Catching up on misc@... Inline reply.
On 03/23/12 15:48, Jan Stary wrote:
On Mar 23 13:58:08, RD Thrush wrote:
On 03/23/12 07:25, Stuart Henderson wrote:
On 2012/03/23 07:11, RD Thrush wrote:
My PORTSDIR is on an nfs server. Mounting the particular nfs
directory on /usr/ports (and setting PORTSDIR accordingly) fails the
new test in bsd.port.mk.
Apparently test -h considers an nfs mount the same as a symlink...
Hmm, that's weird. It doesn't for me, my PORTSDIR is on an NFS server
and it works fine with -current bsd.port.mk.
Probably better to work out why test is broken for you.
Before the recent bsd.port.mk change, I used dpb to build ports with
a symlink to an NFS mounted directory (on a -current NFS server),
ie.:
v1:tmp/rd 130>ls -ld /usr/ports
lrwxr-xr-x 1 root wheel 32 Mar 23 06:21 /usr/ports ->
../a8v/pub/OpenBSD/current/ports
v1:tmp/rd 131>mount | grep /a8v/pub
a8v:/pub on /a8v/pub type nfs (nodev, nosuid, read-only, v3, udp,
timeo=100, retrans=101)
Let me understand this:
The server (a8v) exports /pub and you mount it as /a8v/pub
/usr/ports is a symlink to ../a8v/pub/OpenBSD/current/ports;
i.e., a symlink to /usr/a8v/pub/OpenBSD/current/ports. Right?
No, it's a (relative) symlink to /a8v/pub/OpenBSD/current/ports.
Is this inteded? Why?
Yes, it's intended. I'm not sure what you mean by why?
As a result of the change, I added the following to /etc/mk.conf:
PORTSDIR=/x2/pub/OpenBSD/current/ports
What is /x2 and what does it have to do with any of the previous?
My bad, I corrected the 'x2' mistake in [1].
Why don't you set PORTSDIR to the directory /usr/ports links to?
That's what I thought I did but forgot to set PORTSDIR in the /etc/mk.conf of
the other hosts in the dpb cluster.
I had previously removed the symlink and NFS mounted the ports
directory on /usr/ports.
I believe people do that.
That didn't work (still had the '/usr/ports is a symlink').
I don't believe this. Try it again.
Remove the symlink, mount a8v:/pub/OpenBSD/current/ports
directly to /usr/ports, and show me how it fails.
As noted in [2], the failure was in my config of the other hosts in the cluster.
But a8v is not exporting /pub/OpenBSD/current/ports,
it is only exporting /pub, right?
Yes, with -alldirs, ie.:
a8v:build/packages 502>grep '^/pub ' /etc/exports
/pub -alldirs -maproot=0:wheel -network 10.1.2 -mask 255.255.254.0
As a result, I made the patch in the
previous mail.
However, I didn't like having two mountpoints in the same NFS
hierarchy, ie. /a8v/pub and /a8v/pub/OpenBSD/current/ports. So, I
reverted to using the symlink and PORTSDIR change along with the
patch and seem to be successfully building (still in progress).
With PORTSDIR set in /etc/mk.conf, how could ${PORTSDIR} expand to
'/usr/ports' as reported in the dpb log below? How can I help
further isolate the problem?
I believe your "problem" is probably caused by your convoluted
structure of symlinks and (non)mounted subdirectories, not
by the build process of ports.
The problem was my configuration of the other hosts in the cluster.
WRT to the above 'convolution' and given that I want to use an OpenBSD server
that contains the CVS tree for ports, src and xenocara, how would you recommend
arranging the dpb cluster and nfs mounts?
Thanks.
[1]<http://marc.info/?l=openbsd-ports&m=133253773107507&w=2>
[2]<http://marc.info/?l=openbsd-ports&m=133258871920079&w=2>
