building releng11 on RELENG12 broken ?

[Kyle Evans]

On Mon, Oct 5, 2020 at 9:48 AM Kyle Evans <kevans at freebsd.org> wrote:
>
> On Mon, Oct 5, 2020 at 5:13 AM mike tancsa <mike at sentex.net> wrote:
> >
> > On 10/4/2020 8:21 PM, Kyle Evans wrote:
> > > On Sun, Oct 4, 2020 at 7:02 PM mike tancsa <mike at sentex.net> wrote:
> > >> Not sure exactly when, but it seems building RELENG_11 seems to be
> > >> broken on releng12.  I am trying to buildworld on a RELENG12 image with
> > >> -j4 (12.2-STABLE r366088) and it fails with the errors below
> > >>
> > > Hi,
> > >
> > > I've been battling this one for a couple days, but I really haven't
> > > decided how to move forward. The TL;DR version is that crunchgen is
> > > getting the OBJDIR wrong when building WITHOUT_AUTO_OBJ by assuming
> > > objects will be built in-tree, but the reality is that it's more
> > > complicated than that- it may still use out-of-tree OBJDIR if some
> > > circumstances are hit.
> > >
> > > This is reproducible on -HEAD and stable/12, too, but it's effectively
> > > mitigated by the fact that those branches default to WITH_AUTO_OBJ.
> > >
> > > CC'ing Bryan, because he probably has a better idea of how to fix it.
> > > I almost smacked it with a "shell out for each utility and just ask
> > > what the OBJDIR is" hammer, but I don't want to get smacked. =)
> >
> > Thanks, is it merely defining
> >
> > WITH_AUTO_OBJ
> >
> > to allow things to build ? Would that have just build consequences or
> > would it cause installation issues too ? We crossbuild on a releng12 box
> > and then mount over nfs to do an installworld on some servers.
> >
>
> Yes, I suspect this is sufficient -- you'd need to make sure the
> rescue build is clean, so either a not-no-clean build or make -C
> rescue/rescue clean prior so that it rebuilds the rescue.mk inside
> your .OBJDIR (probably better to just do a full clean build). Note
> that I don't know what kind of behavior may be present that leaves it
> off-by-default in 11; Bryan put a lot of work into the build system
> between 11 and 12.
>

Another report came to me privately that had tracked it down in the
other direction -- building -CURRENT from an earlier 12 -- that
pointed directly at a bug in crunchgen on -CURRENT. I've committed the
fix in r366466 and will MFC in 3ish days.

Thanks,

Kyle Evans