Re: WITHOUT_PF breaks buildworld

On Wed, Dec 22, 2021, 10:02 AM Gary Jennejohn <gljennjohn@gmail.com> wrote:

> On Wed, 22 Dec 2021 16:51:54 +0100
> Gary Jennejohn <gljennjohn@gmail.com> wrote:
>
> On Wed, 22 Dec 2021 16:27:52 +0100
> Kristof Provost <kp@FreeBSD.org> wrote:
>
> > On 22 Dec 2021, at 8:21, Konrad Sewi??o-Jopek wrote:
> > > Hi,
> > >
> > > I think the reason is somewhere in tools/build/test-includes:
> > >
> > > --- net/if_pfsync.o ---
> > > In file included from net/if_pfsync.c:1:
> > > In file included from
> > > [...]freebsd/arm64.aarch64/tmp/usr/include/net/if_pfsync.h:56:
> > > [...]freebsd/arm64.aarch64/tmp/usr/include/net/pfvar.h:65:10: fatal
> error:
> > > 'netpfil/pf/pf.h' file not found
> > > #include <netpfil/pf/pf.h>
> > >          ^~~~~~~~~~~~~~~~~
> > > 1 error generated.
> > > *** [net/if_pfsync.o] Error code 1
> > >
> > > make[3]: stopped in [...]freebsd/tools/build/test-includes
> > > --- net/pfvar.o ---
> > > > In file included from net/pfvar.c:1:
> > > > [...]freebsd/arm64.aarch64/tmp/usr/include/net/pfvar.h:65:10: fatal
> error:
> > > > 'netpfil/pf/pf.h' file not found
> > > > #include <netpfil/pf/pf.h>
> > > >          ^~~~~~~~~~~~~~~~~
> > > > 1 error generated.
> > > > *** [net/pfvar.o] Error code 1
> > > >
> > > > make[3]: stopped in [...]freebsd/tools/build/test-includes
> > > > 2 errors
> > > >
> > > > make[3]: stopped in [...]freebsd/tools/build/test-includes
> > > > *** [test-includes] Error code 2
> > > >
> > > > make[2]: stopped in [...]freebsd
> > > > 1 error
> > > >
> > > > Best regards,
> > > > Konrad Sewi??o-Jopek
> > > >
> > > >
> > > > niedz., 19 gru 2021 o 12:26 Gary Jennejohn <gljennjohn@gmail.com>
> > > > napisa?(a):
> > > >
> > > >> On Sun, 19 Dec 2021 19:05:35 +0800
> > > >> Alastair Hogge <agh@riseup.net> wrote:
> > > >>
> > > >>> On Sunday, 19 December 2021 6:47:23 PM AWST Gary Jennejohn wrote:
>
> > > >>>> Some recent change, probably in a .mk file, breaks builworld on
> HEAD
> > > >>>> when WITHOUT_PF is enabled in src.conf.
> > > >>>
> > > >>> I have had to disable WITHOUT_PF since 2020-07-27, but probably
> earlier.
> > > >>>
> > > >>
> > > >> Hmm.  I did a successful buildworld a few days ago with WITHOUT_PF
> > > >> enabled, so it's new breakge for me at least.
> > > >>
> > > >> I don't enable pf in the kernel and don't need it in userland.
> > > >>
> > > >>>> Disabling WITHOUT_PF results in a successful buildworld.
> > > >>>>
> > > >>>> The reported error is that netpfil/pf/pf.h can't be found.
> > > >>>
> > > >>> Some ports depend on that too.
> > > >>>
> > > >>
> > This is the test-includes target, which validates that include files are
> self-contained (that is, you can '#include <$file>' without prerequisites.
> > The target fails because it looks at all headers in /usr/src/sys and
> then tries to build them, but some of those headers (like the pf headers)
> include other headers that may not be getting installed because they're
> disabled.
> >
> > I'm not quite sure how to best fix this.
> >
> > Note that it is not happening because some pf tools are still getting
> built. This is a validation target that fails.
> >
> > We could potentially add the pf headers to BADHDRS depending on the
> WITHOUT_ flag, but that would mean manually maintaining badfiles.inc.
> > Or perhaps we should keep installing the pf headers even when WITHOUT_PF
> is set, but I'm not actually sure how we convince the build system to do
> that. Or if it's a good idea.
> >
> > Warner might have better ideas on how to fix this.
> >
>
> This simple patch to /usr/src/include/Makefile fixes it for me:
>
> --- Makefile.orig       2021-12-22 13:37:02.817745000 +0100
> +++ Makefile    2021-12-22 13:37:12.177336000 +0100
> @@ -246,9 +246,7 @@
>  INCSGROUPS+=   IPFILTER
>  .endif
>
> -.if ${MK_PF} != "no"
>  INCSGROUPS+=   PF
> -.endif
>
>  .if ${MK_CDDL} != "no"
>  INCSGROUPS+=   NVPAIR
>
> Since pf.h is used so widely in the tree it's probably the simplest fix.
>

I like this. I'd unconditionally remove the test in preference to adding it
conditionally.

Warner