Date: Mon, 30 Jan 1995 10:21:56 -0500
To: djgpp AT sun DOT soe DOT clarkson DOT edu
From: cburke AT mitre DOT org (Carl D. Burke)
Subject: Apparent assembler problems...

With DJGPP 1.12M3, there seems to be a problem with the inb, outb, and les*
assembly codes.  inb and outb aren't reliable; I spent several frustrating
hours trying to install a custom keyboard handler which (depending on whether
I used _set or _chain vector) would either hang the system or return different
values for the same key on each press.  Using the inportb and outportb functions
in libpc, which used the same instruction text, works fine.  This implies I
could have a real problem if I need to rebuild the libraries.

les, lesb, lesw, and lesl aren't recognized as instructions at all.  This is
probably fine in protected mode, since what I really wanted to do was set up
the VGA interrupt to bang 256 palette entries at once and _simulate_int was
the way to go; but it is missing.

Where are the instruction templates for i386?  Is this something in the gas
source distribution?

--

Hope there's a workable solution in V2 for finding video memory, although if I
have to mess with bank switching it may pay to abandon DOS anyway and use a
Windows-based compiler.  The GRX 1.03M1 code just seems too slow for slamming
lots of memory to the screen.

Last note: the event management part of GRX 1.03M1 doesn't return mouse movement
events to me even when I enable all events.  I can get it myself on int33,
but it would be nice to know why the mouse never budges from 0,0.  Do I need to
setup some kind of viewport boundaries or something?

--

Anyway, even with the quirks, so far it's working great!  No crashes or hangs
that I can't explain with my own code, fairly fast compilation, and access to
a level of the machine that TCW never gave me.  And (for the most part) no more
memory model crap.

Carl Burke
cburke AT mitre DOT org