Subject: Re: build failing
To: Jon Olsson <email@example.com>
From: David Hopper <firstname.lastname@example.org>
Date: 08/05/2003 20:25:18
Jon Olsson wrote:
> On Tue, Aug 05, 2003 at 01:43:37PM +0000, Christos Zoulas wrote:
>>In article <Pine.NEB.email@example.com>,
>>Greywolf <firstname.lastname@example.org> wrote:
>>>Okay. Made cleandir; made clean. cvs up about an hour ago.
>>>It is now Mon Aug 4 22:24:05 PDT 2003.
>>>mk.conf has, among other things:
>>>Yet, for some bloody blasted inexplicable reason, my build insists on making
>>>references to bleeding kerberos, somewhere in libexec/telnetd, and it's also
>>>not finding libdes.so.7 for some ridiculous reason.
>>>Can someone *please* explain this?
>>rm -f /lib/libdes* /usr/lib/libdes* and restart your build.
> I had a similar problem yesterday.
> Maybe this should be added to src/UPDATING?
I do hope everyone will forgive my griping this once... But I have always
understood the intention of the build.sh system to be completely
self-hosting. Is this just an unrealistic goal? Why does the build
environment consistently depend upon the state of my userland?
I am at a ratio now where builds upon cvs updates (attempted daily over the
period of a week every couple of months) only succeed perhaps 20% of the
time on systems that only backdate -current by a couple months. Is this
just bad timing? Am I just nailing the build bugs? Complete 'build.sh 
release' runs without userland dependency errors are extremely rare, and
very valuable to me, as the operator of a -current production system at our
company (so shoot me...).
I will say, as a testament to everyone involved, that when -current does
build without error, it tends to be extremely reliable, pthread growing
pains notwithstanding, on my pack of Alphas and Amigas.
So please forgive my ignorance of the build process, but why the dependency
upon userland libraries when the build is bootstrapped?