Re: [PATCH v3] conf: Accept duplicate and conflicting options, the last one wins

[Stefano Brivio <sbrivio@redhat.com>]

On Fri, 21 Jun 2024 11:52:40 +0200
Paul Holzinger <pholzing@redhat.com> wrote:

> On 20/06/2024 19:19, Stefano Brivio wrote:
> > In multiple occasions, especially when passt(1) and pasta(1) are used
> > in integrations such as the one with Podman, the ability to override
> > earlier options on the command line with later one would have been
> > convenient.
> >
> > Recently, to debug a number of issues happening with Podman, I would
> > have liked to ask users to share a debug log by passing --debug as
> > additional option, but pasta refuses --quiet (always passed by Podman)
> > and --debug at the same time.
> >
> > On top of this, Podman lets users specify other pasta options in its
> > containers.conf(5) file, as well as on the command line.
> >
> > The options from the configuration files are appended together with
> > the ones from the command line, which makes it impossible for users to
> > override options from the configuration file, if duplicated options
> > are refused, unless Podman takes care of sorting them, which is
> > clearly not sustainable.
> >
> > For --debug and --trace, somebody took care of this on Podman side at:
> >    https://github.com/containers/common/pull/2052
> >
> > but this doesn't fix the issue with other options, and we'll have
> > anyway older versions of Podman around, too.
> >
> > I think there's some value in telling users about duplicated or
> > conflicting options, because that might reveal issues in integrations
> > or accidental misconfigurations, but by now I'm fairly convinced that
> > the downsides outweigh this.
> >
> > Drop checks about duplicate options and mutually exclusive ones. In
> > some cases, we need to also undo a couple of initialisations caused
> > by earlier options, but this looks like a simplification, overall.
> >
> > Notable exception: --stderr still conflicts with --log-file, because
> > users might have the expectation that they don't actually conflict.
> > But they do conflict in the existing implementation, so it's safer
> > to make sure that the users notice that.
> >
> > Suggested-by: Paul Holzinger <pholzing@redhat.com>
> > Suggested-by: David Gibson <david@gibson.dropbear.id.au>
> > Signed-off-by: Stefano Brivio <sbrivio@redhat.com>  
> 
> Tested-by: Paul Holzinger <pholzing@redhat.com>

Thanks!

> > ---
> > v3:
> >    - override --gateway and --dns-forward with the last given option
> >      for a given address family, too  
> assuming you mean --address here as --gateway already worked

...right, no idea why I wrote 'gateway'. :)

-- 
Stefano