Re: toolchain/42014: Can't cross-compile netbsd 5.0.1 toolchain on snow leopard

[Andreas Wrede <andreas%wrede.ca@localhost>]

On Sep 28, 2009, at 15:40 , Soren Jacobsen wrote:

> The following reply was made to PR toolchain/42014; it has been  
> noted by GNATS.
>
> From: Soren Jacobsen <snj%pobox.com@localhost>
> To: gnats-bugs%NetBSD.org@localhost
> Cc: urumisama%excite.co.jp@localhost, Alan Barrett <apb%netbsd.org@localhost>
> Subject: Re: toolchain/42014: Can't cross-compile netbsd 5.0.1  
> toolchain on
> snow leopard
> Date: Mon, 28 Sep 2009 12:35:19 -0700
>
> On Sep 25, 2009, at 9:50 AM, Alan Barrett wrote:
>
> > This problem was fixed in -current by the changes described in
> > <http://mail-index.netbsd.org/source-changes/2009/02/03/
> > msg216739.html>.
> > Do the same changes fix the problem in netbsd-5?
>
> Yes, they do fix the immediate mtio.h problem, but a tools build still
> fails.

For me too. Can those changes please be pulled up to netbsd-5?

> Log at http://www.netbsd.org/~snj/sllog.txt

I don't see your tool build failure:

dependall ===> binstall
/u1/netbsd-5.0/src/tools/binstall/../../usr.bin/xinstall/xinstall.c:  
In function 'metadat
a_log':
/u1/netbsd-5.0/src/tools/binstall/../../usr.bin/xinstall/xinstall.c: 
1133: warning: format
 '%ld' expects type 'long int', but argument 4 has type  
'__darwin_suseconds_t'
install ===> binstall
...

The build fails in the distribution step:

checkflist ===> distrib/sets
sh: line 2: 95271 Done                    ( cat /u1/netbsd-5.0/src/etc/ 
mtree/NetBSD.dist; echo '/unset all'; cat /u1/netbsd-5.0/destdir.i386/ 
METALOG )
     95272 Segmentation fault      | /u1/netbsd-5.0/tools/bin/nbmtree  
-C -k all -N /u1/netbsd-5.0/src/etc > /u1/netbsd-5.0/destdir.i386/ 
METALOG.new
--- /u1/netbsd-5.0/destdir.i386/METALOG.sanitised ---
*** [/u1/netbsd-5.0/destdir.i386/METALOG.sanitised] Error code 139
1 error

nbmake: stopped in /u1/netbsd-5.0/src/distrib/sets

The segfault occurs in nbmtree, line 336 in mtree/spec.c. It's caused  
by junk being returned from the group_from_gid(), with in turn I think  
is caused by a mixup of where group_from_gid() comes from, nbcompat or  
the native libc. I got as far as the comment staring at lines 332 in  
src/tools/compat/compat_defs.h.

As a work-around, providing a header for group_from_gid() in spec.c  
makes nbmtree work.

nbpax segfaults as well:
--- kernset-XEN3_DOM0 ---
sh: line 1: 60410 Segmentation fault      GZIP=-9 /u1/netbsd-5.0/tools/ 
bin/nbpax -O -zw -M -N /u1/netbsd-5.0/src/etc -f /u1/netbsd-5.0/ 
release/i386/binary/sets/kern-XEN3_DOM0.tgz


I suspect it's the same issue since it uses group_from_gid() as well.

--
    aew