Subject: bulk building v7 binaries on a v8 leads to unusable packages, why???
To: None <port-sparc@netbsd.org>
From: Andreas Hallmann <hallmann@ahatec.de>
List: port-sparc
Date: 02/22/2006 18:50:19
Hi all,
my brain world is hit by disaster, i.e. I don't understand this.
With out any doubt I set up a netbsd-3-0-stable bulk building env for my
slow SS2 gateway on a 170 MHz turbo sparc.
Now building my needed packages only takes a few days .-))
But it results in a lot of garbage.
SS2: netbsd-3.0-stable.
SS5: netbsd-3.0-stable.
Both systems are compiled from exactly the same sources.
The SS2 with compile options "-pipe -mcpu=v7 -mtune=cypress"
and the SS5 with compile options "-pipe -mcpu=v8"
The bulk build sandbox was build on the SS5 (mksandbox) with a fresh
pkgsrc-2005Q4.
/usr/sandbox/etc/mk.conf is set up to use compile options
"-pipe -mcpu=v7 -mtune=cypress"
What I'm unable to understand:
Most packges generated this way dump core on my SS2.
On the other side, they just run as expected on the SS5.
Any ideas?????
I.e. most painful for me bind-9.3.1 dumps core with a signal 8, i.e.
floating point exception. Building the same package with the same
options on the SS2 result in a working bind-9.3.1.
Absolutely irritating.
OK, I should try packages compiled without -mpcu=v7.
But I would realy like to understand my observations.
Andreas