Xref: news2.mv.net comp.os.msdos.djgpp:7715
From: ps AT ocisgi7 DOT unizh DOT ch (Serge Pachkovsky)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: -fomit-frame-pointer dangers: FIRE!
Date: 21 Aug 1996 07:57:10 GMT
Organization: University of Zurich, Switzerland
Lines: 31
Message-ID: <4vefgm$304@rzunews.unizh.ch>
References: <Pine DOT NEB DOT 3 DOT 95 DOT 960816133628 DOT 4971A-100000 AT ananke DOT amu DOT edu DOT pl> <321A06F5 DOT 40E8 AT stud DOT warande DOT ruu DOT nl>
NNTP-Posting-Host: ocisgi7.unizh.ch
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp

Elliott Oti (e DOT oti AT stud DOT warande DOT ruu DOT nl) wrote:

: info says that using -fomit-frame-pointer prevents debugging on i386 machines
: (never tried it meself), but I find Mark's question an interesting one.
: I try to make my programs as Win-friendly as possible, and they crash
: fairly often ( sound routines especially playing mods are the worst
: culprits), but the switch has little to do with it. I now use it all
: the time, as the extra register makes ALL the difference. Still -- where there's
: smoke, there's fire . Only, where's the fire? Any of you guys seen fire?

Oh, sure. Quite a lot of, actually ;-)
The problem with -fomit-frame-pointer under DPMI is that it frees up an
extra register (bp) for general use. x86 use SS: as the default selector
when bp is used as an address base register, and if SS: happen to be
unequal to DS:, and your data is in DS:, you lose with segmentation fault.

So, as long all your pointers are to stack-based objects (or you are not
using pointers at all) you are free to compile with -fomit-frame-pointer.
It should be fairly easy to fix in gcc (just emit an extra ds: segment
override if using bp as a base register and -fomit-frame-pointer is in 
effect). For all I know, it might have already been fixed (my experience
with gcc on x86 ends some time around 2.6.3, and that's history by now).

Regards,

/Serge.P

--

Russian guy from the Zurich university...