nsd users - Jun 2023 - NSD 4.7.0rc1 pre-release

Hi Florian,

The make depend fix was intentional, but we did not intentionally break
OpenBSD. Eventually we should probably switch to dynamically
determining dependencies(?) We'll have a look before doing the actual
release.

Thanks for letting us know.

Best,
Jeroen


On Thu, 2023-06-01 at 10:07 +0200, Florian Obser via nsd-users
wrote:
> Hi,
> 
> this might be a case of "I broke it, I get to keep both pieces."
> 
> Commit 36ae6811f8633c7df32fff40a15e4b05b328c8a9 [1] adds a built
> dependency on compat/cpuset.h to every .o file, even if that is not
> needed on the OS one tries to build.
> 
> In OpenBSD we do not carry the compat/ folder at all because we don't
> want to accidentally pick up alternative implementations over our own
> libc. Having a dependency on compat/cpuset.h breaks the build thusly:
> 
> make: don't know how to make /usr/src/usr.sbin/nsd/compat/cpuset.h
> (prerequisite of: answer.o)
> 
> HAVE_CPUSET_T guards the cpu affinity feature in nsd.c, server.c and
> util.c just fine. So if the OS does not support cpu affinity
> compat/cpuset.h is not needed at all.
> 
> Was that "make depend" change intentional or was that picked up
> because
> make depend was run on a system with cpu affinity and then committed?
> 
> I'm fine with an answer along the lines of "this is OpenBSD's
> problem". I just want to make sure.
> 
> Thanks,
> Florian
> 
> [1]
>
https://github.com/NLnetLabs/nsd/commit/36ae6811f8633c7df32fff40a15e4b05b328c8a9
>

Hi Florian,

Can you make an exception for the cpuset.h header here? I think that'd
be easiest short-term. We prefer not to drop the dependency from the
Makefile and currently lack the time to handle dependencies
automatically at build time.

The alternative is that we move it out of the compat directory, but
then we have to reorder things just for the sake of it not being in the
compat directory, which seems weird because that's exactly why the
compat directory is there(?)

Best,
Jeroen

P.S.
You may have some experience with handling dependencies automatically
in OpenBSD. If so, can you point to how OpenBSD handles that? I have
something for the new zone parser, but I'm not sure it's the
recommended way of doing things.



On Thu, 2023-06-01 at 13:58 +0200, Jeroen Koekkoek via nsd-users
wrote:
> Hi Florian,
> 
> The make depend fix was intentional, but we did not intentionally
> break
> OpenBSD. Eventually we should probably switch to dynamically
> determining dependencies(?) We'll have a look before doing the actual
> release.
> 
> Thanks for letting us know.
> 
> Best,
> Jeroen
> 
> 
> On Thu, 2023-06-01 at 10:07 +0200, Florian Obser via nsd-users wrote:
> > Hi,
> > 
> > this might be a case of "I broke it, I get to keep both
pieces."
> > 
> > Commit 36ae6811f8633c7df32fff40a15e4b05b328c8a9 [1] adds a built
> > dependency on compat/cpuset.h to every .o file, even if that is not
> > needed on the OS one tries to build.
> > 
> > In OpenBSD we do not carry the compat/ folder at all because we
> > don't
> > want to accidentally pick up alternative implementations over our
> > own
> > libc. Having a dependency on compat/cpuset.h breaks the build
> > thusly:
> > 
> > make: don't know how to make /usr/src/usr.sbin/nsd/compat/cpuset.h
> > (prerequisite of: answer.o)
> > 
> > HAVE_CPUSET_T guards the cpu affinity feature in nsd.c, server.c
> > and
> > util.c just fine. So if the OS does not support cpu affinity
> > compat/cpuset.h is not needed at all.
> > 
> > Was that "make depend" change intentional or was that picked
up
> > because
> > make depend was run on a system with cpu affinity and then
> > committed?
> > 
> > I'm fine with an answer along the lines of "this is
OpenBSD's
> > problem". I just want to make sure.
> > 
> > Thanks,
> > Florian
> > 
> > [1]
> >
https://github.com/NLnetLabs/nsd/commit/36ae6811f8633c7df32fff40a15e4b05b328c8a9
> > 
> 
> _______________________________________________
> nsd-users mailing list
> nsd-users at lists.nlnetlabs.nl
> https://lists.nlnetlabs.nl/mailman/listinfo/nsd-users