Hi,

On Tue, Nov 21, 2017 at 01:46:52AM +0100, Roland Hieber wrote:
> [...]
> 
> > diff --git a/rules/nginx.make b/rules/nginx.make
> > index ef98c0adb..0f8ce65f2 100644
> > --- a/rules/nginx.make
> > +++ b/rules/nginx.make
> > @@ -17,8 +17,8 @@ PACKAGES-$(PTXCONF_NGINX) += nginx
> >  #
> >  # Paths and names
> >  #
> > -NGINX_VERSION      := 1.10.3
> > -NGINX_MD5  := 204a20cb4f0b0c9db746c630d89ff4ea
> > +NGINX_VERSION      := 1.12.1
> > +NGINX_MD5  := a307e74aca95403e5ee00f153807ce58
> >  NGINX              := nginx-$(NGINX_VERSION)
> >  NGINX_SUFFIX       := tar.gz
> >  NGINX_URL  := https://nginx.org/download/$(NGINX).$(NGINX_SUFFIX)
> > @@ -44,7 +44,8 @@ NGINX_CONF_ENV := \
> >     ngx_force_have_map_anon=yes \
> >     ngx_force_have_map_devzero=yes \
> >     ngx_force_have_sysvshm=yes \
> > -   ngx_force_have_posix_sem=yes
> > +   ngx_force_have_posix_sem=yes \
> 
> I cannot get IPv6 to work, because nginx complains about "the INET6
> sockets are not supported on this platform" when I try to have it listen
> on [::]:443. This happens because NGX_HAVE_INET6 apparently gets defined
> to "yes" in objs/ngx_auto_config.h, and it works well if I change this line:
> 
> > +   ngx_force_ipv6=$(call ptx/ifdef, PTXCONF_GLOBAL_IPV6, yes, no)
> 
> to
>       ngx_force_ipv6=$(call ptx/ifdef, PTXCONF_GLOBAL_IPV6, 1, 0)
> 
> Can you reproduce this?

Yes, my mistake. I assumed, they use #ifdef NGX_HAVE_INET6 and not
#if (NGX_HAVE_INET6)..
So NGX_HAVE_INET6 is defined to yes in ngx_autoconfig.h, which evaluates
to 0.

It's a little bit misleading because for ngx_feature_run != value, "yes"
is a possible value, so all other ngx_force_foo can be set to yes and
no.

Now we could just use 1 and 0 for ngx_force_ipv6. But maybe it would be
preferable to translate yes and no into 1 and 0 respectively in case of
ngx_feature_run = value (in auto/feature) to be consistent with other
passed ngx_force_ environment variables?

Cheers,
Clemens

_______________________________________________
ptxdist mailing list
ptxdist@pengutronix.de

Reply via email to