Development history of FDXMS:

The version number this time are names of fruits. Fruits taste great
and are nutritious.


Version 0.94.Bananas - 2005-05-28 - Martin Str”mberg

- After chaining INT 0x15 AH=0x87, execute a correct return.
- Functions 0xf and 0x8f (realloc) broken because BX was destroyed
    before used.
- Don't touch BL on a successful calls unless the specification says
    it should be zeroed (bug 1894).
- Thank you to Peter Nagy for testing.


Version 0.93.Strawberries - 2005-05-05 - Martin Str”mberg

- Changed assmebler to GNU as.
- Device driver calls don't mess up registers anymore.
- Added proper handling of input and output status device commands
    according to information from Bob Smith.
- Unecessary pushf/popfs removed from int2f_handler.
- Modify int15_handler to not clobber return flags when function 0x87
    is forwarded to previous int15 handler.
- Rearranged order of the fields in the handle struct for smaller
    resident memory size.
- Merged the variables hma_used and a20state into flags.
- Resident VDISK detection added back. According to the specification
    it isn't sufficient to just check for VDISK at installation
    time. (Note that I mistakenly removed the resident checking, not
    Lixing Yuan.)
- Only check for VDISK until HMA has been allocated.
- Proper stack frame setup made huge space savings possible.
- Improved XXMS valid number check for smaller size. (It should be the
    most optimal possible now, sizewise.)
- XMStest has been removed as it requires a Borland 16-bit C
    compiler. (If you really need the program extract it from an
    earlier FDXMS release.)


Version 0.92.Oranges - 2003-07-10 - Martin Str”mberg

- VDISK detection in check_vdisk corrected by Lixing Yuan.
- Only call check_vdisk at installation time by Lixing Yuan.
   (Evidently those two corrections to VDISK handling make SMARTDRV
   work with FDXMS.) 


Version 0.91.Apples - 2003-02-08 - Martin Str”mberg

- FDXMS now only warns if it can't disable the A20 line instead of
	refusing to load (bug 1352).
- INT15 parameter added.


What's up with the version numbers this time? This time they are a
tribute to SF writer Samuel R. Delany, <http://www.pcc.com/~jay/delany/>.
He has written a book called Stars in My Pocket like Grains of Sand
which is an amazing piece of work. But if you want to try his works, I
suggest you try some of his lighter works (sentences spanning more
than a page can be tiresome) like Babel-17 or the Neveryona series.


Version 0.9.Sand - 2002-03-02 - Martin Str”mberg

- Squeezed out some more bytes by rearranging xms_free_xms.
- Squeezed out some more bytes in trace version by substituting test
	with bt in xms_dispatcher.
- Squeezed out some more bytes in validation of AH in xms_dispatcher.
- Corrected not recognising the BIOS switch in the non-TRACE versions
	(bug 880).
- Corrected 386 detection by Ralf Quint and me.
- Maximum XMS memory is 0xfcb0 KiB, not 0xffff KiB. Changed code to
	reflect this. (Note that maximum XXMS memory is much larger.)
- Use smsw instead of mov cr0 to detect protected mode.
- Now fdxxms.sys has been tested with 512MiB.
- Added PS parameter that modifies the A20 line switching code so
	FDXMS works on PS/2 Model 30 (and perhaps other PS models)
	(bug 753).
- Added GEODE parameter that modifies the delay routine so FDXMS works
	on National Semiconductor's Geode GX1 200MiHz processors (bug
	886). 
- Added DELAY parameter and using it in enable_a20 and disable_a20
	(bug 1038).
- Changed a lot of "or reg1, reg1" to "test reg1, reg1".
- Removed BIOS_FLAG from the flags variable and added insertion of
	near jmp to do_bios_move if the BIOS parameter was given.


Version 0.8.Grains - 2001-07-29 - Martin Str”mberg

- xms_move_xms: add illegal ovelap check in BIOS path.
- Added a flags byte. 
- Made the bios variable a bit in the flags variable.
- Made the initialized variable a bit in the flags variable.
- Really only initialise once by looking at said bit.
- Corrected my erroneous correction of xms_query_free_xms (RBIL is a
	little confusing here).
- Minor adjustments to XMStest.c.
- Corrected xms_query_free_xms which didn't return BL == 0xa if all
	XMS memory is allocated.
- Corrected xms_get_move_addr which could be fooled by large offsets
	or sizes while verifying length was valid.
- Proper copyrights added to cptest.c
- Optimised away the redundant variable xms_size.
- Added printout of how much XMS memory that was found.
- Added an XXMS version (fdxxms.sys) supporting 4GB (4GiB untested,
	but up to 384MiB tested).
- Decreased size by using a register instead of immediate in
	xms_free_xms and xms_dispatcher.
- Made XMStest use a better seed for more randomness.
- Watch out for AX=0xe801 in INT15 handler.
- Properly clear the carry flag in INT15 handler on success.


Version 0.7.Pocket - 2001-07-10 - Martin Str”mberg

- Correct lookup of variable 'bios'. I inadvertently happened to use
	the bios without the BIOS argument (cs: override was missing).
- Added a call to INT15 AX=0xe801 to detect more memory.
- Merged lock and used bytes, so bookkeeping size is now down to five. 
- Simplified and corrected xms_query_free_xms.
- Move some of the initialisation of the handle table to initialize in
	preparation of adding the call to INT15 AX=0xe820.
- Increased minimum number of handles to two due to the above change.
- XMS test program by Tom Ehlert. I include the .EXE file for test
	purposes unitl FDXMS is ready for prime time. Minor
	improvements by me.
- xms_move_xms without BIOS argument didn't work right for lengths not
	evenly divisible by four (i. e. 2, 6, 10, ...) if source
	address was below destination address.  


Version 0.6.Stars - 2001-07-05 - Martin Str”mberg

- Correction of merging code by Tom Ehlert.
- Keep lock counter indicating lock in case of lock overflow by Tom
	Ehlert. Hopefully corrected now. I messed up Tom's earlier
	correction.
- Restore a20 state in xms_move_xms in case of failure.
- Don't uneccessarily enable a20 if it's already is enabled in
	xms_move_xms.
- Improved address moving verification which could be tricked with big
	offsets in conventional ram by Tom Ehlert.
- Now possible to allocate 0 sized EMB if there are handles available
	(but no memory).
- Minor updates to invalid NUMHANDLES option argument strings.
- Now you can install FDXMS.SYS in a directory called FDXMS.SYS.
- Handle ovelapping moves correctly. I'm not sure if BIOS does handle
	overlapping moves correctly though. Feedback wanted!
- Skip move altogether if move length is 0 by Tom Ehlert.
- Skip move altogether if src == dest by Tom Ehlert.


What's up with the version numbers? They are a tribute to Dead Can
Dance <http://www.evo.org/html/group/deadcandance.html>, a superb
music group. Their third album is called "Within the Realm of a Dying
Sun" which was the first piece I heard by them. Check it out some
time. 


Version 0.5.Sun - 2001-06-13 - Martin Str”mberg

- The TRACE_SEPERATOR argument was never implemented. Ooops! Don't go
	looking for it.
- Coalesce xms_realloc_umb, xms_reqest_umb and xms_release_umb into
	one block to save memory by Tom Ehlert.
- Keep lock counter indicating lock in case of lock overflow by Tom
	Ehlert.
- Now possible to compile in the trace code conditionally.
- Detection of end of command line in DOZE corrected.
- Inserted P8086 and P386 at strategic places so that it ought to fail
	gracefully on a processor earlier than 386.
- Check the size of moves to/from XMS. Idea from Tom Ehlert.
- Report interface version as 2.0, which it is as XXMS is not
	implemented, from Tom Ehlert.
- Report driver version properly from Tom Ehlert.
- Protect bios_gdt from reentrancy properly by Tom Ehlert.
- Only copy 0x1000 words using BIOS call without enabling interrupts
	by Tom Ehlert.
- Debug option IFDEFed out for now.
- Check the size of moves to/from conventional RAM.


Version 0.4.Dying - 2001-05-13 - Martin Str”mberg

- NOTE: The license has been changed to GPL version 2

- Using ax for jump to the proper function; thus no need to pop bx in
	all functions anymore.
- Localized most local labels.			
- Command line arguments handling added.
- TRACE argument added.
- DEBUG argument added. Doesn't do anything though.
- NUMHANDLES argument added.
- Improved merging of adjacent free blocks by Tom Ehlert.
- Made to work in protected mode environment (e. g. SICE) by calling
	the BIOS by Tom Ehlert.
- Removed limit of 16MiB in protected mode environment.
- BIOS argument added to always use the BIOS.
- Should be able to handle more than 32MiB now by Tom Ehlert.


Version 0.3.Realm - 2001-04-21 - Martin Str”mberg

- Validation of handle procedure added: handle_valid.
- Use of handle_valid added to functions 0xa, 0xb, 0xc, 0xd and 0xf
- Don't push _all_ registers in xms_move_xms.
- Change XMS_NO_UMB_AVAILABLE to XMS_NOT_IMPLEMTED in xms_request_umb
	by Bart Oldeman.


Version 0.2.Within - 2001-04-15 - Martin Str”mberg

- Corrected function 0xe.


Version 0.11 - 2001-01-?? - Louis P. Santillan

- Correction of A20 switch code.


Version 0.1 - 10/11.95

- initial version, uploaded for testing
- still about 400 bytes bigger than HIMEM :(
- poor feedback until now...
- works with a lot of software, but Watcom Help doesn't work when run with
  Norton Cache of Norton Utilities 8.0