On Mon, Mar 18, 2013 at 09:45:57AM -0400, John Baldwin wrote: > On Sunday, March 17, 2013 10:49:22 pm Jeremy Chadwick wrote: > > Please see PR 177055. This MFC has broken mergmaster for at least a > > couple people. The PR should really be rated high priority:
Sorry I missed this earlier. I've backed out the change for now. Ironically, it only effects people who follow the instructions about mergemaster -p. If you get away with skipping that step then it just works. > > http://www.freebsd.org/cgi/query-pr.cgi?pr=177055 > > > > http://lists.freebsd.org/pipermail/freebsd-stable/2013-March/072848.html > > Probably breaks etcupdate as well. Both of these tools use the host's tools > and not the cross-built tools as they just run 'make distribute'. The right > fix might be as simple as making those make targets use XMAKE so that if a > world is built they use the cross-built tools instead of the host's tools? I was wondering why they sets MAKEOBJDIRPREFIX at all. If they don't set it, then they would use the bootstrap tools that were already created. Regardless of the final fix in mergemaster and etcupdate, we also need to change the instructions in UPDATING. Users need to run "sh usr.sbin/mergemaster/mergemaster.sh -p" as that's the only way we can reliably survive requirement changes in the source tree. Does this seem like a generally acceptable change? -- Brooks P.S. In addition to requiring install -l, I'm planning to change HEAD to require nmtree in the not to distant future so that we can eliminate the last duplicate entries in METALOG by only listing each directly once in a .mtree file so it would be good to get this sorted out soon.
pgpgzIPjqgPNj.pgp
Description: PGP signature
