history.txt

GNU FreeDOS兼容MS DOS很好的东东.

2006 May 20 - Build 2036 cvs and 2037 test
-------- Eric Auer ([e-mail])

WARNING: This changelog entry is a draft! The cvs log contains
  data for both HEAD and UNSTABLE branch while a cvs checkout
  seems to return only the HEAD branch. This file history.txt
  should describe "2036 cvs" HEAD and "2037 test" UNSTABLE
  separately rather than in one merged changelog item.

In addition, the "Changes Eric" items yet have to be ported to
  the UNSTABLE branch and the version.h for UNSTABLE should say
  "03 Mar 2006" "2037 test" (revision 37)...!

Most recent changes to source code (see version.h) are:
  HEAD 9 Sep 2005 (not counting the Korean stub and execrh change
    by Eric, and not counting the version number change itself)
  UNSTABLE 3 Mar 2006 (includes SYS changes, more Win3 support,
    COM port output for debug print, the SHARE tool...)
    The head to unstable diff in general is quite large. As a rule
    of thumb, use UNSTABLE if you need NLSFUNC or Win3 386 mode or if
    you need a very new kernel, and use HEAD if you need stability.

+ Changes Jeremy
	* build.bat, config.b: UPX and WIN3 build options, for BUILDALL
	* boot/oemboot.asm: set backup boot sector on FAT32, add update
	  option, fix oemboot to work with msdos 7+, in UNSTABLE branch.
	* docs/bugs.txt: update Bugzilla URL
	* docs/config.txt: document SWITCHES, based on suggestions from Eric
	* docs/contrib.txt, docs/fdkernel.lsm: new email address
	* docs/sys.htm, docs/sys.txt: see oemboot.asm
	* docs/sys.txt: better check for valid backup boot sector # and add
	  option to skip setting it
	* hdr/debug.h, kernel/prf.c: slight cleanup, send debug prints to COM
	  port, allow printf to work correctly when SS!=DS, debug changes
	* hdr/device.h: add comments about what flags mean
	* hdr/portab.h: from dev: add MK_PTR and MK_SEG_PTR macros
	  add/use halt() wrapper for issuing "hlt" asm instruction
	* hdr/version.h: update version to 2035b-cvs (= not a release yet)
	* kernel/config.c: fix read calls to detect error, from Eric / Brad
	* kernel/console.asm, newstuff.c, prf.c, task.c: compile time option
	  to send debug output to COM port ("serial debugging console")
	* kernel/dosfns.c: do not compares filenames to name field in block
	  devices, extra win3 support, misc debug output changes, IsDevice
	  should not compare filenames against block device driver headers
	* kernel/dsk.c: works towards better SS!=DS support, centralized
	  debug macros, some comments
	* kernel/entry.asm: more win3 support, debug output changes, comments
	* kernel/execrh.asm: protect 386 registers, using hdr/stacks.inc
	  macros (Watcom: FS, GS Others: high 16 bits of EAX EDX, for MSCL8
	  also ECX, for BC5 also EBX), for ALL device driver calls. Without
	  this, some versions of free-EMM386 crash 386 optimized kernels.
	  See list archives 24 Jul 2005 and later.
	* kernel/fatfs.c: don't leave fd open when failing DIR bit check
	  bug 1874 fix, do not consider setting DIR bit an error for
	  int 21/4301=set attribute, if really is a directory
	  Added some disabled debug messages, improved win3 support.
	* kernel/init-mod.h: allow printf to work correctly when SS!=DS
	* kernel/initdisk.h: insure bpb_huge is initialized (zeroed) [merge
	  from dev kernel], fake floppy ddt so hd access works correctly on
	  floppyless systems, based on suggested fix from Eric
	* kernel/int2f.asm kernel.asm main.c: initial BIOS disk int 13h wrapper,
	  only passively detects and notes floppy disk change, more win3 support
	* kernel/inthndlr.c: potential fix for spurious near fnodes!=0 msg with
	  shsufdrv and >2 near fnodes in use for bug 1879, use halt() wrapper,
	  more win3 support, some debug output changes.
	* kernel/intwrap.asm: assembly glue to wrap misc interrupts with C
	  handler, cleaner push immediate from Arkady and note big bug to be
	  fixed, later moved to UNSTABLE branch (also affects makefile)
	* kernel/proto.h: fix prototype, add FAR
	  allow printf to work correctly when SS!=DS
	* mkfiles/generic.mak: undo a local only change, build list option
	  is proper way to do this. more win3 support, upx build option.
	* mkfiles/watcom.mak: allow printf to work correctly when SS!=DS
	* share/MAKEFILE, share/share.c, share/share.hlp: FreeDOS SHARE
	  (initial revision, later moved to UNSTABLE branch)
	* sys/fdkrncfg.c: sync printf prototypes (uncorrupt executables)
	* sys/sys.c: sync printf prototypes (uncorrupt executables)
	  better check for valid backup boot sector # and add option to skip
	  setting it. set backup boot sector on FAT32, add update option,
	  fix oemboot to work with msdos 7+
+ Changes Eric
	* kernel/globals.h changed the os_release version message to be
          "build 2036 cvs [version Sep 09 2005 compiled May 20 2006]" as
	  opposed to the old
	  "version 1.1.35 (Build 2035b-cvs) [Jul 23 2005 12:34:56]"
	  -> removed the "1.1.35" and compile TIME, added SOURCE DATE
	* hdr/version.h removed some no longer used version defines
	  Changed build to "2036 cvs" (revision 36)
	  WARNING: Version should be "2037 test" (revision 37) for UNSTABLE!
        * docs/history.txt: summarized log -d '>Feb 25 2005' kernel for this
	  WARNING: some of the described changes might be available only
	  in the UNSTABLE cvs branch (like SYS OEMBOOT, Win3, COM port
	  output for debug print, the SHARE tool, ...)
	* kernel/execrh.asm: removed the "protect 386 registers" patch
	  again, a fixed version of free-EMM386 is available. This reduces
	  device driver call overhead in terms of CPU and stack usage.
	  Older EMM386 versions destroyed eax ebx ecx edi esi ebp fs gs,
	  now go_driver_entry in emm386.asm saves those registers. Only the
	  INIT was broken, but the execrh.asm patch protected ALL functions.
	* share/share.c: changes to reduce TSR memory footprint
	* kernel/inthndlr.c: added stub for function 0x6301/0x6302 Korean
	  Hangul keyboard input methods (0x63xx is DBCS support related)
	* docs/intfns.txt: updated list of supported DOS functions. When
	  DOSLFN is loaded for int 21.71, only remaining unsupported int
	  21 functions are 21.4b05 (set execution state), int 21.63xx
	  (DBCS / Korean Hangul, see above) and int 21.5d01 - 21.5d05
	  (close files by name/computer/task, list open files, commit all)
	* made linebreaks in all batch files use CR LF, as Unix line breaks
	  confuse command.com, leading to "early abort" or "endless loop"
+ Changes Lucho
	* kernel/chario.c: Fix Ctrl-Z bug and retro-optimise a bit
	* kernel/config.c, sys/sys.c: Fix BC 3.1 CONFIG.C compile and
	  SYS Abnormal Program Termination error for BC
	* sys/sys.c: fix update option (compared too many characters)
+ Changes Eduardo
	* kernel/config.c: fix Spanish (34, not duplicate of 33)
	* kernel/inthndlr.c: add int 2f subfunc 122B and 122D
+ Changes Jason
	* kernel/fatfs.c: extended DPBs and Set Extended Error function
	* kernel/inthndlr.c, kernel/proto.h: same
+ Changes Arkady
	* kernel/inthndlr.c: use direct access instead of reading from copy
	  made on stack
	* kernel/intwrap.asm, kernel/kernel.asm: cleaner push immediate from
	  Arkady and note big bug to be fixed
+ Changes Michael
	* share/share.c: bugfixes
+ Changes Japheth
	* share/share.c: HXDOS related improvements
+ Changes Bart
	* kernel/main.c: patch to make device=f:\foo.sys work. String printf
	  optimisation


2005 Feb 25 - Build 2035a
-------- Jeremy Davis ([e-mail])
+ Changes Bart
        * int2f.asm: call_nls: DI register was not preserved (bug 1812)
+ Changes Arkady
	* floppy.asm: fix epilog for FL_SETMEDIATYPE & FL_SETDISKTYPE
	  to use 'ret' instead of 'ret 8'; improve comments
	* entry.asm: use unsigned comparision instead of signed
+ Changes Lucho
	* dsk.c: text in play_dj
	* fatfs.c:
          - zero creation/access stamp on directory entry write as MS-DOS 7.10
          - prevent removal or renaming of the current directory of that drive
	* ioctl.c:
	  - r_si/r_di contents added as documented in PC-DOS Technical Update
	* inthndlr.c
	  - fix: for INT21/5F07 and 5F08, before changing bit CDSPHYSDRV,
	    MS-DOS checks if physical device associated with drive letter. Without
	    this check MS-FORMAT under FreeDOS was destroys RAMDISK.
	  - zero serial number of Int 21h/30h (else buggy 32RTM thrashed stack) [& Bart]
	  - set AX = ES after Int 21h/4Ah (undocumented but expected by BRUN45)
	    (the last 2 changes needed to fix bugs discovered by Michael Devore)
	  - added Int 2Fh/2Fh processing to set DOS version as per MS-DOS 4.0
+ Changes Eduardo
	* inthndlr.c: added Int 2Fh/26-29h processing for NLSFUNC (with Lucho)

2004 May 30 - Build 2035
-------- Bart Oldeman ([e-mail])
+ Changes Lucho + Tom
	* config.c, fatfs.c, globals.h, init-mod.h: From Tom: must
	  set r_bpfat for C_BLDBPB, otherwise USBASPI.SYS, DI1000DD.SYS
	  won't like us. From Lucho: use DiskTransferBuffer for deblock_buf.
+ Changes Lucho
	* docs/fdkernel.lsm, stacks.inc, version.h,
	  entry.asm, inthndlr.c, io.asm, nlssupt.asm: no need to preserve
	  high part of ecx for Borland, ebx for MSVC
	* device.h, dsk.c, init-mod.h, ioctl.c:  MSC fixes and small cleanup
	* ioctl.c: Cache attr and control flow optimizations (~120 bytes)
	* device.h, dsk.c, ioctl.c: deal with the correct
	  structure for generic and query ioctls. Fixes issues with
	  3rd party device drivers (or ones that use dsk.c)
	* dosfns.c, inthndlr.c, proto.h: Merge dosmkdir and dosrmdir
	* config.c: strcasecmp and set_strategy common code optimizations.
	* main.c: Add necessary cast for BC
	* portab.h: Avoid problem with BC declaring __emit__, en/disable twice.
	* dsk.c: DJ mechanism without memcpy.
	* newstuff.c: replace loops by fmemset/fmemcpy (43 bytes)
	* dosfns.c, fatdir.c, fcbfns.c, ioctl.c: Some small cleanups.
	* config.c: GetNumArg and misc other cleanups
	* time.h, fatfs.c: converted TM_ENCODE into a function.
	* inthndlr.c, proto.h, fatfs.c: Inlined dos_setdta.
	  Various bug fixes and cleanups from Lucho and Arkady.
	* inithma.c: moved fmemcmp's into a checkHMA function.
	* inithma.c: HMAFree == 0xfff0 - bytesToAllocate
	  is valid and may happen if the buffers occupy exactly all HMA
	  space up to and including the last byte at ffff:ffff. Failing
	  caused all buffers to go low ...
	* inthndlr.c: cleanups (merge common code into long_check and
	  short_check). Made int21/3301,2e,54 consistent with MSDOS
	  behaviour.
	* globals.h, task.c, main.c: Convert setvec into a function
	  for the resident code
+ Changes Arkady
	* initdisk.c: initdisk optimizations (65 bytes) and is_suspect fix.
	* config.c, init-mod.h, intr.asm: optimized allocmem()
	* fattab.c, kernel.asm: optimizations
	* main.c: Make signon one big printf function call.
	* ioctl.c: bugfix for INT21/440b (change NetRetry if DX is _non_zero).
	* inthndlr.c: cntry == 0 doesn't make sense for Set Country
	  Information. Fix int21/5e02+
	* sys/talloc.c: Turbo C really wants unsigned __brklvl. Only shows
	  up if you enable stack checking.
+ Changes Bart
	* memmgr.c, proto.h, inthndlr.c: DosUmbLink no longer
	  tries to merge free blocks. Also cleaned up a bit.
	* inthndlr.c: Fix int21/3302 (the swap was wrong, Lucho & Bart)
	* config.c: Pointer-ise specificCountriesSupported accesses.
	* config.c: Reorganized GetBiosKey a bit -- inlining
	  GetBiosTime saves a slight amount of code, converted timeout < 0
	  to a blocking read instead of polling.
	* config.c: Use strcaseequal instead of strcasecmp since we
	  never care about order anyway.
	* chario.c: Fix problem with redirected input and
	  int21/ah=0a (debug < foo.txt) -- could get you an invalid opcode!
	* console.asm, main.c, prf.c: Use int29 for kernel console
	  output. Enables ansi escape sequences in config.sys once you load
	  nansi.sys. It also saves a couple of bytes.
	* newstuff.c, proto.h: made dosmktmp to return a long:
	  consistent with DosOpen et al and avoids the cast.
	* mkfiles/tc2.mak: Let TC2 warn for the utilities.
	* config.c: Disabled EBDA moves by default. Automatic EBDA
	  moves now happen using switches=/e:-1
	* config.c: Pointer-ize MenuStruct[MenuSelected]... (from
	  Lucho). and InstallCommands[i].  (from Bart). Some other small
	  optimizations.
	* memmgr.c: Some small optimizations (46 bytes) and dead
	  code removal from Lucho and Bart
	* init-mod.h: Need to define strchr to init_strchr otherwise
	  the wrong one may be called.
	* initdisk.c: Do a bit of printf merging.
	* filelist, drivers/makefile, init-mod.h, inthndlr.c,
	  task.c, globals.h, main.c: Convert getvec into a C function.
	* int2f.asm: Clean up int2f assembly a bit (merge some
	  common bits, clarify label names)
	* fatfs.c, fattab.c: fattab.c now remembers if the old FAT
	  entry was free or not, so the free space adjustment could be
	  moved back to fattab.c.
	* init-mod.h: Replaced setvec macro by a function.
	* entry.asm, inthndlr.c: Made int21/ah=25,35 reentrant.
	  Solves problem with Intel PRO/1000 driver.
	* portab.h, globals.h, main.c: Initialise all int
	  vectors except the ones at 70:xxxx using a table.  Convert setvec
	  into a function that disables/enables ints (init code only for
	  now, resident code will follow later); cli/sti (disable/enable)
	  are now macros (inlined).
	* fnode.h, fatdir.c, fatfs.c: dir_read() no longer updates the
	  diroff counter. This eliminates the NEW field,
	  simplifies remove_lfn_entries(), and avoids a bug if you delete
	  the first entry in the root directory on FAT32.
	* fattab.c: Stop abusing "idx". The OW optimizer likes this better,too.
	* fattab.c: Clean up FAT12 handling -- there's a portable
	  approach that's small too.
	* fattab.c: Inlined much of getFATblock() into link_fat.