Date: Mon, 4 Mar 1996 15:42:57 +0000 (MET)
From: C M Marka <matriu AT gim DOT ktu DOT lt>
To: djgpp AT delorie DOT com
Subject: C++ packing/DPMI
Message-ID: <Pine.HPP.3.91.960304154104.4217A-100000@santaka.sc-uni.ktu.lt>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

!!!please ignore the return address, since this message is being posted
   from my friend's account!!!
Reply to: Martynas DOT Kunigelis AT vm DOT ktu DOT lt


	I found out how to make GNU C++ compiler pack the ffblk structure in
<dir.h> correctly, so you can use findfirst() and findnext() from your C++
programs. I know my method is undocumented and non-elegant, but it *works*. You
just need to edit your <dir.h> file like this:

----dir.h------
...
...
...

#ifdef  __cplusplus
extern "C" {
#pragma pack(1)        <---- here's our hack!
#endif

...
...
...
...

#ifdef __cplusplus
}
#pragma pack()         <---- here we go again 
#endif

...
...

	In any case, you'll be able to remove #pragmas from your file when GNU
fixes the bug in cc1plus [ and/or removes support for #pragma ].

	I've also some questions about DPMI, so I think they go to C. Sandmann:

[1] When try to allocate a real mode callback and pass the address of a real
    mode call structure to DPMI function 0x0303, does it copy structure's 
    contents to some internal buffer or does it use the original one (the one I
    passed)? If so, do I need to lock it?

[2] It is said in the FAQ that changing the PM int vector automatically changes
    the appropriate RM vector. Why and how?

 
I've spent whole weekend trying to get a mouse event handler to work, but all
I got were hundreds of reboots. Here's what I did [ in assembler ]:

1. lock the handler code and RM Call structure
2. allocate a real mode callback (passing a locked struct to DPMI)
3. `simulate real mode interrupt' 0x33 with AH=0x0C and ES:DX = return
   values of step 2 (plus other parms, and SS:SP = 0) in RM call struct.

I check  for errors after each step, but step 3 never returns. I though it
could be caused by too small stack, but it's not, I tested it: I allocated
some DOS mem, put a RETF opcode there, and did step 3, and it was ok.

After hundreds of failures I tried to write a RM timer int handler. The handler
just increased a counter on the screen (0xB8000 byte). It got called only 2
times, and then -- my my, reboot again. I think this is not caused by high 
interrupt frequency and reentrance problems -- default 18.2 secs is not that
high, is it?

So can someone help me or show a `live' example of setting a mouse event
handler with int 33/AH=0C? Or maybe a RM hw interrupt handler? Please?

P.S. I tried to check sb_lib, but it's rather messy. Please help.

Martynas Kunigelis

P.P.S. Could someone mail me any info about the vertical retrace interrupt?
All I know about vert. retr. for now is just how to wait for it... :(