On 27/05/2009 3:40 PM, Eric Haszlakiewicz wrote:
On Wed, May 27, 2009 at 10:34:11AM +1000, Sarton O'Brien wrote:On 27/05/2009 12:48 AM, Patrick Welche wrote:On Sat, May 23, 2009 at 02:25:08PM +0000, Andrew Doran wrote:- httpd fails because semctl() is broken, so nxr.netbsd.org is down.So no one mentioned this one. In my case it was "you upgraded netbsd so now you can't see your postgresql data". Recompiling postgresql allowed me to run a database, but not to look at my old data. I got around it with make new directory, unpack old base.tgz into it, chroot to it (so use old libc which I think was the crux), run old postgresql, pg_dumpall, load into new postgresql.With apache there was some libtool/apr related brokenness as well. Not sure which one specifically.I think the brokenness with apr that you are referring to is that if you update only apache you get a message: Undefined PLT symbol "apr_socket_accept_filter" and it is difficult to "upgrade" apr, because the version number of the package isn't actually any different, even though it got built differently. Then again, that is what I am seeing when upgrading from netbsd-4 to netbsd-5 so maybe the netbsd-5 -> netbsd-current problem is something different.
Yeah it's definitely different. It's not related to a stale dependency but to a kernel/libc change. The symptoms are warnings that you should increase sysctls, I can't remember the error exactly.
I rarely encounter what you have mentioned so long as I stick with pkg_rolling-replace -suv. When I know things are safe I'll drop the 's'. :)
Every now and then when a major change occurs I'll pkg_chk -g and pkg_delete -ff '*-*' just to be safe and compile it all again ... but as Patrick mentioned, there have been changes (time_t anyone?) that have made data unreachable without back-pedaling. We were used to BDB changes being haphazard but postgresql caught us by surprise. I always keep one domu out sync for such an occasion :)
Sarton