3.7b4:
  - ML Decomp fixes:
    - Arrays with a column count. 256
    - $'s on level 1 that are exactly 23 characters ( including the last " )
      what happens is there is just a white space where there should be an
      elipse
    - Tagged %'s were the combination of the lengths of the name and the %
      is > 22 characters.
    - Fixed problems ->STR mode and working on COL's

3.7b3:
  - Fixed list decomp found by running ->LIST on RANM output in istack
  - Added OFF key entries back since FASTOFF code is now conditional
  - Modified `mass' script to take flavour from command line
  - Added Makefile with a few handy targets
  - All flavours now included in betas

3.7b2:
  - Removed ill formed line in gentries.a -- reads OffK (Detlef)
  - Fixed long tag bug in ML Decomp (Will)
  - Improved conditional compilation support using GNU m4 (Tony)
  - Converted mass to sh script (Tony)
  - Added Tony as another co-author
  - Updated docs for 3.7

3.7b1:
  - Added FSTCALL support to avoid ROMPTR lookup (Tony)
  - Started work on external user key library

3.6a:
  - Added user functions to Cappuccino flavour to match 3.4a and docs
  - Temporarily removed FASTOFF as it was causing problems
  - Fixed MIN and MAX within symbolics in ML Decomp
  - Updated documentation slightly

3.6:
  - Release version

3.5b17:
  - Added MLa%>$ to fix decomping of reals within symbolics and units (Tony)
  - Optimized the NS keyhandler for the fast lib menu (Tony)

3.5b16:
  - ML Decomp updated to version 42e:

    29.11.1998      WKL
    - Fixed rounding errors
    - Fixed complex % handler when working in RRP's and lists
    - Added all other changes made by Tony.

3.5b15:
  - Fixed checking of sGX2 flag in GetCOLIndent (Tony)
  - Fixed display refresh flags for DA1 in interactive stack (Tony)
  - Fixed referencing of VAR menu replacement key handler (Tony)

3.5b14:
  - Added extension to source in RPLCOMP line for GNU-Tools 3.x
  - Added support for xNEGNEG in algebraic decomp (Tony)
  - Changed OVER#2+UNROL to OVER#2+UNROLL (Tony)
  - Uncommented and updated xEDIT$ML

3.5b13:
  - Added user accessible xlib xFASTOFF for custom fast off routines
  - Fixed inconsistent ROMPTR backup (Tony)
  - Fixed strength length problem with xWHERE in algebraics (Tony)

3.5b12:
  - Fixed precedence problem with logicals in algebraics (Tony)
  - Changed null-name/hidden char to minus sign `-' (Tony)

3.5b11:
  - Fixed tagged lists for Cappuccino/Frappuccino flavours (Kresimir Pervan)
  - Added DirLabel: support in menu code so FNT1 usage is more consistent (Tony)
  - Improved support for null-name/hidden dirs (Tony)

3.5b10:
  - Fixed VAR menu/UNDO key interaction (Tony)
  - Updated Will's address to wlaughlin@ntsi.com

3.5b9:
  - ML Decomp updated to version 42d:

    31.10.1998      WKL
    - Fixed the 20 character tag problem
    - Fixed the ->STRML tag problem
    - Fixed the ->LIST of array's in the ISTK (it was a decomp problem, not
      a Java problem)
    - Fixed % formatting ( hopefully for the last time )

3.5b8:
  - Lib menu replacement (Tony)

    Handles libid's < 256 and skips built-in rom libraries properly
    Fixed corrupt menu if no visible libs were present

  - Fixed PICTURE environment (Tony)

    A MLDC temp variable was overwriting a HP48 display var

  - Fixed ML Decomp bugs for arrays with > 2 dimensions (Tony)

3.5b7:
  - Added fast OFF key replacement (Jean-Yves)
  - Modified to call SetThisRow on exit instead of GetMenu% InitMenu%.
    Not only is this shorter, but is needed for with the new lib menu.

3.5b6:
  - Improved lib menu replacement (Tony)

    Now automatically updated when libraries are added/deleted
    Fixed problem with duplicate lib ids
    Fixed miscellaneous bugs

3.5b5:
  - Updated lib menu replacement to match Lib 1.2 (Fabrice)

3.5b4:
  - Added fast library menu replacement, currently G/GX only (Fabrice/Tony)

3.5b3:
  - Updated Richard's address and version number in docs
  - Fixed array decomp (Tony)
  - Fixed decomp of xPI (Tony)

3.5b2:
  - Fixed spawning of istack from built in editor (Tony)

3.5b1:
  - Fixed editing of null character; reset edit modes in case of error (Tony)

3.4a release:
  - Fixed FNT1 path display to be S/SX compatible

3.4 release:
  - Removed another redraw put in place for RainEQ
  - Updated docs etc...

3.3b10:
  - ML Decomp updated to version 42b:

    fixed some bugs related to edit mode

3.3b9:
  - ML Decomp updated to version 42a:

    from v41 to v42 05.28.1998      ber
    - Changed mode passing affected source files are:
       decomp.s
       init.a
       miscsubs.a  (Make$, added GetDecompWidth)
       and all object handeling files.

  - FNT1 now standard for path display, $>grob, and menu.  Original
    smallfont is still available as a compile time option.

  - Fixed operator grob shorter than operand grob(s) assumption (Tony)
    a.k.a 'inf+inf+inf' bug

  - Fixed corrupt path display of hidden dir in FNT1 version

  - ML Edit now standard when ML Decomp is present

  - Now calls ML Edit for built in editor as well (Tony)

  - Call xEdit instead of xEDIT when using ML Edit and StringWriter,
    but still fall back on ID EDIT

  - Removed DA3 redraw when exiting equation writer (was for old RainEQ)

3.3b8:
  - Fixed mask to clear cache when wordsize or base is changed - needed
    because there can be binaries within lists, which in the scroller
    and/or Cappuccino/Frapuccino are displayed with AGrob

  - Fixed redraw problem when entering scroller when DA2 was invalid,
    for example if review key was pressed

3.3b7:
  - ML Decomp updated to version 41a:

    from v41 to v41a        05.12.1998      ber
    - Added support of edit mode for alg's and units

    * V41 never released.

    from v40v to v41        01.01.1998      ber
    -Added entries to Gentries.a file
    -Added edit mode for all objects ( excluding alg's and units )
    -Fixed previous changes by Tony in this department
    -Fixed bugs in rrp handeling.

3.3b6:
  - ML Decomp updated to version 40v:

    from v40u to v40v       15.12.1997      ber
    -Incorportated fixed by Tony
    -Fixed vector decompilation in lists
    -Fixed patch.s to reflect updates in decomp.s

  - Made year in lib title year 2000 compliant

3.3b5:
  - Added external equation writer support (e.g. RainEQ)

  - Fixed cache bugs (Tony)

  - Fixed missing out of memory check in array decomp (Tony)

  - Updated FNT1 and ML Edit patches to match new original files

3.3b4:
  - Added entries for ML Edit to source

  - Fixed typo in AGrob user function optimization

  - ML Decomp fixes and improvements:

    During ML decomp in DO>STR mode, tagged objects created as
    ":tag: object". Otherwise, a single tagged object becomes "tag: object",
    which won't parse properly.

    In DO>STR mode, strings within lists or progs are checked for embedded
    " characters. If found the string is decomped as "C$ nnn string". Such
    objects can then be parsed properly.

    Fixes a decomp bug with xFCNAPPLY. Previously, if the argument was a
    direct bint (i.e. # 1 as opposed to ONE) the whole thing would decomp to
    'UNKNOWN'. To test this enter 'FUNC(A)', then ED it, change ONE to # 1.

  - Cleaned up menu code

  - Fixed status line redraw problem in FNT1 version and cleaned up source

  - Made patches for FNT1 and ML Edit source relative to this version

3.3b3:
  - Small optimizations to AGrob courtesy Balazs

  - Tweaked submitter's name in some of the CVS entries

  - Two alternate libraries provided, one to test out ML Edit routines
    the other to test out FNT1 routines

3.3b2:
  - ML Decomp updated to version 40ua:

    from v40t to v40u       23.11.1997      ber
    -fixed xwhere bug, and review mode

  - Another tiny optimization to istk

  - Another new beta tester:

    Robert Tiismus          <robert@physic.ut.ee>

3.3b1:
  - ML Decomp updated to version 40s:

    from v40r to v40s       3.11.1997               ber
    -Fixed Rounding errors in %'s

  - Some slight byte optimizations courtesy Balazs

  - Two new beta testers:

    Balazs Fischer          <Balazs.Fischer@studbox.uni-stuttgart.de>
    Tony Ambardar           <tonya@ee.ubc.ca>

3.2a release:
  - Fixed ON exit in scroller for Frappuccino/Cappuccino

3.2 release:
  - Added Frappuccino flavour

3.1b5:
  - ML Decomp updated to version 40r:

    from v40q to v40r       22.10.1997      ber
    -Fixed DcSymb to support user command supported in algebriacs.
    -Changed the ST flag structure for dcsymb.a
    -Fixed bug where: 'SQ(SIN(M)*ATAN(U))' wouldn't display correctly in SL mode.
    -Fixed Patch.bin to do type checking, and cleaned up error handeling.

3.1b4:
  - Altered location of hash table for ML Decomp patch routine.

    The included patch routine will only work with 3.1b4 or greater!
    That means for this beta you will have to download the whole lib.

  - ML Decomp updated to version 40q:

    from v40p to v40q       15.10.1997      ber
    -Fixed character decomp
    -Fixed BINT decomp
    -Fixed Patch.s to work correctly on cards on the SX.

    from v40o to v40p       15.10.1997      ber
    -Fixed the % rounding errors.
    -Fixed patch.s to work correctly.

  - From now on the latest patch.bin and decomp.bin will be available for
    beta testers from my web page for use between betas where only the
    decomp changes.

3.1b3:
  - ML Decomp updated to version 40o:

    from v40n to v40o       9.10.1997               ber
    -Fixed arrays again.
    -Fixed patch routine to update decomp xlib in java libs that support it
     ( currently 31.b2 and > )

  - Added `patch.bin' and `decomp.bin' to package.  You can use use this
    to update your decomp, so you don't have to download the whole Java lib
    if you don't want to.  See the enclosed README.PATCH file for details.

3.1b2:
  - ML Decomp updated to version 40n, fixes the following bugs:

    from v40ma to v40n      5.10.1997               ber
    -Fixed lists of units/algebraics
    -Fixed , in % decomp, user flag 56 toggles this option ( when 56
     set comma's are used in STD mode )
    -Fixed arrays with long % on level's > 1
    -Fixed ENG display error
    -Fixed tagged algebraics

  - Moved ML Decomp to last xlib in library for Will's patch routine

3.1b1:
  - Display loop and DISPN optimizations from Paquot
  - Version strings updated to 3.1

3.0 release:
  - ML Decomp updated to version 40ma, fixes last minute bug
  - Added ML Decomp README to source: mldecomp.doc and source docs/mldecomp.sgml
  - Updated all checksums etc... to match

3.0 pre-release:
  - ML Decomp updated to version 40m, fixes all known remaining bugs
  - Added faster KeyInBuffer? routine, courtesy Paquot
  - Slightly optimized display routines, also from Paquot
  - Updated package for release

3.0b14:
  - ML Decomp updated to version 40k, fixes some more bugs, but there are
    still two known bugs left:

    o Algebraic/VLM condition
    o Warmstart/weird display on some simple SRPL I have

  - Integrated EQStk 9.2's CRC, cache clearing code, and UNDO exit fix.  Now
    handles changes in system flags that affect display (e.g. SCI/FIX/ENG),
    as well as any CRC changes (e.g. arrays poked by MATRIX)
  - Merged two temporary environments into a single temporary environment
  - Fixed compile of Mocha and Decaf libs
  - Strings are now scrollable

3.0b13:
  - ML Decomp updated to version 40j, fixes all known bugs
  - Tagged objects are now scrollable if they are of correct type
  - Updated the docs; added Frequently Asked Questions for Java
  - TO DO: clear cache if certain flags are changed (planned for EQStk 9.1)
  - TO DO: truncate grobs before committing them to cache (maybe in EQStk 9.1)
  - TO DO: Check StringWriter in SysRPL mode

3.0b12:
  - Made 4 flavours available, like we discussed.  See the docs for more info.
  - Added flavour string to SOL screen
  - Included new SGML docs in the source tree and compiled versions in the
    main Java package
  - Integrated EQStk 9.0's cache.  All flavours with AGROB also contain the
    cache.
  - Modified STK to wipe cache on exit in case the user changes the size flag
  - Added CLRCACHE hidden xlib in case they don't use STK

3.0b11:
  - ML Decomp updated to version 40ha, fixed one reported bug
  - Removed xMLDC for now to avoid inconsistencies later (it is not 100%
    compatible with the built in decomp yet, so if we release it as is
    it will screw up people doing things like "MLDC" palparse later)
  - Minor update to AGrob wrapper to be consistent with EQStk 9.0.  I have
    yet to implement the ML cache, though.
  - `nokeyob.lib' no longer included - I think I'll make it the default
    unless anyone still has problems with it?

3.0b10:
  - ML Decomp updated to version 40h, fixes all reported bugs and adds
    support for ENG mode.  Please continue testing and let us know!

3.0b9:
  - AGrob updated to EQStk 8.0

  - Fixed review key replacement to run built in one if it is non-standard,
    replacement one otherwise (e.g. plot menu, solver, etc...)

  - Fixed long standing quirk: If on-stack algebraics were disabled and you
    still asked to scroll level one, levels > 1 would not be redrawn properly
    (not serious, but it has been bugging me for AGES)

  - Major overhaul to ML decomp, courtesy Will Laughlin.  NEEDS EXTENSIVE
    TESTING SO WE CAN FINALLY RELEASE THIS BABY!

  - Added Will as another co-author for all the work he's done

  - Review key now less of a hack, uses special mode in new decomp

  - Added xMLDC to user menu, also for new decomp.  Give it any type of
    object, and it will act like x->STR (aka DO>STR)

  - Main display loop, istk modified to accomodate new decomp and new AGrob

3.0b8:
  - Now that I'm back in classes again, time to procrastinate and work on Java!

  - Added review key replacement.  Uses ML decomp, built-in decomp, or SRPL
    decomp as appropriate.  Uses UFL FNT2 for display.  Especially nice in
    the VAR menu using the ML decomp, it's much faster :^)

    This isn't quite complete... I need a routine that return me a flag T/F
    on whether the current menu has a non-standard review key (for example,
    the plot menu).  If so, I want to run the built in one rather than this
    replacement one.  One might call it DoReviewKey@ (along the lines of
    MenuKeyNS@ etc...).  There doesn't appear anything to do this ROM, so
    can anyone help me here?

    As an added bonus if flag -59 is set, the review key will only show
    variable names when in the VAR menu... this was suggested by Manoj
    based on what the built-in G/GX file browser does.  Also, like most
    other things in Java, key strokes pre-empt display of the review key, so
    if it's only drawn half the screen and you press a key, it will exit and
    service the key instead.

  - Fixed bug in STK and FHOOK that left an External on the stack.

  - STK and JAVA commands are now smarter on when Java is actually running
    and when it isn't... it's harder to confuse them by manually changing
    flag 64 now.  For example, previously if Java wasn't running and you did
    did 64 SF JAVA, the replacement stack wouldn't start because it'd think it
    was quitting and clear flag 64 instead.  Now it actually starts.

    Also STK will clear flag 64 if you start it with flag 64 set and Java
    wasn't actually running, so that you can check "Java Active" in order to
    actually start it.. well just try it and you'll see!

  - Eliminated some overhead in SRPL mode.  Now if you want don't use STK to
    enable SRPL mode (i.e. just 60 SF) you also have to run FHOOK if you want
    anything to happen.  Also, if you exit SRPL mode and you have your tables
    in TEMPOB (i.e. due to a covered port) you should get your 30K or so freed
    up iff you use STK and/or FHOOK.

  - Reverse ordered this CHANGES file

3.0b7:
  - Detlef fixed SRPL mode so it doesn't make a copy of the Jazz tables
    if they are covered and flag 60 is not set.  Still requires a copy
    if covered and you set flag 60, however.  Of course if you put the
    tables in a "normal" port you won't have any of these problems.

  - Added the "nodecomp" lib back into the betas after I found out testers
    weren't using the latest lib because of its instability.  At least if
    you use the nodecomp lib you will be testing the non-decomp specific
    changes.

  - Converted Java source to CVS, which is a concurrent version control
    system that wraps around RCS.  The archive files are still 100% backwards
    compatible with RCS, but CVS allows me to split archive and working
    directories and has more powerful commands.  If you are interested in
    RCS and/or CVS let me know!

3.0b6:
  - Added support for LS-view key on GX (same as down arrow when edit line
    doesn't exist)

  - Fixed SX visit key so it doesn't crash when an edit line exists

  - Merged SX and GX edit key replacements to save a few bytes

  - Redid VAR menu again, this time with an NS handler instead of a track
    replacement.  There were still some cases where the old one wouldn't
    redraw when it should have.

3.0b5:
  - Slightly newer GetKeyOb from Detlef, this one will force a status area
    update after ANY key is pressed when you change the R/W status on a card.
    This allows the port annunciator to get updated faster when you switch the
    R/W status.  Also it will give an audible "chirp" to acknowledge a R/W
    switch.

  - Support for Arnold's MATRIX and NEWMAT commands for built-in to editor
    and matrix keys.  Will run them automatically if they exist instead the
    built-in matrix writer.

3.0b4:
  - New GetKeyOb, courtesy Detlef Mueller!  Yes believe it!  Commented and
    should fix the problems that have been associated with the GetKeyOb in
    the past.  Please test it, especially if you were previously forced to
    use the nokeyob version of Java.  I still include it just in case though..

  - 1D lists now horizontal instead of vertical in AGrob

  - Add another set of brackets added to 2D symbolic matrices in AGrob, like
    previously done for conventional matrices

  - Added page preservation code to UPDIR (only applies in VAR menu unlike
    some others); still includes smarter redraw setting

  - Fixed VAR and LIB menu redraws

3.0b3:
  - Added improved VAR and LIB menus.  They are smarter about when to redraw
    the stack and when not to redraw the stack.

  - Converted entire Java source to Revision Control System (RCS).  History
    goes back to the 2.9 release; decomp history goes all the way back to 30a.

3.0b2:
  - Updated STK for same bug as fixed in 2.9a

3.0b1:
  - ML Decomp back for revenge
  - Fixed annoying lib ellipses problem... was a result of my changing
    "Library" to "Lib" and not changing the truncation code to compensate

2.9 release:
  - Added UFL and FNT1 check to xAGROB (forgot)
  - Removed ML Decomp temporarily for release

2.9b12:
  - Improved low mem handling in scroller some more (fixes problem of
    brief redraw by built in stack)
  - New decomp from Will (does NOT include any algebraic fixes yet though)
 
2.9b11:
  - Removed palparse from UFL? for config (was hanging on warmstart - oops!)
  - I didn't like slowdown I introduced a while ago in SysRPL mode, so I changed
    it again.  Should be faster again; you should only get a slowdown if you
    put your tables lib in a covered port.

2.9b10:
  - Altered for new UFL LID #101

2.9b9:
  - Changes for the Universal Font Library (UFL).  Requires a UFL containing
    FNT2 (currently only Font Levels 2 or 3)
  - Removed ->FNT, FNT->, cvFNT as they will now be part of the UFL's font
    maintenance package
  - Added code for checking existence of UFL, and making sure it isn't purged
    while Java is running

2.9b8:
  - Yes this beta is actually slightly smaller and faster, believe it or not!
  - Lots more internal code cleanups, source doesn't really look the same
    anymore - more like Jazz
  - Redid AGROB low mem handling
  - Scroller should avoid booting you out of Java, but if there isn't
    enough mem it displays the object in 4 line mode, then blinks back
    to Java... weird!  At least it doesn't quit you now.

2.9b7:
  - Fixed ticking clock in istack
  - Renamed "ED" to more appropriate "ED/TED" in STK (for new editor stuff)
  - Re-organized code into separate directories (more cleanups to come)
  - Added SOL's cvfnt, faster FNT-> and ->FNT

2.9b6:
  - AGrob updated to EQStk 6.1
  - Minor fixes to 7-line istk (still screws up with clock on though)

  - Rather than saving the Jazz tables directly, Java now saves GetRplTab
    and GetDisTab and calls them to get the tables when it needs to do
    a SRPL decomp.  This means a slight performance decrease in SRPL
    mode, but it should solve the problem of losing 30K or more when storing
    the tables in a covered port.

    Actually, you still need enough RAM for a copy of the tables, but
    only if you want to use SRPL mode.  Anyone have any better ideas?
    I can't do the same as Mika does in SSTK because SRPL mode can be
    enabled/disabled at any time by the user setting flag 60...

  - Multiple libraries included in order to track down specific bugs to
    either the GetKeyOb routine, the ML Decomp, or a combination.

    java29b6.lib	: Normal lib with latest everything
    nodecomp.lib	: No ML decomp, but contains GetKeyOb routine
    neither.lib		: Neither the ML decomp or the GetKeyOb routines
    olddecomp.lib	: Old ML decomp (pre algebraics), contains GetKeyOb

    If some other combination is necessary let me know so we can fix these
    these things.

2.9b5:
  - Interactive stack now 7 lines
  - Altered how startup/exit works:

    o now preserves user mode on exit
    o only sets user mode and var menu on warmstart

    Should be a bit faster starting/quitting Java now.

2.9b4:
  - Newer decomp from Will, fixes some bugs.

2.9b3:
  - Newer decomp from Will, fixes some bugs.

  - Updated editing for new ED/TED.  Requires JAZZ 6.0 or later.

2.9b2:
  - New decomp that supports algebraics and units directly

  - Interactive stack slightly improved from SOL source

2.9b1:
  - Same as released 2.8, except includes the ML decomp.

2.8b2:
  - Changed FHOOK to a tNAME'ed xlib

  - Improved docs and dirs a bit to prepare for release

2.8b1:
  - Fixed JazzHook retry in STK.  SysRPL mode NOW REQUIRES JAZZ 5.9 OR
    HIGHER.  USING VERSIONS 5.5 TO 5.8 WILL PROBABLY CAUSE A CRASH.

  - If using ED in SRPL mode, strings on level 2 can take a real offset on
    lev 1.  This is most typically used after xASS gives an error.  You
    can now just press down-arrow to jump directly to the problem.

  - Fixed pre-mature exit from main display loop.  There might have been
    problems updating the screen with the faster UPDIR and HOME key
    replacements if you were typing real fast.

  - Added Jack's font to jvfonts.zip

2.7b5:
  - Modified JazzHook to retry if the user runs STK

2.7b4:
  - AGrob updated to EQStk 5.8
  - Small changes to smallfont, now supports tab, newline, FF, CR.
  - Now automagically calls DIS/ASS for editing when in SysRPL mode

2.7b3:
  - $>grob2 now returns 0x0 grob instead of 0x6 grob for NULL$ (yay!)
  - AGrob parenthesization fixed when left argument in multiplication is
    a unit object

2.7b2:
  - Newer decomp (fixes lists)
  - Added $>GROB2

2.7b1:
  - Support for new JAZZHOOK, like having a better SSTK! (Me)
  - $>grob2 fixed for very long strings (Rich)

2.6b4:
  - Newer decomp
  - Smaller exit from STK

2.6b3:
  - Newer decomp
  - AGrob updated to EQStk 5.6
  - More edit line fixes
  - Rev. J Equation Writer bug fixed in key replacement

2.6b2:
  - Newer decomp

2.6b1:
  - AGrob updated to EQStk 5.5

  - Error trapping for AGrob in xAGROB, scroller, and main display loop.

  - Replacement GetKeyOb routine now SX compatible, allows on-the-fly port
    lock/unlock, courtesy Hervi Couplet.  Apparently still has problems with
    32K cards, but is untested.

  - ML Decomp mostly debugged and in place, courtesy Will Laughlin

  - Added tab character to medium standard font

  - Added array size cap of 20 for on-stack arrays being passed to AGrob

  - Keyboard Buffer is now checked AFTER EVERY LEVEL on stack, so if you type
    something while the stack is being updated, the main display loop will
    be exited and the key will be handled.  The stack will be updated later
    once you stop typing.  This improves keyboard response a bit.

    Previously, KEYINBUFFER? was only checked between entire stack redraws
    (all five levels)

  - Miscellaneous stuff I ran into

