From xemacs-m  Thu Jul  3 14:45:46 1997
Received: from black-ice.cc.vt.edu (black-ice.cc.vt.edu [128.173.14.71])
	by xemacs.org (8.8.5/8.8.5) with ESMTP id OAA03234
	for <xemacs-beta@xemacs.org>; Thu, 3 Jul 1997 14:45:46 -0500 (CDT)
Received: from black-ice.cc.vt.edu (LOCALHOST [127.0.0.1])
	by black-ice.cc.vt.edu (8.8.6/8.8.6) with ESMTP id PAA21190
	for <xemacs-beta@xemacs.org>; Thu, 3 Jul 1997 15:45:47 -0400
Message-Id: <199707031945.PAA21190@black-ice.cc.vt.edu>
To: xemacs-beta@xemacs.org
Subject: Life on the bleeding edge - 20.3-b10 loses on AIX 4.2.1
From: Valdis.Kletnieks@vt.edu
X-Url: http://black-ice.cc.vt.edu/~valdis/
Mime-Version: 1.0
Content-Type: multipart/signed; boundary="==_Exmh_817810932P";
	micalg=pgp-md5; protocol="application/pgp-signature"
Content-Transfer-Encoding: 7bit
Date: Thu, 03 Jul 1997 15:45:46 -0400

--==_Exmh_817810932P
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: quoted-printable

I finally got around to building a beta since The Great Autoconf overhaul=
,
and hit a bunch of problems.

Scenario:  AIX 4.2.1, xemacs-20.3-b10.

1) IBM xlc 3.1.4.6 *loses* when using --use-union-type=3Dyes.  I have rep=
orted
this to IBM.

2) autoconf problem - down in src/m/ibmrs6000.h, there's a link spec that=

includes $(srcdir)/ in it.  This is *NOT* expanded correctly, and the lin=
ker
then complains (quite correctly) that it can't find the .inp file specifi=
ed.
I suspect a shell quoting problem someplace.

3) autoconf problem - it fails to auto-detect X11 (and probably a pile of=
 other
stuff) because the link spec in ibmrs6000.h specifies '-bnso,-bnodelcsect=
'.
This in effect says "ignore shared objects and link statically".  This ho=
wever
loses when looking for X11, because it looks for Xt routines, which then =
fail
to link because Xt is a shared object - and the -bnso loses the fact that=
 Xt
has pointers to the *other* libraries needed.  So we don't have a pointer=
 to =

-lX11 either in the command line or in the shared object, and we lose.  T=
he
fix here is to either lose the -bnso, or include *all* the libraries need=
ed
to compile the configure test cases.  I decided to lose the -bnso, and fo=
und:

4) You probably don't need the syscalls.ext in ibmrs6000.h either if you'=
re
not linking statically - the libc.a object incorporates pointers to these=
 so
you get duplicate definition warnings if you include this.

I "fixed" (2)-(4) by hacking the ibmrs6000.h link specs - I ended up usin=
g:

define LD_SWITCH_MACHINE "-Wl,-bI:/home/spd/valdis/src/xemacs-20.3-b10/sr=
c/src/
m/ibmrs6000.inp,AIX_SMT_EXP"

which got it through the link process.  After all that, I had the followi=
ng
problems:

A) Compiled with -O2 it lost big time in temacs while doing loadup.el.

B) Compiled with -g to debug problem (a) it undumped OK, but dies while
easily.  'xemacs -q -no-site-file', then typing control-h i to get into
info, dies gloriously.  IBM's dbx itself SEGV's trying to start up, and
gbd 4.16 says thusly:

(gdb) where
No stack

Off to go install gcc 2.7.2.2 to see if it loses any differently....





-- =

				Valdis Kletnieks
				Computer Systems Senior Engineer
				Virginia Tech



--==_Exmh_817810932P
Content-Type: application/pgp-signature

-----BEGIN PGP MESSAGE-----
Version: 2.6.2

iQCVAwUBM7wBadQBOOoptg9JAQF89gQAw+tWbit/IK/4FtOEH5LgusdQvOvX7v6M
YA9HO2hSIFFn9Rkna3C5U8oUbhRk/nPgOxkQnMTtdTFzLE8Nebo7MLunGi3uL3xQ
EhWLY2xiwaVT32YmMtya/VDut1O3XwK4+1CkrxD65Zbuv0EPYqbCd3uR3rThfqAP
Z5UaT1WtdMM=
=HGCp
-----END PGP MESSAGE-----

--==_Exmh_817810932P--

