Date: Fri, 29 Dec 1995 22:56:59 -0500
From: Andrew Szymkowiak <aes AT rosserv DOT gsfc DOT nasa DOT gov>
To: djgpp AT delorie DOT com
Subject: profiling in V2 beta 4

  First, let me thank DJ and all the other developers for all the work
that you have done on DJGPP.  It appears to be an excellent effort.
  I have two observations about operations with V2, which I though I
would send in so that the FAQs/documentation/info files could be
updated for the release (what does FCS stand for anyway?).  I have
read the FAQs, performed some searches on the mailing list archives,
and generally made some attempt to not appear as a total newbie,
but apologize if I am not the first to cover this ground.
(These observations based on the beta 4 release, using CWSDPMI
as the dpmi server, from DOS ( i.e. not within a Windows DOS box).

[1] If you "gcc -p" the ld will fail, unable to find mcrt0.o.  Since
the distribution does not come with "prof" (only "gprof"), everyone
should be using "-pg" instead anyway, but I thought I would mention
the absence, in case it is easy to add to a library somewhere.

[2] GDB chokes on any programs GCC'ed with "-pg".  It is easy for
me (in hindsight) to think of reasons why this might be (timer
interrupt is hooked and paging might be occurring, etc), but I
wasted some time before I discovered this.  If the profiled program
is run under GDB, you may get a SIGSEGV; if you try to step within
it, you may get a GPF, with no addresses that symify can identify.
Any chance that GDB could recognize that "-pg" was used and sound
an alarm?

Thanks,
   Andy S.

Andrew Szymkowiak
X-ray Astrophysics Branch
NASA/Goddard Space Flight Center