[ Moved to -current ] On Tue, Jul 29, 2003 at 07:12:09PM -0700, Peter Wemm wrote: > John Birrell wrote: [...] > > The installworld check for a kernel with the new sigaction promptly > > core-dumped sh with the unsupported syscall. I think this check should > > be based on sysctl kern.osreldate, not just running the new shell to see > > if it core dumps. Thats kind of sub-optimal because I normally associate > > core dumps during build/installworld with a dodgy build. In this case, > > the build wasn't dodgy... I just hadn't realised that I was missing a toe. > > The problem is that some of the install scripts run the host /bin/sh *after* > src/bin has been installed. If your sh in ${OBJDIR} wont run, you *WILL* > be hosed later on in the build with about a 50:50 split between old and new > worlds. I dont recall which scripts are the culprits, but it is a side > effect of "#! /bin/sh" that is on a script that is run during installworld. > > In theory, the 'make installworld' stuff copies everything needed for an > installworld to a /tmp directory, but it still misses a bunch of > "#! /bin/sh" type things. This isn't a cross build issue because the shell > scripts are run on the host. > > For what its worth, this test saved me yesterday when I was trying to update > an ancient ia64 box to -current. > That wasn't the point John was trying to make, IMO. I'm regularly upgrading 4.0-RELEASE to 5.x-CURRENT to check that the upgrade path wasn't broken, and this check comes very handy. I wouldn't object though if it was done using kern.osreldate. I'd happily test any patches in this direction for you, John. Cheers, -- Ruslan Ermilov Sysadmin and DBA, ru_at_sunbay.com Sunbay Software Ltd, ru_at_FreeBSD.org FreeBSD committer
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:37:17 UTC