NetBSD Problem Report #26146
Received: (qmail 11264 invoked by uid 605); 2 Jul 2004 23:14:46 -0000
Message-Id: <200407022314.i62NEhQL009654@knox.mines.edu>
Date: Fri, 2 Jul 2004 17:14:43 -0600 (MDT)
From: jbernard@mines.edu
Sender: gnats-bugs-owner@NetBSD.org
Reply-To: jbernard@mines.edu
To: gnats-bugs@gnats.NetBSD.org
Subject: Changes to make parallel/pvm3 build under Darwin
X-Send-Pr-Version: 3.95
>Number: 26146
>Category: pkg
>Synopsis: Changes to make parallel/pvm3 build under Darwin
>Confidential: no
>Severity: non-critical
>Priority: low
>Responsible: darwin-pkg-people
>State: closed
>Class: sw-bug
>Submitter-Id: net
>Arrival-Date: Fri Jul 02 23:15:00 +0000 2004
>Closed-Date: Sun Jul 18 18:06:20 +0000 2004
>Last-Modified: Sun Jul 18 18:06:20 +0000 2004
>Originator: Jim Bernard
>Release: pkgsrc as of Fri Jul 2 09:37:43 UTC 2004
>Organization:
>Environment:
System: Mac OS X 10.3.4
Architecture: ppc
Machine: ppc
>Description:
The pvm package needs a few modifications to get it to build
under Darwin.
>How-To-Repeat:
Try it.
>Fix:
arch.mk: Recognize Darwin and set _PVM_OPSYS and _PVM_ARCH
appropriately.
These introduce 4 new patches to the patches directory:
work/pvm3/conf/DARWIN.def: Incorporate RSHCOMMAND like the
NetBSD .def files do. Eliminate FAKEXDRFLOAT (no longer
needed---this source is pretty old) and the -I flag (also
no longer needed). Add flags to eliminate /usr/local/include
from the system paths searched by cpp (see PR 26143 for
details; since aimk is used here, the change to sys.mk
given in that PR is not sufficient to eliminate
/usr/local/include in this build; but maybe buildlink3
could be modified to handle it).
work/pvm3/tracer/{tracer.h,trclib.h,trcsort.c}: Include
stdlib.h instead of malloc.h if HASSTDLIB. This is not
Darwin specific, but I believe it is generally correct.
And, of course, distinfo must be modified.
Here are the 5 patches:
--- arch.mk-dist Thu Mar 20 16:34:36 2003
+++ arch.mk Tue Jun 29 15:32:41 2004
@@ -24,6 +24,9 @@
.elif ${OPSYS} == "SunOS"
_PVM_OPSYS= SUN4
_PVM_ARCH= SOL2 # Solaris (SunOS 5.*)
+.elif ${OPSYS} == "Darwin"
+_PVM_OPSYS= DARWIN
+_PVM_ARCH= # empty
.else
_PVM_OPSYS= UNKNOWN
_PVM_ARCH= # empty
--- conf/DARWIN.def.orig Thu Sep 27 13:56:33 2001
+++ conf/DARWIN.def Fri Jul 2 16:33:29 2004
@@ -1,9 +1,8 @@
ARCHCFLAGS = -DSOCKADHASLEN -DNOREXEC \
- -DRSHCOMMAND=\"/usr/bin/rsh\" \
+ -DRSHCOMMAND=\"${SSH}\" \
-DHASSTDLIB -DNEEDMENDIAN -DHASERRORVARS \
- -DFAKEXDRFLOAT -DSYSERRISCONST \
- -I/System/Library/Frameworks/System.framework/Headers/bsd/sys
- # this last -I flag needed for tracer build...
+ -DSYSERRISCONST \
+ -nostdinc -isystem /usr/include/gcc/darwin/default -isystem /usr/include
ARCHDLIB =
ARCHDOBJ =
ARCHLIB =
--- tracer/tracer.h.orig Fri May 11 15:19:43 2001
+++ tracer/tracer.h Tue Jun 29 14:23:21 2004
@@ -40,8 +40,12 @@
#include "pvmwin.h"
#include <time.h>
#endif
+#ifdef HASSTDLIB
+#include <stdlib.h>
+#else
#ifndef IMA_MACOSX
#include <malloc.h>
+#endif
#endif
#include <signal.h>
#include <ctype.h>
--- tracer/trclib.h.orig Fri May 11 15:19:43 2001
+++ tracer/trclib.h Tue Jun 29 14:31:42 2004
@@ -34,8 +34,12 @@
#include "pvmwin.h"
#endif
#include <time.h>
+#ifdef HASSTDLIB
+#include <stdlib.h>
+#else
#ifndef IMA_MACOSX
#include <malloc.h>
+#endif
#endif
--- tracer/trcsort.c.orig Fri May 11 15:19:44 2001
+++ tracer/trcsort.c Tue Jun 29 14:41:00 2004
@@ -31,8 +31,12 @@
#include <sys/types.h>
#include <sys/stat.h>
+#ifdef HASSTDLIB
+#include <stdlib.h>
+#else
#ifndef IMA_MACOSX
#include <malloc.h>
+#endif
#endif
#ifdef SYSVSTR
>Release-Note:
>Audit-Trail:
Responsible-Changed-From-To: pkg-manager->darwin-pkg-people
Responsible-Changed-By: recht
Responsible-Changed-When: Fri Jul 2 23:33:31 UTC 2004
Responsible-Changed-Why:
over to darwin-pkg-people
From: Jim Bernard <jbernard@mines.edu>
To: gnats-bugs@gnats.netbsd.org
Cc:
Subject: Re: pkg/26146: Changes to make parallel/pvm3 build under Darwin
Date: Thu, 8 Jul 2004 16:58:53 -0600
It turns out that, in addition to the patches needed to make pvm3 build,
one more thing is necessary to make it usable: ranlib. Under Mac OS X,
the timestamp on the library index must be newer than the creation time
of the file, so when the libraries are installed, the copies are not
usable for linking. This could be fixed by copying them with the
creation time preserved, or by running ranlib on them after they are
installed. I suspect this is not an issue for binary packages, but I
haven't checked.
Here's a patch to the pkg Makefile that does ranlib on the installed
libraries:
--- Makefile-dist Mon May 31 21:09:09 2004
+++ Makefile Thu Jul 8 11:53:53 2004
@@ -67,6 +67,9 @@
${INSTALL_PROGRAM} ${WRKSRC}/lib/${PVM_ARCH}/pvm* \
${PVM_DIR}/lib/${PVM_ARCH}
${INSTALL_DATA} ${WRKSRC}/lib/${PVM_ARCH}/*.a ${PVM_DIR}/lib/${PVM_ARCH}
+.if ${OPSYS} == "Darwin"
+ ranlib ${PVM_DIR}/lib/${PVM_ARCH}/*.a
+.endif
for f in `${LS} -1 ${WRKSRC}/lib`; do \
if [ -f ${WRKSRC}/lib/$$f ]; then \
${INSTALL_SCRIPT} ${WRKSRC}/lib/$$f ${PVM_DIR}/lib; \
--Jim
From: Dan McMahill <dmcmahill@NetBSD.org>
To: jbernard@mines.edu
Cc: gnats-bugs@gnats.NetBSD.org
Subject: Re: pkg/26146: Changes to make parallel/pvm3 build under Darwin
Date: Thu, 15 Jul 2004 18:19:29 -0400
> >Number: 26146
> >Category: pkg
> >Synopsis: Changes to make parallel/pvm3 build under Darwin
> work/pvm3/conf/DARWIN.def: Incorporate RSHCOMMAND like the
> NetBSD .def files do. Eliminate FAKEXDRFLOAT (no longer
> needed---this source is pretty old) and the -I flag (also
> no longer needed). Add flags to eliminate /usr/local/include
> from the system paths searched by cpp (see PR 26143 for
> details; since aimk is used here, the change to sys.mk
> given in that PR is not sufficient to eliminate
> /usr/local/include in this build; but maybe buildlink3
> could be modified to handle it).
I was about to make some other changes to pvm3 and wanted to get this
PR addressed first. So what happens if pvm3 looks in /usr/local/include?
It is causing a problem here?
What if you are using a different gcc, will these changes break it
because of the -isystem /usr/include/gcc/darwin/default part?
-Dan
--
From: Jim Bernard <jbernard@mines.edu>
To: Dan McMahill <dmcmahill@NetBSD.org>
Cc: gnats-bugs@gnats.NetBSD.org
Subject: Re: pkg/26146: Changes to make parallel/pvm3 build under Darwin
Date: Thu, 15 Jul 2004 21:41:25 -0600
On Thu, Jul 15, 2004 at 06:19:29PM -0400, Dan McMahill wrote:
> > >Number: 26146
> > >Category: pkg
> > >Synopsis: Changes to make parallel/pvm3 build under Darwin
>
> > work/pvm3/conf/DARWIN.def: Incorporate RSHCOMMAND like the
> > NetBSD .def files do. Eliminate FAKEXDRFLOAT (no longer
> > needed---this source is pretty old) and the -I flag (also
> > no longer needed). Add flags to eliminate /usr/local/include
> > from the system paths searched by cpp (see PR 26143 for
> > details; since aimk is used here, the change to sys.mk
> > given in that PR is not sufficient to eliminate
> > /usr/local/include in this build; but maybe buildlink3
> > could be modified to handle it).
>
> I was about to make some other changes to pvm3 and wanted to get this
> PR addressed first.
Thanks for looking at this!
> So what happens if pvm3 looks in /usr/local/include?
> It is causing a problem here?
Good question. In my particular case, IIRC, there was no specific
trouble with pvm3 itself. But, as you can imagine, whenever any
source file in a pkg build includes a system header file and cpp looks
in /usr/local/include before looking in the real system directories,
there's potential for bad things to happen, and whether they occur
depends on what happens to be in /usr/local/include at the time of the
build. So to be sure there's no possibility of breakage, we need to
remove /usr/local/include from the list of system include directories.
An example from the pkgsrc bootstrap problem described in 26143 is
that the bootstrap succeeded on one system and failed on another.
The reason turned out to be the presence of db.h from a db4 installation
in /usr/local/include on the box where the bootstrap failed.
One of the changes I suggested in 26143 was to add the relevant
flags in sys.mk, but since pvm3 uses aimk instead of bmake to build,
if we want to be sure to exclude /usr/local/include, it's not enough
to rely on the change to sys.mk; the change has to be made locally.
Perhaps more precisely, CPPFLAGS is the appropriate place to add the
flags in sys.mk for bmake, but for pvm3/aimk, it's in ARCHCFLAGS,
which gets added to CFLAGS (and CPPFLAGS is not used, so even if
bmake passes that to aimk in the environment, it doesn't help).
I have a suspicion that buildlink3 may really be the right tool
to accomplish all this, but I'm not familiar enough with it to be
sure.
> What if you are using a different gcc, will these changes break it
> because of the -isystem /usr/include/gcc/darwin/default part?
I think they would break it, yes. Unfortunately, there doesn't seem
to be any way to delete a directory from the list of system include
directories. It's apparently only possible to delete all of them
and to add new ones to the list.
I wonder if it's perhaps best just to declare Apple's compiler broken
and make all of pkgsrc require lang/gccwhatever. But are there also
important beneficial features in Apple's compiler?
--Jim
From: Dan McMahill <dmcmahill@NetBSD.org>
To: Jim Bernard <jbernard@mines.edu>
Cc: gnats-bugs@gnats.NetBSD.org
Subject: Re: pkg/26146: Changes to make parallel/pvm3 build under Darwin
Date: Fri, 16 Jul 2004 08:46:56 -0400
> Good question. In my particular case, IIRC, there was no specific
> trouble with pvm3 itself. But, as you can imagine, whenever any
> source file in a pkg build includes a system header file and cpp looks
> in /usr/local/include before looking in the real system directories,
> there's potential for bad things to happen, and whether they occur
> depends on what happens to be in /usr/local/include at the time of the
> build. So to be sure there's no possibility of breakage, we need to
> remove /usr/local/include from the list of system include directories.
> One of the changes I suggested in 26143 was to add the relevant
> flags in sys.mk, but since pvm3 uses aimk instead of bmake to build,
> if we want to be sure to exclude /usr/local/include, it's not enough
> to rely on the change to sys.mk; the change has to be made locally.
> Perhaps more precisely, CPPFLAGS is the appropriate place to add the
> flags in sys.mk for bmake, but for pvm3/aimk, it's in ARCHCFLAGS,
> which gets added to CFLAGS (and CPPFLAGS is not used, so even if
> bmake passes that to aimk in the environment, it doesn't help).
>
> I have a suspicion that buildlink3 may really be the right tool
> to accomplish all this, but I'm not familiar enough with it to be
> sure.
buildlink3 can probably be used within pkgsrc to fix up the include
path. I guess the issue is that some users may use aimk to build
a program which uses pvm3 and that program may reside outside of pkgsrc.
In this case bl3 is not around. Of course I really dislike aimk
and its possible no one is doing that anyway. It turns out that the
two packages within pkgsrc which depend on pvm do not use aimk and
hence end up ignoring $PVM_ARCH.def anyway (which I suppose may be
questionable).
> > What if you are using a different gcc, will these changes break it
> > because of the -isystem /usr/include/gcc/darwin/default part?
>
> I think they would break it, yes. Unfortunately, there doesn't seem
> to be any way to delete a directory from the list of system include
> directories. It's apparently only possible to delete all of them
> and to add new ones to the list.
>
> I wonder if it's perhaps best just to declare Apple's compiler broken
> and make all of pkgsrc require lang/gccwhatever. But are there also
> important beneficial features in Apple's compiler?
Are they using gcc or something else? Have people complained to apple
about including /usr/local/include by default and in fact including it
before the system libraries?
--
From: Jim Bernard <jbernard@mines.edu>
To: Dan McMahill <dmcmahill@NetBSD.org>
Cc: gnats-bugs@gnats.NetBSD.org
Subject: Re: pkg/26146: Changes to make parallel/pvm3 build under Darwin
Date: Fri, 16 Jul 2004 08:23:36 -0600
On Fri, Jul 16, 2004 at 08:46:56AM -0400, Dan McMahill wrote:
>
> buildlink3 can probably be used within pkgsrc to fix up the include
> path. I guess the issue is that some users may use aimk to build
> a program which uses pvm3 and that program may reside outside of pkgsrc.
I'd say that's outside the "responsibility" of pkgsrc. This problem
with the Apple compiler could cause trouble to anyone building anything,
regardless of whether pkgsrc is involved. Maybe we should document it
in a README somewhere, so Darwin users of pkgsrc will at least be aware
of the potential for lossage.
> > I wonder if it's perhaps best just to declare Apple's compiler broken
> > and make all of pkgsrc require lang/gccwhatever. But are there also
> > important beneficial features in Apple's compiler?
>
> Are they using gcc or something else? Have people complained to apple
> about including /usr/local/include by default and in fact including it
> before the system libraries?
It is gcc (3.3, IIRC), but with some Apple-specific additions (clearly
marked in the gcc man page), and with Apple's choice of "system" include
directories. I don't know if anyone has complained or not (I'm not at
all attuned to the Apple user community), but I think someone should :-).
--Jim
State-Changed-From-To: open->closed
State-Changed-By: dmcmahill
State-Changed-When: Sun Jul 18 18:05:34 UTC 2004
State-Changed-Why:
patches (except the compiler include path part) applied.
thanks.
>Unformatted:
(Contact us)
$NetBSD: query-full-pr,v 1.51 2026/08/10 02:28:17 riastradh Exp $
$NetBSD: gnats_config.sh,v 1.10 2026/05/13 22:00:09 riastradh Exp $
Copyright © 1994-2026
The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.