Re: [Qemu-devel] [PATCH v2 2/2] build-sys: do not compile net/vhost-user.c if vhost-user is disabled

[Marc-André Lureau]

----- Original Message -----
> On Tue, Aug 01, 2017 at 12:15:16PM -0400, Marc-André Lureau wrote:
> > Hi
> > 
> > ----- Original Message -----
> > > On Fri, Jul 28, 2017 at 04:13:09PM +0200, Marc-André Lureau wrote:
> > > > This adds two extra #ifdef that have fairly limited conflict potential.
> > > > 
> > > > Signed-off-by: Marc-André Lureau <address@hidden>
> > > 
> > > OK but why can't we keep all ifdefs in net/ where all the
> > > rest of the features are ifdef'ed?
> > > 
> > 
> > We would need fallback code for vhost_user_get_* functions (see below).
> 
> Not if you merely add ifdefs around name arrays in net/net.c
> I think.

vhost_user_get_* are defined in net/vhost-user.c, linked from hw/net/vhost_net.c

> 
> > > My concern is no one is going to test this weird configuration
> > > so we might accidentally re-enable it without noticing.
> > > net/net.c is where all command line play happens so
> > > people know a bunch of configs need to be tested when
> > > changing it.
> > 
> > I am not sure I understand your argument there. You are afraid we
> > 'accidentaly re-enable' vhost-user... and forget to remove those #ifdef?
> > 
> > I can imagine people don't like having code clutter with #ifdef, but I
> > think when it doesn't change much the logic (as in here with clean
> > per-case switch/if), it is fine, ymmv.
> 
> Let's assume all this code is refactored, we'd have to test
> with ifdefs.

This is true whether there is one or ten #ifdef, isn't it?

It only gets more complicated the more #ifdef there is. But here it's fairly 
limited impact since it's just 2 more #ifdef for the proposed CONFIG_VHOST_USER 
from previous patch.

> 
> > > 
> > > 
> > > > ---
> > > >  hw/net/vhost_net.c | 4 ++++
> > > >  net/Makefile.objs  | 2 +-
> > > >  2 files changed, 5 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/hw/net/vhost_net.c b/hw/net/vhost_net.c
> > > > index e037db63a3..96c4da49e4 100644
> > > > --- a/hw/net/vhost_net.c
> > > > +++ b/hw/net/vhost_net.c
> > > > @@ -193,6 +193,7 @@ struct vhost_net *vhost_net_init(VhostNetOptions
> > > > *options)
> > > >          }
> > > >      }
> > > >  
> > > > +#ifdef CONFIG_VHOST_USER
> > > >      /* Set sane init value. Override when guest acks. */
> > > >      if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
> > > >          features = vhost_user_get_acked_features(net->nc);
> > > > @@ -203,6 +204,7 @@ struct vhost_net *vhost_net_init(VhostNetOptions
> > > > *options)
> > > >              goto fail;
> > > >          }
> > > >      }
> > > > +#endif
> > > >  
> > > >      vhost_net_ack_features(net, features);
> > > >  
> > > > @@ -414,10 +416,12 @@ VHostNetState *get_vhost_net(NetClientState *nc)
> > > >      case NET_CLIENT_DRIVER_TAP:
> > > >          vhost_net = tap_get_vhost_net(nc);
> > > >          break;
> > > > +#ifdef CONFIG_VHOST_USER
> > > >      case NET_CLIENT_DRIVER_VHOST_USER:
> > > >          vhost_net = vhost_user_get_vhost_net(nc);
> > > >          assert(vhost_net);
> > > >          break;
> > > > +#endif
> > > >      default:
> > > >          break;
> > > >      }
> > > > diff --git a/net/Makefile.objs b/net/Makefile.objs
> > > > index 67ba5e26fb..7cac7ed1e4 100644
> > > > --- a/net/Makefile.objs
> > > > +++ b/net/Makefile.objs
> > > > @@ -3,7 +3,7 @@ common-obj-y += socket.o
> > > >  common-obj-y += dump.o
> > > >  common-obj-y += eth.o
> > > >  common-obj-$(CONFIG_L2TPV3) += l2tpv3.o
> > > > -common-obj-$(CONFIG_POSIX) += vhost-user.o
> > > > +common-obj-$(CONFIG_VHOST_USER) += vhost-user.o
> > 
> > So we could have something like:
> > 
> > +common-obj-$(not CONFIG_VHOST_USER) += vhost-user-stubs.o
> > 
> > ?
> > 
> > > >  common-obj-$(CONFIG_SLIRP) += slirp.o
> > > >  common-obj-$(CONFIG_VDE) += vde.o
> > > >  common-obj-$(CONFIG_NETMAP) += netmap.o
> > > > --
> > > > 2.14.0.rc0.1.g40ca67566
> > > 
>