possible -mdpmi extensions

[stsp]

Hi.

Currently it seems -mdpmi does the
same as -mdosx, and -mdpmi32 seems
not available.
I'd like to propose a few changes to
that picture to get more value from -mdpmi.

1. add -mdpmi32
2. in mdpmi mode prefer the existing DPMI.
   Currently cw32 does everything possible
   to avoid the installed dpmi: it is trying to
   install its own dpmi from real-mode or via
   VCPI. This is perfectly fine for -mdosx, but
   should probably be changed for -mdpmi.
3. query for "MS-DOS" extension, and if it is
   there - restore all interrupt vectors to their
   original values, effectively disabling cw32.
   It is not needed in -mdpmi mode when
   "MS-DOS" extension is there, as it means
   the extender is already there.
4. add a function to get the old value of an
   interrupt vector (i.e. the value before cw32
   set it to its own handler). This will allow to
   communicate with the DPMI server and
   bypass cw32 when needed. If there is no
   "old" value (i.e. cw32 is not hooking that
   vector), then such func should return NULL
   or some error.

Thanks!

[tkchia]

Hello @stsp,

In mdpmi mode prefer the existing DPMI. Currently cw32 does everything possible to avoid the installed dpmi: it is trying to install its own dpmi from real-mode or via VCPI.

CauseWay will prefer using an existing DPMI server if it sees a CAUSEWAY=DPMI environment variable setting. But indeed, it will be good if there is a way to specify this as a system flag setting in a compiled program.

query for "MS-DOS" extension, and if it is there - restore all interrupt vectors to their original values

I do not think there is any simple way to figure out if a DPMI host does full int 0x21 translation — if you know of such a way, I would like to hear about it. The best I can think of so far is to maybe pass an invalid pointer to some int 0x21 service, and see if a crash happens (?).

Thank you!

[stsp]

Just like I said, you need to query
dpmi for "MS-DOS":
http://www.ctyme.com/intr/rb-4624.htm
But of course if you provide the func
to get the old int vector value (and err
if its not hooked by cw32), then it would
be easy to implement in client's code.
So whether or not it should be automatic,
is probably a matter of the taste.