Return to PORT_TO_WIN32 CVS log

Up to [GPL Stacker compression] / dmsdos

DMSDOS 0.9.2.2

Porting dmsdos to Win32
-----------------------

If you want to use dmsdos under a 32 bit Microsoft Windows system such as
Win95/98 or Win NT 4.0 there are two possibilities: try it at user level
(easy) or at driver level (difficult but more interesting).

Don't try a 16 bit environment. You won't get it work as dmsdos needs more
than 1MB of temporary memory. Buy the appropriate MS-DOS or Stacker version 
instead. They're also faster because they are said to contain hand-optimized
assembler code.

At least, there's no Win NT version of Doublespace, Drivespace or Stacker.
This makes dmsdos somewhat interesting for Win32 :)

Well you can also try a 32 bit extended Dos environment such as DJGPP.


A. User Level
-------------

I tried, and I found Win32 programming is not fun. At least if you are
used to programming in a Unix environment.

It _should_ be easy to compile libdmsdos and, for example, dcread under any 
32 bit environment. Well, I tried Win NT 4.0 with Cygnus' Win32 B19 version 
of gcc, but almost nothing worked. The configure script, which runs under
pure bash, didn't run with the bash found in that package. So what, I then
configured dmsdos under Linux and copied the config files. With a lot of
hacking, I finally managed to compile libdmsdos, but it was of no use.
Whatever I did, when linking any dmsdos utility against the library I just 
managed to crash the linker :(( 
I have really no idea what stupid things I may have done...
Looks like this gcc port is not yet ready for daily use. Sigh. Though it's 
really an impressive work: all gnu tools under NT...

Okay, now Cygnus have released B20 version of their Win32 gcc port, and -
guess what - it _worked_, almost out of the box. Even the Configure script
runs correctly now. There is a special Makefile, Makefile.cygwinb20,
that should be used for this compiler. One thing, though, is bad: the
compiler does neither support Unix flock nor Win32 sopen. This means the
generated code does not do file locking. Be sure that there is never
more than one process that is using a CVF through libdmsdos.

The next chance I gave to Microsoft Visual C++ 5.0. Again almost nothing
worked (and yes, the IDE also crashed awaking Dr. Watson when I just clicked
around in the help in order to learn how to use the IDE...). No need to say
that I had to configure dmsdos under Linux. The complete Makefile was even
rejected by nmake. So what. I finally wrote a script win32_msc50.bat
that compiles the necessary things in a Win32 _console_. Currently, the
only useful application is mcdmsdos: it can extract files from a CVF in
a Win32 environment. (Just tested: the executable runs even under wine in 
an xterm :-) )

I did not try other compiler versions. It may work with the DJGPP Dos port 
of gcc and the EMX OS/2 port of gcc. But you may find that the long filenames
the dmsdos sources use cause a lot of problems there...
Now there's a script called "prepare_dos_8.3" in the src directory that
translates all relevant dmsdos files to use 8.3 names. There's no further
documentation about this. You are expected to take a look at the script.
You are also expected to write your own Makefile for compilation in such an
environment. There are lots of problems that may occur in a DOS environment.
Sorry, but I can't test everything.

Okay, that's what should work under every compiler:

* Configure, compile and install dmsdos under Linux.
* Copy the whole directory where it compiled to a vfat partition.
* Boot a Win32 system and edit the script win32_msc50.bat (on my system
  I had to setup include and library paths there). If your compiler is not
  Microsoft Visual C++ 5.0 this may not work and you may have to write
  a new script from scratch.
* Run the script with errors directed into a file.
* Hack the source where it fails. Please be sure to surround your changes
  with appropriate #ifdef's (unless they are real bugs) so I can integrate
  them into next dmsdos release. Let me know what you had to change and why.


B. Driver Level
---------------

That's indeed possible. At least, there has been some discussion about using
libdmsdos for writing a Win95/98 VXD or a Win NT 4.0 FS driver.

Well, I refuse to do that. I'm not a Win32 driver hacker and my knowledge
about the Win95/98 or Win NT driver interface is almost zero. But some
notes may be interesting, though.

Note that libdmsdos is *not* reentrant. If the Win32 system does not ensure 
exclusive execution of the driver code you must write your own *exclusive* 
locking interface around all libdmsdos calls, even in read-only mode.

Beyond that, you will have to write a full-featured FAT filesystem
driver that calls the libdmsdos cluster read and write routines, and link it
against libdmsdos. That's almost all. That's not difficult.
As FAT is a simple filesystem layout, and the CVFs are FAT12 or FAT16 but
never FAT32, each experienced Dos programmer should be able to write that 
from scratch, at least for 8.3 filenames which would be enough for 
functionality :) The first trial is always read-only, and that should not be
too much work.

Note that libdmsdos references some standard C library functions that are
not available at driver level. You must write an emulation for all these 
functions and link it to the driver instead of the standard C library. The 
emulation needn't be full-featured like the C library. Look at the source 
if you are in doubt. Well, the read/write/lseek emulation is actually all 
the low-level disk access :)

See doc/libdmsdos.doc for further information about the dmsdos library
(including a list of referrenced C library functions and a detailed
description of the dmsdos functions the library exports).

Let me know if you are a Win32 hacker and want to start such a project :)

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.