Date:	Thu, 25 Apr 1996 00:50:43 -0800 (GMT)
From:	"Rafael R. Sevilla" <rsevilla AT upd DOT edu DOT ph>
To:	mdruiter AT cs DOT vu DOT nl, DJGPP newsgroup mail <djgpp AT delorie DOT com>
Subject:	Re: djgpp compiling in 4 megs
Message-ID:	<Pine.SOL.3.91.960425002529.1163B-100000@sauron>
MIME-Version:	1.0

The only thing I can think of that causes that (print-screens in
unexpected places) is a very odd instruction which Intel introduced with
the 80186, namely BOUND. Officially, what this instruction does is check
array bounds with its two arguments an index and a pointer to the start of
the array (which should begin with the array's limits), and if the index
happens to be outside the array bounds, it issues an INT 5, which, quite
unfortunately, IBM appropriated back in '81 for its BIOS printscreen
interrupt. This didn't cause problems for the original PC and XT, which
both used the 8088 and consequently didn't recognize the BOUND
instruction. But come the 80286 and the AT, and subsequent more advanced
processors, this caused some fairly serious problems. A bug that caused a
jump to a byte that contained 0x62 (the BOUND opcode), will probably cause
the "bound exception," as its two arguments will hit some random spot in
memory, and the success or failure depends on what's in the next few
bytes... And what's worse, the CS:EIP pushed by the exception interrupt
will be the address of the errant instruction, and once PrtSc returns it
will of course fail again, print the screen again... DISASTER! I don't
think BOUND gets a lot of use, as C/C++ code doesn't need it (does the C
standard say anything about checking array bounds?), and that probably
excempts all the programs you've mentioned, *provided they were written
correctly*. Now if there were a bug in some program you're using that did
that fatal jump to a BOUND opcode, now we're getting somewhere. Blame Big
Blue and their BIOS. Intel warned us that it reserved INT 00-1F for its
own internal use, but Big Blue just *had* to use Int 5! 

And btw, I'm also using djgpp under 4MB, and on a 386 to boot! But then 
again, I've never used GRX20, nor cc1plus. Almost all my work is confined 
to C, and I haven't had any serious problems. Yet :)

-------------------------------------------------------------------------
University of the Philippines                           Rafael R. Sevilla
PABX/OVCA: Diliman Networking Project          rsevilla AT sauron DOT upd DOT edu DOT ph
-------------------------------------------------------------------------

- Raw text -