forgot to fix closing brace after the 2003-12-11 mods.

2003-12-21    Fixed Mk4too sorting on subview of length 1

    There was a silly bug when sorting on subviews in the new Mk4too
    interface (not in Mk4tcl), which returns a view, instead of a list
    of integers.  In the case of 1 row, optimization was done wrongly,
    omitting the sort as well as the remapping.  Fixed.
    
    Oomk needs to be patched to work around this (don't sort if n=1).

2003-12-13    Tweak to avoid two unisgned/signed compiler warnings

    In remap.cpp, compiled in debug mode on Win32 (DWORD vs. t4_i32).

2003-12-11    Checked in numerous changes to Mk4py by Nicholas Riley

    All changes are local to the python/directory - for details, see
      http://www.equi4.com/pipermail/metakit/2003-September/001407.html
    With a big thank you to Nicholas for contributing these improvements.

2003-11-23    Bumped to Python 2.3, doc tweaks, lots of name fixes

    Adjusted makefile to now look for Python 2.3 by default.
    Regenerated config files.  Fixed obsolete links in the doc files.

    Got rid of tons of Metakit "interCap" cases, should be Metakit.
    This affects a huge number of source files and one #define (which
    has been defined twice for compatibility: d4_MetakitLibraryVersion).
    Use "cvs diff -i" to see real changes, other than case swapping.

2003-10-28    Get rid of --enable-python, check in c22.txt

    It's not valid, but some files mentioned it.  Use --with-python.
    Forgot to check in the new c22.txt file added earlier this month.

2003-10-16    Added note to Tcl docs

    Added a note to doc/tcl.html on how to load a MK datafiles stored in
    another one, as needed when using VFS for example.

2003-10-10    Added c22 test

    Added test to make sure groubpy/select bug is not present in the C++
    core (bts #75 reports the bug found for Mk4tcl/Mk4too).

2003-10-01    Fixed bugs in Tcl test suite

    The tests in tcl/test/ were incorrectly closing non-test datafiles,
    such as the "exe" one open while tclkit runs, and needed when a
    slave is created which needs to re-run init.tcl etc from VFS.
    
2003-09-30    Python 2.3.1 & cleanup

    Tweaked PWONumber.h and PyRowRef.cpp and MSVC6 project so it can
    compile Mk4py for Python 2.3.1 again.  Windows binary uploaded.

    Removed the win/msvc60/tests/ directory, it's no longer needed.

2003-09-20    Autoconf and libtool rebuilds

    In an attempt to stay ahead of version trouble and other nonsense,
    the autoconf / configure and libtool files have been regenerated.
    Autoconf is at 2.57 and libtoolize is at 1.4.3 on this dev system.

2003-08-26    Documentation fix

    Fixed install comments for Mk4py, as reported in bts #59.

2003-07-17    Fixes to Mk4py (Gordon)

    Fix crash when db.description(nm) is called for missing view "nm".
    Fix wrap - for the special case of wrapping a list of primitives
    (eg, ints, floats, strings) in a single property "view".

2003-07-11    Fix for Linux not finding .lai file

    Posted for RH 8.0 on MK mailing list by Jeff Web on July 3rd.

2003-07-01    Fixed "Metakit" (preferred) vs "Metakit" (obsolete)

    Get rid of InterCappedWords.  Time to grow up, d00d...

2003-06-06    Fix to Mk4py for case (in)sensitivity.

    When using a dict (or keyword args), makeRow now gets the
    names from the dict's keys and gets the properties by name.
    This allows case insensitive matching. Note that using an
    instance still requires that the instance attributes have
    names that match (case sensitive) with the view.

2003-05-15    Add distutils setup.py script (Gordon).

    Tested on Linux & Windows (MSVC).
    You can now do a plain configure / make (without python)
    then cd python; python setup.py build | install

2003-05-08    Fixed array bound bug when not using mmap-ed files

    There was an incorrect test, when dealing with a new file and
    memory mapping is not enabled (which is uncommon, these days).
    Thanks to V Demarco for catching and resolving this bug.

2003-04-28    Sourceforge

    Synced to SF's CVS, see http://sourceforge.net/projects/metakit

2003-04-25    Autoconf/libtool update

    Did "autoreconf -force" with autoconf 2.5.7 and libtool 1.4.3, as
    suggested by Gerfried Fuchs/Ryan Murray.  Some Makefile.in tweaks.

2003-04-22    Fixes to Mk4py (Gordon).

    Make view.append(instance) work again.
    Fix recursively adding subview data.
    Make properties compare properly.
    Initial (incomplete) unittest based test script.

2003-03-16    ############################################    MK 2.4.9.2

    Also allow passing pairs to "mk::row append" as a list (Mk4tcl).

2003-03-10    Fixes for sparc64 configure and AIX strcasecmp

    Both changes contributed by Andreas Kupries.

2003-03-07    Fix more bugs in blocked view, add 64-bit Sparc support

    The blocked viewer deletion was STILL incorrect.  Fixed bad offset
    calculations, added several more assertion checks, and added a new
    m07 test case which checks for numerous cases of deletion overlap,
    i.e. cases where multi-row deletions cross subview boundaries.

    Added __sparc9 #define to mk4.h (thanks again, Andreas K)

2003-03-05    Fixed two potential races in C++ threaded build

    There was a serious bug in view.cpp, whereby Unix (p)thread locks
    were non-functional in release mode (#define NDEBUG).  Yikes!

    In addition the logic was flawed (both Unix and Windows), failing to
    protect from indexing while sPropNames / sPropCounts were being
    resized (i.e. a realloc).  Added explicit locking to all affected
    paths, and removed the now obsolete count locks (i.e. AddRef).

    Thanks to Murat Berk for chasing this and explaining the problem.

    This bug "only" affects C++ builds.  Tcl and Python each have an
    extra layer on top which means there can be no races inside MK.

2003-03-03    ############################################    MK 2.4.9.1

    Moving to a new 4-level bugfix release number.

2003-03-02    Fixes to Mk4py (gmcm)

    Modify some recent patches that were Python 2.2 only so they would
    still work with 2.1. Add a view.properties() method (returning a
    dict of propname -> Property) for cases where a property is masked
    by a method name.

2003-03-01    Reverted changes in Mk4py/scxx, avoid Mk4tcl warning

   One was in SCXX, PWONumber.h - ouch and apologies (jcw).  Changed
   comparison logic in mk4too.cpp (64-bit ints, new code).

2003-02-28    Fixed relaxed tqlayout in Mk4tcl, several Mk4py fixes

   Use list operators to convert a Tcl tqlayout to MK format.  This used
   to crash, Tcl test 6.7 added to verify the fix.  Changed test 3.1 to
   use a different notation for nested defs (^).

   Added to ViewerAsSeq, to allow v1+v2+v3 (thanks Michael Scharf).
   Also fixed several mem leaks - thx again!

2003-02-27    Added support for HPUX aCC "long long"

   Added #ifdef's to mk4.h to match autoconf HAVE_LONG_LONG settings.
   Tweaks to configure.in by Andreas K to better support HP/UX (thx!).

2003-02-26    Added 64-bit int support to Mk4tcl, fix mingw

   The "L" type was not exposed, though Tcl >= 8.4 has "wide" ints!
   Will now support 64-bit ints, if built against a Tcl 8.4+ header, but
   still defaults to without for use in Tcl's <= 8.3.

   Fix _strdup #ifdef in univ.cpp when compiled from mingw on win32.

2003-02-24    Break was missing in switch Mk4py

   Fixed a serious problem, which must have come from editing too
   wildly (Mk4py needs a test suite!).

2003-02-20    Remove a stray include, remove CR's

   Remove "#include <stdio.h>" from remap.cpp, leftover from debugging.
   Also removed CR's left behind from editing under Windows (doh!).

2003-02-18    ##############################################    MK 2.4.9

   This is mostly a bug fix release (some features added to Mk4tcl).

2003-02-18    Fixed a bug in blocked view deletion, and hash byteorder

   There was an off-by-one error in the deletion of multiple rows
   which span more than an entire subview block.  Test m06 is ok now.
   Also added consistency checks (when compiled in debug mode).

   Fixed a too-strict assertion in mk4tcl.cpp, Item destructor.

   Fixed a byte-order sensitivity in hash views for numeric properties.

2003-02-17    Configure tweaks for hpux/ia64

   Re-instated hpux changes by A. Kupries, for ia64, also -lpthread.

2003-02-14    Bug found in blocked viewer modification

   Added test m06 to catch a problem first detected in the Tcl test
   suite (test 5.7) - recoded to C++.  This is a show-stopper for the
   upcoming 2.4.9 release - fixing it is now a top priority.

2003-02-14    Some changes to OO interface in Tcl

   The OO "select" cmd now returns a view, not a list of row #'s.
   Added "product" and "rename" operators, they were not exposed.

2003-02-14    Enable stdio buffering

   On platforms where stdio is used, the setbuf(..., 0) calls have been
   removed (fileio.cpp).  This improves performance, and lets caller set
   up whatever buffering they like.

2003-02-07    Tweaks to restore broken MK ports

   Fixes by A. Kupries to fix MK builds I broke (Itanium/HPUX).

2003-02-07    Changed code to avoid compiler warning

   In remap.cpp, LookDict(), add return at end to avoid a mistaken
   compiler warning (on AIX 5.1).  No effect on runtime behavior.

2003-02-02    Work around optimizer bug in gcc 3.2.1

   The sign-extension in c4_ColOfInts::Get_16r was not being done right
   with -O2 on Linux.  This only manifests itself with files having
   reversed bytes (e.g. PowerPC/Mac).  Changed code to use a local temp
   buffer instead.  This bug was the reason why some starkits failed to
   load when created on Mac and used on Linux.

2003-01-24    Fixed cleanup order bug in Mk4tcl

   There was a long-standing bug in Mk4tcl, in which cleanup order of
   MkPath objects caused them to access their MkWorkspace parent even
   though it was already deleted.  This may have caused some of the
   crash-on-exit bugs in the past.

2003-01-22    Add missing -lstdc++

   For unknown reasons, the current make failed to produce shared libs
   with libstdc++ linked in - manually added to LDFLAGS again.

2003-01-19    Tweak to temp object use

   Two changes in handler.cpp to avoid bug in compilers which get the
   cleanup logic of temp objects wrong (thx, Dan Gregory).

2003-01-17    Add synonym for mk4tcl "info" command

   To avoid a name clash, "$view info" can now also be written as 
   "$view properties" (or an abbreviation of that).

2003-01-16    Allow access to root view in Mk4tcl

   Access to root view (i.e. the storage object itself) was not
   allowed - added a small change to allow this (MkPath::AttachView).

2003-01-15    Use strdup

   On Unix, use strdup i.s.o own code for _strdup, see univ.cpp.
   Perhaps it's time to start reversing the names, and make unix std.

2003-01-10    Build improvements, Mk4py long and Mac improvements

   Merged changes submitted by Nicholas Riley (thanks!) into CVS:
   - changes to build Mk4py on MacOS X (framework/dylib)
   - support 'L' fields, handle overflow, throw exceptions
   - the beginnings of a test suite for Mk4py (in python/test/)

   Separated PPC and 68K projects (Mac classic), because PPC is now
   built with 8.3 (Carbon support), while 68K requires 6.3.  The x86
   cross-compile-from-mac projects have been dropped.

2003-01-09    String compare tweak, Mac Carbon runtime mmap code

   Use strcasecmp on unix, instead of own code (string.cpp).

   Added code based on Paul Snively's contributed patch to recognize
   running under OSX, so carbon apps can benefit from mmapped files.
   All changes are within fileio.cpp, requires CW8 to build this way.
   Added "PPC Carbon" target to cw.mcp, derived from "PPC DLL" one.

2002-12-23    Tweak for Borland builder 5 & 6

   Changed #ifdef line 22 in univ.cpp (fix by S. Cusack).
   
2002-12-09    Fixed bug in selection view change propagation

   Fixed a bug when a row is inserted in a view on which a selection
   depends, when the inserted position is not part of the derived one.
   This looks like an oversight in 2.3/2.4 changes, and must have been
   in there for quite some time.  Added new n14 test to verify this.

2002-12-02    Fixed bug in MK old-file format conversion

   Fixed bug in on-the-fly conversion of old 1.x/2.0.x format files.

2002-11-24    Fixed Mk4tcl threaded build

   Release and reclaim mutex lock while calling eval inside loops.
   Added "--enable-threads" option to configure script.

2002-11-22    Configure tweak for HPUX/Itanium

   Removed ia64 check in configure.in, now that libtool has been fixed.

2002-11-16    Tweaks to compile on Mac

   Small changes to the source code to avoid errors from the MWCW 8.x
   compilers.  Omit stricmp and strdup with CW version >= 8.
   
2002-11-04    Fixed typo in Makefile

   This prevented Mk4py from getting installed (thanks F. Majid).

2002-11-03    ##############################################    MK 2.4.8

   Reverted non-portable change in Makefile to copy to "Mk4tcl.*".

2002-10-28    Workaround for bugs on ARM/WinCE and HP/UX, Mk4tcl/WinCE

   Turn off c4_ColofInts::Set_8i optimization when compiling for ARM on
   WinCE using EVC3.  All regression tests now pass (on ARM & x86emu).

   On HP/UX, enable copying in fileio.cpp, line 265, so a write is not
   done directly from a mmap'ed file (this can hang a process, hard).

   Added a "mktcl" subproject to msevc3, to build "Mk4tcl.dll".
   
2002-10-27    Added multi-thread support for Unix

   Added changes to support same appartment-threading model on Unix as
   on Win32 (pthreads based).  This merely adds support to allow safe
   use of the current "each datafile in one thread" design on Unix.  All
   changes in view.cpp (thank you, M. Berk).  Define q4_MULTI to enable.

2002-10-26    Merged WinCE changes

   Merged changes to build MK for WinCE using MS Embedded VC 3.0.
   These changes are based on Joseph Tate's modifications (thanks!).

2002-10-21    Updated autoconf & libtool (on teevie)

   Files in unix/ regenerated with autoconf 2.53a and libtool 1.4.1.

2002-10-16    Added "dup" subcommand to Mk4tcl OO

   There was no way to duplicate a view in the mk4tcl OO interface.
   Needed to properly deal with re-use / ref counts - added "dup".

2002-10-11    Cast widened in Mk4tcl, support q4_TINY def for Mk4tcl

   Changed cast from int to long, to avoid compiler warning on some
   64-bit machines (mk4tcl.cpp, line 2013).

   Remove float/double code when q4_TINY is defined, also in Mk4tcl.

2002-10-10    Makefile tweaks

   Removed duplicate flags from CXXFLAGS definition.

2002-10-09    Fixed blocked/subview bug, tweak for the ARM platform

   Blocked viewers were not doing the right thing when rows had subviews
   in them - fixed the logic, and add a new m05 test to verify it all.

   Fixed a signed-char bug which prevented MK from passing all tests on
   the ARM, in this case the Compaq iPaq PDA with Linux and gcc 2.95.4.

2002-10-07    Tweak to prevent gcc compiler bug

   Added intermediate temp var in derived.cpp to prevent gcc -O failure.

2002-10-04    Config and makefile adjustments

   Adjusted configure.in to use ".sl" for HP-UX, and ".dll" for Cygwin.

   Changed "install -d" to "mkdir -p" (2x) in Makefile.in, since the
   former is not supported by all incarnations of install (A Kupries).

   Added @CXXFLAGS@ to the end of "CXXFLAGS =" lines in Makefile.in,
   as suggested by Donal K. Fellows.  Used for his IRIX(64) builds.

   Added logic to build properly on HP-UX, including a small assembly
   file (!) which allows loading C++ shared lib from C, as needed in
   Tcl - with thanks to Andreas Kupries for solving and patching this.

2002-09-25    Build tweaks for Mac OS X

   Applied patches by Daniel Steffen to deal with ".dylib" (thanks!).

   Use -O i.s.o. -O2, which caused test b27 to fail on OS X (gcc 3.1).

2002-09-09    More 64-bit platforms recognized

   Added #ifdefs for more 64-bit platforms, thanks to Reinhard Max.

2002-09-08    Make tweaks for HP-UX

   Tweaks to build on HP-UX / 9000 (added __hpux #ifdef in header.h).

2002-09-03    Fixed Mac OSX build problem

   Compile Mk4tcl lib with stubs only if building the shared lib.
   Ignore strip errors (fails with Mk4tcl.so on OSX / dyld libs).

2002-07-01    Python and Tcl installation improved

   Now installs Python Mk4py.so and metakit.py in the most common dir
   location by default, i.e. "/usr/local/lib/python2.2/site-packages".
   Locations can be overridden through $pyincludedir and $pylibdir.

   For Tcl, "make install" now constructs a standard package dir, i.e.
   "/usr/local/lib/Mk4tcl/" with entries "Mk4tcl.so" and "pkgIndex.tcl".
   Locations can be overridden through $tclincludedir and $tcllibdir.

2002-05-30    ##############################################    MK 2.4.7

   Fix CONST84 logic so source compiles under both Tcl 8.3 and 8.4.
   Fix creation of tests/CVS/ so diff in regression tests won't fail.

2002-05-14    Fixed an adaptive int insert/delete bug

   Another bug in the 2.3/2.4 codebase, related to adaptive integers.

   Symptom: one int entry has incorrect bytes after insert/delete.

   Scenario: ints are 1..8 bits, and an odd number are added/removed,
   leaving an odd-sized internal "gap".  Then store a 16/32 bit value,
   forcing resizing.  Once this is done, there will be one value which
   cannot be properly read or set because its data is split *across* the
   gap (commit is ok and removes the problem).  Fix in column.{h,cpp},
   with a new regression test s46 added to make sure things are ok now.

   Let's hope that this is truly the *last* deeply embarrassing bug...

2002-05-06    ##############################################    MK 2.4.6

   Recent bug fix was critical enough to warrant a new revision.
   It is safest to avoid using versions 2.4.2 .. 2.4.5 altogether.

2002-05-05    Fixed major bug in string/bytes after multiple commits

   Finally found a way to reproduce spurious bugs reported in the latest
   revisions.  It turns out that one of the optimizations of the past 2
   months (no exact details) caused memo's to be tracked incorrectly in
   their free space use.  This causes trouble with strings over 10 Kb
   (or smaller if there are over 1000 rows).

   The bug is forced by new (frightfully short) test cases s44 + s45.
   Fixed by performing a slightly less agressive optimization in the
   c4_FormatB::Commit (which is also shared with strings, type 'S').

   Removed a "--exclude" from diff, which is not portable enough.

2002-05-01    Added support for Windows CE

   Patches submitted by Joseph Tate (thank you!), with minor tweaks.

2002-04-29    ##############################################    MK 2.4.5

   Various bug fixes, fixed a number of platform issues.

2002-04-28    Fix small-int re-use view bug

   A nasty bug was reported by VPI which caused upper bytes to be
   truncated from int values.  The problem appears when storing ints of
   1..4 bits in a view, then clearing the view (so a gap is placed past
   the end of the column), then adding a row with an int of 2 or more
   bytes.  This uncovered a bug in forgetting to truncate columns with
   sub-byte int storage.  Now fixed, added regression test s43.

   This bug could also have affected string and byte storage, since
   these use int columns to store item sizes.  Under very specific
   conditions, it may have lead to truncated or even mixed-up values.

2002-04-27    Fix nested mk4tcl loop bug

   Loops would exit prematurely when nested - due to objc/objv being
   overwritten in the inner loop.  Affects mk4tcl.cpp and mk4too.cpp.

2002-04-10    Fix bug introduced in recent blocked viewer optimization

   Murphy was at it again.  The bug affected the way row inserts and
   deletes were done, and can cause incorrect data to be copied.  It is
   relatively hard to reproduce (the test dataset was 12 Mb), but the
   change explains things fully and is in fact very small.
   
2002-04-02    Fix bug in debug code

   In remap.cpp line 531, a debug assertion was moved in the wrong way.

2002-04-01    Backed out to libtool 1.4d, fix test diff and Tcl const

   Backed out to the more official 1.4d release of libtool (instead of
   the CVS version, which is adding "tags" we will not need anyway).
   The unix/Makefile.in has been simplified, back to how it used to be.

   Fixed the "diff" call at the end of "make test" so that it no longer
   generates extra output if things match and now fails if they do not.

   Added fix to allow compiling mk4tcl.cpp with "pre-constified" Tcl
   code, thanks to a tip by Don Porter (see "grep CONST84 mk4tcl.*").

2002-03-31    ##############################################    MK 2.4.4

   Various bug fixes and (blocked view) performance enhacements.

2002-03-28    More blocked optimizations, IRIX tweaks

   Switched Slot() to binary search.  This seems to slow down a few
   percentage point for smaller views, but with 5 millions rows this is
   reported to make a huge difference (from code by Zhang Dehua).

   Added header "bool" fix by Erik Hofman so MK compiles on IRIX (SGI).

2002-03-27    Added definitions for AIX

   Added six operators defs before c4_Cursor class, to avoid compile
   errors on AIX.  With apologies to Murat Berk for taking so long...

2002-03-26    Re-instated the c4_View::RelocateRows operation

   Re-enabled code which supports efficient moves of rows from one view
   to another - avoiding copying of subviews.  Strings/memos are still
   being copied for now.  Also moved a slow test out into a new call
   "c4_View::IsCompatibleWith", this must be checked to make sure that
   RelocateRows can work.  Passing incompatible views will still cause
   an assertion in debug mode, but must be avoided in release builds.

   This change ought to have a dramatic effect on inserts/deletes for
   blocked views with large subviews.  Added examples/blockdels.tcl to
   thoroughly exercise this new code and all boundary conditions.

2002-03-22    Fixed a serious bug in serialization code

   When serializing string/bytes columns with large strings using the
   c4_Storage::SaveTo function, memo's would sometimes not be written,
   leading to a *damaged* datafile (and incorrect free-space tracking).

2002-03-15    Better configure logic, "mk::view tqlayout" fix

   Many tweaks to configures, makes, and libtool setup.  Get rid of the
   library version numbers and the special Mk4*.so targets.  Instead of
   Mk4tcl.so use libmk4tcl.so (likewise Mk4py.so is now libmk4py.so).

   Fixed "mk::view tqlayout", it was broken by the 10-2-2002 change.

2002-03-13    Extend partial access 'B' usage

   Added code to Access and Modify to simulate partial data access the
   hard way when the underlying view does not support it (c4_Bytes
   only), i.e. copy via full temp.  Added test s41, doing partial access
   on a blocked view.

2002-03-12    Add test for serialized input

   The c4_Storage::LoadFrom() call (mk::file load) can be tricky.  After
   a commit, the header is adjusted so that reading from the start
   works.  This is not the case in commit-extend mode, where a load will
   read th state as it was before extending - it cannot know there is
   more.
  
   Added test s40 to verify this case.  Also reorged a few more files
   (tbasics and tstore4 split into 2 each).

   Added test in mk4tl.cpp so LoadFrom failure generates an error
   return, instead of being ignored.

2002-03-10    Tweaked March 8 fix a tad further

   The workaround implemented below wasn't 100% complete.

2002-03-08    Workaround for file extend on Win

   There seems to be a bug when extending a file on at least NT4 with
   NTFS (file server works fine).  When extending the file in such a way
   that a gap is created to save new data during commit, the file
   appears to get rounded up to the next 512-byte boundary.  At that
   point, the datafile no longer has a valid end marker and is sort of
   corrupt.

   Wrote workaround for Win32 builds, see persist.cpp notes names "March
   8, 2002".  One way to force this bug is to run "mkhash Dmhs 250000" -
   after row 150000, it asserts on bad filesize when built in debug mode
   and run on NTFS.

2002-02-19    Fixed over-aggressive optimization

   Change made on Feb 7 was releasing too many objects in a commit,
   causing it to detach empty top-level views, even when in use.  Fixed,
   added s39 to test for this case.

2002-02-10    Improve Mk4tcl's "mk::view tqlayout"

   Avoid crash when asking for the tqlayout of a non-existent view.  Now
   returns an error instead.

2002-02-07    Fast commit with many empty subviews

   Avoid creating subviews when committing, if these are empty anyway.
   Added new s38 test for this case.

2002-02-02    Small optimizations

   Changed a few c4_Property instances to const& references to avoid
   copying (2 in view.cpp, 2 in custom.cpp).

2002-02-01    ##############################################    MK 2.4.3

   Bug-fix release, mostly.

   Python include path now upgraded to python2.2 (was 2.1).

2002-01-31    Cross-platform serialization, Tcl

   There was a bug with serializing a datafile (SaveTo) when it was
   created on a platform with reverse endian-ness.  Fixed so serialized
   data also flags reversed byte order.

   Tweaks to fix const changes in the latest Tcl cvs branch.  Fixed a
   recently-introduced UTF8 path bug in mk4tcl.cpp.

2001-12-21    Optimized GetAs

   Now that GetAs is used so much more, optimize the common case where a
   description does not require restructuring.  Can lead to order-of
   magnitude speed improvement in cases where a storage contains many
   views.

2001-12-20    Fixed bug in Locate, comparison issue

   Testing for the Mk4py changes uncovered a serious bug in
   c4_View::Locate, causing it to sometimes return zero, even if there
   are matching rows (thanks, Gordon).

   But a very fundamental weakness also showed up, being that row
   comparisons (and that includes the C++ operators) have the confusing
   property of not being symmetric in all cases.  The problem occurs
   when left- and right-hand sides do not contain the same (number of)
   properties.  In that case, the *left-hand* row participating in the
   comparison determines which properties take part in a comparison.

   In the case of Locate, this caused improper comparisons.  And it is
   very easy to get bitten by this, such as here: c4_RowRef key = ...;
   int n = view.Search(key); bool match = n >= 0 && view[n] == key; The
   above code is *wrong*.  The last line needs to be: bool match = n >=
   0 && key == view[n];

   This is very clearly a design mistake.  Comparisons should have been
   *either* implemented *or* named differently.

   A new "m04" test has been added to the regression suite.

2001-12-19    Changes to mk4py by Gordon McMillan

   Several changes and cleanups.  Mk4py now has logic to track different
   view categories, e.g. to make sure a R/O view is not being written
   to.  This should greatly reduce the number of silently ignored
   incorrect calls, as well as crashes, and will produce appropriate
   error messages instead.

2001-12-18    Cleanup

   Cleaned up source comments and got rid of yet more warnings.

2001-12-14    Fixed yet another case of crash-on-exit

   The new Unmapped() code of 2.4.2 forgot one case of cleaning up,
   which has now been fixed (in c4_FormatB::Commit).

2001-12-12    ##############################################    MK 2.4.2

   Better portability, hashing improved

   This release marks the consolidation of a number of changes, mostly
   relating to better portability & hashing.  The speed of commits with
   many strings and subviews should be notably better.  The Tcl
   extension no longer needs a "stub" library to compile (it now has
   that code itself), just std headers.

2001-12-08    Changes to commit cleanup, Mac stuff

   Changed the logic of how ReleaseAllSegments gets called at the end of
   commits.  There was at least one case of leaving a column pointing
   into mapped file space when it was about to be remapped.  This should
   fix a very long-tanding bug which shows itself as freeing unallocated
   memory during commit or cleanup of the storage object.

   Changes to test coding and PyRowRef.cpp to deal with builds on
   Macintosh (different issues for MacOS 9 and MacOS X).  The
   mac/cw.sea.hqx project has been upgraded to MW CW 6.3.  Verified Tcl
   8.3.2 and Python 2.1.1 builds with CW6 on Mac.

2001-12-06    Tweaks to Tcl interface

   More robust, added "$vw loop var ... {body}" object command.  The
   "$vw size" command now takes an optional newsize arg.

2001-12-04    Tweaks to makefile and configure

   Tweaks, in preparation of an upcoming 2.4.2 release.

2001-12-03    Changes in M4py, avoid gcc problem, hash

   The "-fomit-frame-pointer" option for gcc has been turned off,
   because it causes problems with exception handling in Mk4py.  All
   failures in Mk4py now propagate properly to Python AFAIK.

   Simplified Mk4py - by removing a layer of exception handler classes
   in scxx.  Errors now set info and throw a plain int.

   The mkhash.cpp sample program exposed a problem with multi-key
   hashing: the order of properties in the search key must match exactly
   the order in the hash view itself.  For now, this has been left as is
   (it's easy to do, once you are aware of it).

2001-11-30    Win MT fix, commit tweak, indent cleanup

   Drop static buffer in fileio.cpp (DataWrite), uses stack now, so the
   code can be used in MT context even on Win 95/98.  It does more
   copying than would be needed for NT (2K?), alas.

   More changes in c4_FormatB::Commit to properly detect memo use in
   blocked views.  The recent changes introduced a bug which shows up
   only with blocked views and large string/bytes items.  Reported by
   Steve Baxter with demo, new "m02" regression test.

   Changed fileio.cpp to turn off file buffering, this avoids a few
   reads when writing and seeking a lot.  It does not have as big an
   impact as one might expect, but every little bit helps.

   Cleaned up new 2-space indentation in several source files.

   Added new "mk::file space" in Mk4tcl, to inspect the current file
   space usage.  This is only intended for internal testing.

2001-11-28    Fixed memory leak in string/memo cleanup

   There was a mem-leak in c4_FormatB::Commit which showed up due to
   yesterday's more extensive testing.  Only showed up after a commit,
   in string columns with widely varying item sizes.  This caused
   regression test s37 to fail in MFC-debug compile.

2001-11-27    Major performance bug fix, and MT strings

   A serious problem has been resolved, which slowed down commits, and
   prevented blocked views from committing efficiently.  The reason was
   that for string props, the string size was always being saved anew,
   even if no changes in that view took place.  This did not affect
   proper operation, just speed, and was most noticeable with many
   (sub-)views containing many string props.

   The solution is in src/format.cpp, the examples/mkhash.cpp code was
   further adjusted to better expose and measure the effects.  Thanks to
   P. Baspeyras and S. Baxter for helping me resolve it.

   Another change was to alter the way empty strings are allocated in
   the src/string.cpp code, making it compatible with multi- threaded
   use and removing the remaining memory leak.

2001-11-26    Fix in assertion check (blocked view)

   Corrected an off-by-one bug.  Only shows up with debugging on, since
   it's inside an assert().  Thanks to Steve Baxter.

2001-11-25    More arg checking in Mk4py, locate

   Added more checks against incorrect usage, based on sample code by
   Mitch Chapman.  The "throw" code appears to be inconsistent when
   called at top of *some* calls, using a workaround for now.

   Added "view.locate(key)" wrapper, returns (pos,count) tuple.

   Various source code formatting adjustments (indents and such).

2001-11-04    Added alternate calls to c4_CustomViewer

   Added extra defs to "mk4.h" of Lookup and InsertRows which take a
   "const c4_RowRef&" i.s.o. a c4_Cursor.  Inlined in "mk4.inl".  The
   "c4_Cursor" datatype might one day become obsolete.

2001-11-03    Removed tcl/kit/ and copyright notices

   Removed the entire tcl/kit/ tree, Tclkit will be distributed as
   separate package from now on (the 2.4.1 release still builds ok).

   Also replaced all copyright notices by version Id's and an URL.

2001-11-02    Fixed partial memo commit bug, makefile fix

   Modifying a small item as memo (i.e. through Modify) properly
   committed a change, but subsequently left an incorrect pointer after
   the commit.  Fixed, and added test s37 to catch this case.

   Don't strip symbols from installed static libs (whoops!).
   
2001-10-31    Fixed Mk4py error flag clear on delete

   When deleting rows, a slice was constructed from a PWOSequence, which
   generates an otherwise harmless error when its length is checked.
   The flag was not cleared, causing errors in subsequent Python
   statements.  Changed to a PWOTuple in PyView.cpp (2x).

2001-10-26    Fixed Mk4tcl re-open test

   The tcl/test/mk1basic.tcl test #6 was reporting open failure on
   non-Linux systems.  Fixed, the code works, the message was wrong.

2001-10-19    Rearranged some demo files, Lua binding

   Rearranged some of the demos/samples from the python/ and tcl/ areas,
   and placed them in examples/ instead, for consistency.

   Added "selmap.tcl" to illustrate how to turn select into a view.

   Added a basic Lua binding (incomplete, but it's a start) in lua/.

2001-10-18    Fixed recent hash bug, and add to ordered

   The recent "fix" to deal with hash misses introduced a huge bug,
   causing the mapping to be recalculated on each insertion.

   Adding to an ordered view did not always work, because the code was
   based on SetAtGrow, which is not suitable for an ordered view in
   which the row position is determined implicitly.  Fixed by changing
   c4_View::Add to use c4_View::InsertAt instead.

2001-10-14    ##############################################    MK 2.4.1

   Custom-extended Tclkit, and threading

   Minor kitInit.c change allows wrapping apps as a "custom-extended"
   version of Tclkit.  Enable threaded Tcl build on Unix.

   This release consolidates the most recent fixes and code tweaks.

2001-09-27    Fix temp storage open in Mk4tcl

   A recent change made it impossible to open temporary storages from
   Tcl ("mk::file open db") - <blush> - now fixed again.

2001-09-19    Bug fixes

   Fixed a problem during commit, when memo's are modified, causing them
   to be stored inline again.  This caused a late reference to mapped
   memory - leading to intermittent crashes on commit.

   Changes/fixes in mk4tcl.cpp, to prevent a problem when a datafile is
   re-opened again under the same name later on.  Old references could
   in some cases end up stale - leading to crashes much later.

2001-09-05    Bug fixes

   Hashing bug fixed: failed to terminate when looking for a missing key
   after a hash collision.  The fill count was not being tracked
   properly to enforce that there is always at least one unused slot.

   Fixed bug when setting a string from a higher row: the data gets
   trashed, because too little copying was done while creating a gap.
   This bug may have led to some other cases of "damaged" datafiles.
   Solved by copying input data in c4_FormatB::SetOne().

   Fixed long-standing bug with commit-after-load, by now doing a full
   view/subview copy.  This is inefficient, but it avoids all sorts of
   mapping/strategy problems.  This also fixes test s33 (at last!).

   Added regression tests b27, c21, m01, and s36 to check above fixes.

2001-06-29    ##############################################    MK 2.4.0

   More changes to the Mk4py interface

    - allow setting a row to a value if the row has a single property
    - there is an unexplained crash when setting slices with wrong type

2001-06-22    Changes to the Mk4py interface

    Modifications and fixes gratefully accepted from John Barnard:
    - c4_LongProp support ('L' datatype)
    - row.__attrs__ returns the list of all properties
    - row.__view__ returns the container view of the row
    - row.__index__ returns current position in view
    - view.setsize(n) added, extends/truncates number of rows
    - generalized makerow to allow any sequence, not just lists
    
    Changed PyRowRef to inc/dec the reference it has to the underlying
    view.  Should prevent dangling pointer problems, such as deleting a
    storage while rows are still in use.  This adds a little overhead.

    Fixed a PyErr_Clear issue when accessing non-existent properties.

2001-06-12    Close DB filedesc on exec

    Added an fcntl call to fileio.cpp, so that on Unix database file
    decriptors are closed on exec (relevant when doing "exec ... &").

2001-05-30    Fixed mem-leak in Mk4tcl

    A long-standing bug, in TclSelector::ExactKeyProps.

2001-05-28    Security fix in Tclkit

    See end of the tcl/kit/README file for deatils.

2001-03-30    Fixed long-standing commit bug

    There was an intermittent bug in c4_Persist::Commit, when properties
    were "restructured away" (dropped).  The bug was hard to track down,
    because it depended on what address ranges the O/S assigned to
    mem-maps.  Might also fix other spurious commit/exit crashes.

2001-03-29    ##############################################    MK 2.3.4

    The "last" release candidate

    Now checked into its *new* CVS home at equi4.com.  Mailings list(s)
    have also been moved to this site.

2001-03-28    Fix Win build, broken on Mar 27

    Dropped kBufMax from mk4.h, it caused compiler errors in MSVC6 -
    switched to "sizeof" in a couple of places.  This error was
    introduced by the double-fix of 3/27.

2001-03-28    Allow builds of Tclkit with Sun CC

    Integrated a few changes provided by D.J. Hagberg.  Note: the
    M-solaris.sh and M-dyn.sh scripts need to be manually edited when
    choosing between CC and gcc.

2001-03-27    Double-alignment bug on Solaris

    Two changes (c4_Bytes in mk4.h and src/column.h) to fix an alignment
    problem for 8-byte doubles on Solaris.  This caused tests b17, b23,
    b24, s22, and s28 to fail.

2001-03-26    Fixed cross-platform commit bug

    There was a nasty bug in the 2.3.x code, which wrote incorrect field
    sizes when committing to a datafile with a different byte order
    (i.e. created on a machine using different endian-ness from the one
    doing commit).

2001-02-14    Removed file events from Tclkit

    Changed kit/rechan.cpp to not generate file events.  This avoids a
    bug in Tclkit whereby an open file can generate a continuous stream
    of file events as long as the file is open - the console will seem
    to be frozen, though "close $file" does work and fixes it.

2000-12-13    Added missing c4_LongRef export

    Added a line in mk4dll.h to resolve references to the new
    c4_LongProp/c4_LongRef datatype on Windows.

2000-12-04    Fixed conversion bug

    There was a bug in c4_FormatV::OldDefine, causing MK to crash when
    trying to convert an old-format file with empty subviews in it.  Bug
    introduced on 14 Nov.

2000-12-02    Fixes in Tclkit code

    Fixed a Tcl-level bug causing memory leaks for all compressed files
    stored with MK opened for reading.

    Disabled event sources (mk4tcl.cpp and rechan.cpp) to avoid a GUI
    freeze-up while a fake file is open.

2000-11-16    Simplified c4_Storage

    Made some changes so that a c4_Storage no longer has a separate
    c4_Persist* copy.  As a result, a storage can now be reconstructed
    from any root-level view.  This generalizes views, and prepares for
    a merge.

2000-11-14    Fixed mem leak in conversion code

    The conversion code from pre 2.3 files had a memory leak in
    c4_FormatV::OldDefine, causing f06 to leak, as well as several
    subsequent erroneous leak reports.  Changed a few "new" to "d4_new"
    calls along the way.

    Remove FlipBytes members, they are no longer needed.

2000-11-08    Tclkit now in the distribution

    The "Tclkit" project has been merged into Metakit.  Details and
    updated build info at "tcl/kit/README".  The SourceForge CVS
    repository is up to date again.

    Mk4py: improved number conversion and error handling.
    
2000-11-03    ##############################################    MK 2.3.3

    First final release candidate

    Yes, there will probably be a second one as well...

2000-10-31    Improved error handling

    More logic added to catch errors in flush and streams.  This affects
    the C++ core as well as Mk4py and Mk4tcl.  API of c4_Stream::Write
    changed to return success flag.

2000-10-30    Added autocommit call to Mk4tcl

    To better support VFS, Mk4tcl now has a new command "mk::file
    autocommit <db>" to force commit on close.

2000-10-26    Changes to compile with Borland C++

    Minor tweaks to compile with Borland C++ Builder 4.0, which does not
    support "long long".  A new Kitviewer has been built (new code to
    replace c4_View::Describe).

2000-10-03    Fixes for Alpha Unix

    Moved _item in column.h up to fix alignment sensitivity.  Config.h
    did not get SIZE_LONG right on Alpha Unix (0?).

2000-09-27    Contributed fixes and Python 2.0b2

    Adjusted makefile to build with Python 2.0b2 release.

    Source code tweaks to avoid DEC CXX 5.7 compiler errors.  Add
    no-inherit flags for Win32 to not leak file handles.
    
2000-08-27    Allow derived row deletes in Mk4py

    Added code to PyView.cpp to handle deletes (and slice deletes) in
    derived views, see "examples/derived.py".

2000-07-30    Major auto-convert 1.8.6 file bug

    Bug in on-the-fly conversion of bytes properties ('B') in pre-2.0
    datafiles (i.e. 1.8.6 and earlier) resolved.

    Unfortunately, this bug can not be 100% unambiguously fixed.  The
    new code *will* properly detect most cases, and convert both 1.8.6
    and 2.0 datafiles on the fly, but especially for views with only a
    few rows and at most a few bytes of data per row - the conversion
    *might* fail.

    In this case, MK will have to be compiled with a define to force it
    to either assume all old datafiles are 1.8.6 (-Dq4_OLD_IS_PRE_V2),
    or to assume that they are always 2.0 (-Dq4_OLD_IS_ALWAYS_V2).  If
    you are currently using MK 1.8.6, then you should *skip* the update
    to 2.01, and consider updating to 2.3.x.  This way you never have
    any 2.0 files around, and can force all your code to handle 1.8.6
    files properly (by using "-Dq4_OLD_IS_PRE_V2").

    See src/format.cpp, c4_FormatB::OldDefine for details.

    This bug *only* applies to bytes properties in pre-2.0 data files.
    Conversion of 2.0x files is unaffected.
    
2000-07-25    Fixed new self-referential views

    The new recursive / self-referential view definition style has been
    fixed, e.g. "view[data:S,self[^]]" will now let you store a tree of
    arbitrary depth, with each 'self' subview having data and self
    properties.  See the demo in "examples/selfref.py" to see how this
    all works.

2000-07-22    Fixed bug in double restructuring

    Solved a very long-standing bug in restructuring, which caused
    incorrect (non-zero, small) default values when a c4_DoubleProp was
    added to a view which already had rows.  

2000-07-18    Added remapwith and pair to Mk4py

    Exposed C++'s c4_View::RemapWith as v1.remapwith(v2), and
    c4_View::Pair as v1.pair(v2) in the Mk4py Python binding.  Added
    pair.py, remap.py, and wrap.py in "examples/" dir.

2000-07-12    Added metakit.py wrapper

    Added "metakit.py" script to wrap Mk4py, including a new
    metakit.dump() to pretty-print views.  More utility code will be
    added over time.  The preferred way to use Metakit from Python is
    now "import metakit".

2000-07-06    Conversion fix, warning cleanup

    Fixed on-the fly conversion of old datafiles.  The free space was
    not managed properly - changed to never touch any data inside the
    file during conversion.

    Some source code change to get the compile through gcc flags
    "-fguiding-decls -Wall -pedantic -Wno-unused".  The only remaining
    complaint is about using "long long".

2000-07-04    MkSQL subtree, "indexed" mapped viewer

    Added the sql and mksql subtrees to the distribution, with Gordon
    McMillan's MkSQL engine, written in Python.  The "isql.py" script is
    a simple interactive shell around it.

    Started work on a new viewer which maintains a persistent index (as
    a one-int-prop permutation), see src/remap.cpp.

2000-07-03    Mk4tcl fixes

    Fixed view rename problem and "delete end" (Matt Newman).  Adjusted
    the tests in tcl/test/mk5object.tcl accordingly.

2000-06-30    Tequila fixes

    Close fix and failure handler (Steve Landers).

2000-06-29    ##############################################    MK 2.3.2

    First beta release

    The new release is 99% feature-complete.  What remains is to further
    document C++/Python/Tcl use and to fix bugs.

    Python sample code in "python/aside.py" and "python/find.py".  Tcl
    samples in "tcl/test/mk5object.tcl" and "tcl/mapped.tcl".

2000-06-28    Hash/blocked/ordered: changes and fixes

    Changed hash insertion to insert at specified position.  This makes
    it possible to use hashes "under" ordered views.  For best
    performance, insert rows at end of hash views.

2000-06-26    Documentation, example, Mk4tcl OO fixes

    Moved C++ member documentation out of "mk4.h" header.  Added
    examples/ directory, with a find.py timing example.  Fixed bugs in
    new Mk4tcl: "$vw find" and "$vw delete end".  Added tests for new
    Mk4tcl OO interface: ":mk5object.test".

2000-06-16    Improved modifiable custom viewers

    The Pair and Slice viewers now support set/insert/remove, while
    RemapWith/Concat/Rename support setting values.  

2000-06-15    Many changes to the Tcl code, hashing

    Contributed by Matt Newman, it adds support for most custom viewers,
    including the new hash etc.  Added a fast find.  Caveat: most old
    custom viewers are still not modifiable.

    Hash calculation improved, far less collisions than before.

    Blocked viewer seems to work.  Ordered on top is sub-optimal.

2000-06-12    GetUpperLimit, Blocked, Ordered, mk4too

    Removed c4_View::GetUpperLimit (it's equivalent to GetSize-1 and was
    not being used anywhere).

    Start implementing c4_BlockedViewer, a simple balanced/blocked
    nested data structure.  Also started on a c4_OrderedViewer, which
    keeps the underlying view sorted during changes.  These two can be
    combined to implement an efficient 2-level btree.

    Adopted code by Matt Newman for oo-cmd's for Mk4tcl views.  First
    trials work, started to extend with new custom viewers.

2000-06-09    Change case of a few Mk4py members

    Changed all top-level members in the Python interface to lower case:
    storage, property, view, wrap (will break existing code).

    Fix bounds check in Mk4tcl.cpp for commit/rollback (new code).

    Whoops, forgot to add new src/remap.{h,cpp} to the cvs tree.

2000-06-08    Implemented hash lookup

    Added a new virtual c4_Sequence::RestrictSearch, which lets a view
    take over searching (used by c4_View::Find).  The result is that the
    new hash viewer gives a huge speedup for finds.  Find requests which
    require linear scanning are unaffected.

2000-06-07    Documentation extraction based on Doxygen

    The automatically generated output from Doxygen is working out well
    and looking pretty, added "src/doxy.h" with more comments.

2000-06-05    Started hash and btree custom viewers

    The hash implementation is nearly done, also usable from Python.
    Btrees are being implemented as fixed 2-level for now.

2000-06-01    Fixes

    Fixed crash when opening missing file r/o new in 2.3.1 alpha.  Fixed
    incorrect on-the-fly-conversion of 2.0 format subviews.  Added
    "storage.aside(storage)" to the Python interface.

2000-05-30    Fixed commit-aside

    The new commit-aside code was botched by recent changes.  Fixed
    SetAside to pick up new root seq, changed by implicit rollback.
    Note that a commit-aside is not finished until you *also* commit the
    secondary file containing all newly generated changes.

    Fixed bug in c4_BytesRef::Access, introduced in 2.3.1 alpha.

2000-05-29    Added new 64-bit long datatype

    Added support for 64-bit longs (type 'L'), and c4_LongProp, etc.
    This type is not autosizing, it always uses 8 bytes per entry.  Uses
    "long long" or "__int64", else defines struct with 2 longs.  This is
    not yet correct for platforms which have no 64b ints.  No regression
    tests or Python/Tcl interfaces yet.

2000-05-28    Better file mark scanning

    Added c4_Strategy::EndOfData, to determine the logical end of a
    Metakit datafile.  This call can be used to check whether a file
    contains any data, and whether a commit-extend has been performed.
    Old-style scripted documents (with preamble) can now be opened.
    Changed strategy class, DataSeek has been merged with read/write.
    Removed c4_LoadContext, LoadIt member moved to c4_Persist.

2000-05-27    Tweaked configuration define's

    Make the release build the default (no assertions, use inlines).
    Enable booleans for gnuc by default (it's pretty standard by now).
    Added extra include path to better find Python's includes.

2000-05-26    Fixed Tcl dependency

    Changed configure script to no longer look for Tcl if the
    "--with-tcl=..." parameter is not specified.

2000-05-25    ##############################################    MK 2.3.1

    First alpha release

    Officially, this is "Metakit 2.3.1 alpha" (ignore "2.3.0" in mk4.h).
    All alpha's are 2.3.1 (beta's will be 2.3.2, finals start at 2.3.5).
    Builds on Linux/Mac/Win appear ok - as do Mk4py, Mk4tcl, and Tclkit.

2000-05-06    Massive changes to the core

    To summarize the main issues: management info is now stored in such
    a way that it need not be read in right away - file open is now
    instant.  The S(tring) datatype is now stored as B, making it far
    more scalable (API/use is unchanged).  Storage objects now derived
    from views, both can be initialized from a stream (data will be kept
    in a buffer, beware of potentially large memory use).

    Several file format changes are "for future expansion".

2000-05-05    Fixed builds without Tcl

    In 2.0.1, the make would fail if there was no Tcl to build with, or
    not an appropriate release.  Changed Makefile to report and skip Tcl
    builds in that case.

2000-04-06    Fixed a nasty restructure/mmap bug

    When a property is deleted by a restructure, then committed, then
    later committed again, a problem can cause MK to crash.  It has only
    been detected in debug builds, but the problem turns out to be a
    fundamental one (only happens with memory-mapped files, if the file
    is resized).  Fix in next rel (c4_HandlerSeq::DetachFromStorage).

2000-04-02    Memo properties are no longer needed

    The M datatype is gone from the public API (and now illegal).
    Everything binary should now be stored in B(ytes) properties, which
    then adaptively decide which internal format to use, based on a
    simple heuristic (which will be refined later).  Existing datafiles
    will automatically convert from M to B.  The partial Access and
    Modify calls now also work on B items.

2000-03-30    Minor change in c4_Strategy

    Dropped the _keepAfterCommit flag in c4_Strategy, it probably has
    never been used and it interferes with new features.

2000-03-27    Change in API for creating storages

    It is no longer possible to create a storage and define its
    structure with a single call.  Instead, open a storage in r/w mode
    (i.e. "1") and then call the (now public) SetStructure member to
    define the structure of all views.  This change is necessary to
    prepare for the upcoming "commit-aside" logic.

2000-03-23    File format changes (in progress)

    The new file format has an incompatible header, so old code will not
    recognize new datafiles.  Major changes are: added a file tail
    marker, the serialized format is now a very good way to compress
    datafiles, since it can be efficiently opened in on-demand/mmap'ed
    mode.  The new format supports several planned features.  Code to
    convert existing files on-the-fly will be added before this change
    is released.

2000-03-19    Added c4_Strategy::FileSize

    The Strategy::FileSize call is used for a file format change.

2000-03-18    Added c4_View::Locate

    Locate returns the number of matching rows, and optionally the
    position of the first one, using binary search.  Like the
    c4_View::Search function, it requires the view to be sorted.

2000-03-17    ##############################################    MK 2.01

    Maintenance release, it's solid

    Updated MK version number to 2.01, this maintenance release
    represents a very stable version.

    Small change to b07 test to avoid evaluation order problems.

    Added unix/metakit.spec file for RPM, thanks to Sean Summers.

2000-03-16    Drop Store, fix deep copy, drop segments

    c4_Storage::Store never worked properly under all conditions.  It's
    been deprecated for some time and has now been removed.  Made a
    handful of changes to test- and demo code to drop it.

    Duplicating a view with deep copy never worked, because it used the
    buggy Store call as well.  Changed to use recursion.

    Dropped support for segmented tree-walk storage, which hasn't been
    used since 1.5 (use a commit with 2.0 to convert files).  This is
    necessary to prepare for some file format changes.

2000-03-15    Modifiable custom viewers, other tweaks

    A start has been made with making custom viewers updatable.  The new
    methods are Set, InsertAt, RemoveAt, and Move, but the number of
    viewers which implement this is still limited.  Mk4py has been
    adjusted to allow "set" on wrapped views.

    Removed c4_Strategy::DataLoad, it was only used in one place.  Small
    optimization of the 2 calls to c4_Streamer::NextByte.  Get rid of /
    disallow read calls on memory mapped files.

2000-03-14    Makefile tweaks, non-commits smarter

    Changed from --enable-tcl to --with-tcl=DIR, because the old
    approach only worked with Tcl installed in a standard place.

    Commits of a R/O file now fail.  Also, if no changes have been made,
    a commit will no longer write anything to file.

2000-03-13    Several new commands added to Mk4py

    Several changes were submitted by Gordon McMillan, which add better
    support for his upcoming SQL engine.

2000-03-12    Allow embedding MK datafile at end of EXE

    Mk4tcl was changed to look for an optional trailer for quick access
    to the start offset.  This makes it possible to append datafiles to
    executables, even if they are larger than 4 Kb.

2000-02-29    Fixed rare bug with lots of memo fields

    There was a bug in free space management (persist.cpp), which can
    only occur when exactly 7500 free space gaps are present, and a
    commit crosses the threshold.  There was also a small mistake in
    that same code causing a bit of free space waste.

2000-02-24    Added proxy support to Tequila

    Tequila can now be used as basic client/server setup for Tcl
    scripts.  See tcl/tequila/README for details and an example.

2000-02-04    Fixed mk::views (Mk4tcl)

    The mk::views command failed to list the first view in the file
    (this bug was introduced by changes in MK 1.99).

2000-01-02    Adjusted y2k

    Copyright and license dates adjusted.

1999-12-26    More Mk4py changes

    Added sortrev, and fixed "select(low,high)".  Docs updated.

1999-12-23    New view operators in Mk4py

    Added rename, project, groupby, and counts operators to Mk4py.
    These were already part of the C++ core.

1999-12-22    Avoid GetId inline warning

    Reordered GetId in "mk4.inl" to avoid (harmless) inline warning.

1999-12-21    Checked-in Catfish and Kitviewer sources

    Added win/catfish and win/kitviewer areas.  Catfish was built with
    MSVC 1.52, so the win/msvc152 area has also been added.

    Kitviewer requires Borland C++ Builder 4.0 to build (using VCL), it
    has been adjusted to now also recognize scripted documents.

1999-12-20    Bug fix in set-after-get situations

    A bug has surfaced when setting string/byte/memo values which span a
    4 Kb block boundary.  The bug can only happen if data is first
    fetched and subsequently changed.  The affected code is in
    src/format.cpp (3x).  Added Tcl test 5.5 to catch this.

1999-12-19    Mac tweaks

    Changes to make the Mac versions build from the CVS repository.  The
    Mac can also cross-compile Windows libraries using MWCW 5.

1999-12-17    Add the Tequila example

    Added the Tequila global Tcl array data server, see tcl/tequila/.

1999-12-15  MK 2.0      Official Open Source release

    Removed a bad assertion from FormatX::Compare.

    Several new services set up on the excellent SourceForge.com site.

1999-12-14    Documentation added

    The C++ API documentation has been added to the distribution, as
    well as a document describing the file format details of Metakit.

1999-12-13    Bug fix affecting c4_View::Description

    There was a problem with c4_ViewScanner::Describe, due to a change
    from c4_String to (const char*).  Now c4_View::Store works again.

1999-12-12  MK 1.99     New release, as open source software

    The major change is that Metakit has been released as open source
    software, based on the liberal X/MIT-style license.  Commercial
    support remains unchanged for all recent commercial customers, and
    for those who purchase the Enterprise License.  The Universal Source
    license has been terminated, because full source code is now freely
    downloadable by anyone from the website.

    Sources and documentation files have been adjusted accordingly.

1999-12-08  MK 1.9h     Bool support for gcc/egcs, minor fix

    Clear _field after delete in c4_HandlerSeq::DefineRoot.  This
    triggered an assertion on Linux, when compiled in debug mode.

    Added pre-processor logic to detect whether gcc supports bools.

    Removed all indentation from #define's, #ifdef's, etc.  This was
    done after a report that some compilers can get confused by this.

1999-12-06    Derived view row copy fix

    There was a problem when using SetAt with derived views as source,
    due to a remapping problem.  Fixed viewx.cpp, added new test b25.

1999-11-25  MK 1.9g     Makefile changes, thread-safe Mk4tcl

    Renamed options to --enable-python and --enable-tcl, both now off by
    default, since most people probably don't want to enable both.

    The new "-shared" changes to make Mk4tcl thread-safe have been
    folded into the main source code.

1999-11-22    Channel improvements Mk4tcl

    There was a close conflict in mk::channel, also several changes to
    improve mk::channel fileevent handling.

1999-11-19    Bug fix in Mk4tcl

    There was an array overrun when mk::get was called without fields.
    Added code to avoid this.

1999-11-11    Mk4tcl exit handling

    Simplified Mk4tcl exit handling, fixes "interp delete" crash.  Added
    new mk5fixed.4 test to catch this case.

1999-11-10    Mk4tcl shared and multi-threaded access

    Made a first experimental version of Mk4tcl (1.2.1, not announced)
    which allows sharing a database between interpreters and threads.

1999-11-09    STL, MkWrap, compare caching

    Tweaked the makefile to support STL builds (also adjusted README).

    Fixed bug in MkWrap, calling storage.description() without args.

    Yet more fixes in comparisons, this is all one problem, caused by
    changes in caching for ints, floats, and doubles.  Added B24 test.
    These (last?) problems occurred in custom / compound views.

1999-11-08  MK 1.9f     Fixed sort comparisons

    There was a cache problem with comparisons of int / float / double
    sorts.  Fixed, also added new B23 regression test to check for it.

1999-11-07    Little nasty details

    Tweaked some make/project files.  All regress tests and Tcl tests
    now reported to really work on Solaris, Linux, Windows, and Mac.
    Regression tests required more memory on Mac to get past L03 case.

1999-11-06  MK 1.9e     Big oops: Mk4tcl and MkWrap were broken!

    Fixed a c4_Strategy / c4_FileStrategy mixup in Mk4tcl and MkWrap.

1999-11-05    Simplify c4_Storage

    Moved state out of c4_Storage and into c4_Persist, to prepare for
    multi-thread wrappers.  Made a new, simpler design to achieve that.

1999-11-04  MK 1.9d     Factor out stdio dependencies

    Moved all stdio dependencies out of core into new "fileio.cpp" src.
    The mk4.h header no longer includes <stdio.h>, added new "mk4io.h"
    header with a c4_FileStrategy class, derived from c4_Strategy.

    Also factored c4_Stream/c4_FileStream out of the c4_Strategy class.
    This alters the API slightly, but makes it 100% portable/embeddable.

    Added "tcl/iohan.tcl", a simple wrapper for generic storage access:
    locally, on a FTP server, in a local MK datafile, or using a remote
    Tequila server (see http://www.equi4.com/tclkit/tequila.html).

1999-11-03    Improved detach/restructure handling

    Changed detach to drop all persistent formathandlers, but leave the
    number of rows intact.  It will be much faster to destroy columns
    than to delete (and propagate) rows.  The result is that a pointer
    to a view of which the underlying storage object goes away will end
    up with the same number of rows as before, but no properties at all.

    Fixed a problem which would have occurred when properties are being
    "restructured away" and then committed.  The solution is to check
    for this and delete all such properties at commit time.

    Moved Buffer() out of c4_HandlerSeq and c4_CustomSeq, and made the
    new version in c4_Sequence allocate the c4_Bytes object lazily.
    Faster, and decreases sequence object size (for lots of subviews).

    Mk4tcl, mk::tqlayout now returns the proper tqlayout even if the views
    are empty.  Solved by adding extra arg to c4_Storage::Description,
    to return structure of a single top-level view.  Fixes "mk8basic.1",
    and the returned string now has one bogus list layer stripped off.

    MkWrap also adjusted with optional arg for storage.description().

1999-11-02    New Wrap code in MkWrap

    MkWrap, added new Wrap(seq,props,byPos) member, a c4_CustomViewer
    which wraps any Python sequence as MK view (for use in joins, etc).

1999-11-01  MK 1.9c     Mk4tcl changes

    Mk4tcl, several changes: added "-size" option to mk::get to return
    the size of prop value without fetching it (see new basic.9 test).
    Added "-globnc" for case-insensitive globbing (for regexp, this is
    available through the new (?i) metasytnax of Tcl 8.2 (see basic.10).

1999-10-31  MK 1.9b     Solved shared lib unload with Tcl 8.2

    Changed property symbol table setup to avoid static initializers, to
    work around a problem with shared library cleanup from Tcl 8.2.
    Added new c4_Property::CleanupInternalData call to clean up 100%
    (this code need not usually be called, only if memory is tracked).
    This fixes the crash-on-exit bug in Mk4tcl (Linux and Solaris).

1999-10-29    Make support for MkWrap and Mk4tcl

    Both MSVC and MWCW now also build Mk4tcl and MkWrap extensions.  The
    MWCW project compiles for Win and Mac *on* either Win or Mac.
    Makefile extended to build Mk4tcl and MkWrap, added dist target.
    Updated to latest Perceps 3.5 beta, started generating docs again.

1999-10-27  MK 1.9a     New build / directory structure

    Completely reworked the directory structure to simplify all builds.
    Created new "builds/" area for all intermediate and output files.
    New MSVC 6.0 project structure created in "win/msvc60/".

    Moved Mk4tcl to the "tcl/" top-level directory, and MkWrap to the
    "python/" dir.  The MkWrap code has been removed from the project.

    Removed c4_View::Match and the regular expression package, since it
    can just as efficiently be done with a wrapper around MK, now that
    string results no longer allocate a temporary copy.

    Started writing a Tcl-based test suite for use with Mk4tcl.

MAJOR CHANGE SINCE 1.8.6

    Merge handler.cpp and format.cpp classes to get rid of special-cased
    in-memory version of handlers.  This has major effects on how data
    is kept for unattached views (they still exist, but no longer
    special).  The code is leaner and meaner, it passes all regression
    tests.

ALSO LISTED IN THE RELEASE HISTORY

    Added c4_MemoRef::Access and c4_Memoref::Modify for partial access
    to memo fields.  Avoids copying and allows inserts/deletes anywhere.
    MkWrap and Mk4tcl have both been extended to allow using Memo fields
    for simulated file IO (mk::channel for Tcl, MkMemoIO.py for Python).

    Added c4_Reference::GetSize to determine the size of a value without
    accessing it.  For ints, returns negated bit width if 1/2/4 bits.

    Added experimental c4_View::RelocateRows to move rows from one view
    to another (both must be in same storage, with the same structure).
    Moves do not involve any data copying w.r.t. subviews and memo's.

The old release history is at http://www.equi4.com/metakit/history.html

# vim: tw=72