[PATCH] resolv: Avoid duplicate query if search list contains '.' (option 2)

[Florian Weimer]

* Carlos Peon Costa:

> On Fri, Jan 9, 2026 at 11:40 AM Florian Weimer <fweimer@redhat.com> wrote:
>
>> I think the traditional assumption is that "." in the search list is
>> unsuported and should be removed.  I checked the documentation of
>> multiple implementations, and no one seems to document the intended
>> behavior of ".".
>>
>> Are you using "." to suppress search list processing?
>>
>
> Yes, I'm using "." to suppress search list processing. Not setting the
> "search" option implies its default value: the local domain name, so
> there is no trivial configuration to disable search list processing.

Thanks for confirming.

> IMO "." in the search list looks weird and I think an empty search
> list (without failing back to local domain name) should be a valid
> configuration to disable search list processing, but that would break
> default behaviour and probably is not an option.

I think the "." approach has leaked into multiple tools that generate
/etc/resolv.conf by now, so we're probably stuck with it.  For example,
systemd does this:

        if (ordered_set_isempty(domains))
                fputs("search .\n", f); /* Make sure that if the local hostname 
is chosen as fqdn this does not
                                         * imply a search domain */
        else
                write_resolv_conf_search(domains, f);

> IMO unsupporting "." in the search list and keeping supporting setting
> hostnames without dots (which both lead to the same scenario: no
> search list processing) seems unnecessary to me.

I agree.  It seems that "." mostly does what some users want.

Regarding your patch, ideally we would have a test case for it that
counts the number of queries the system sends, with various
configurations and functions.  I think having a container test with
/etc/resolv.conf in it and that "search ." line would work once we add
support for disabling the search path initialization in
resolv_test_start (new bool field in struct resolv_redirect_config
is probably the way).  Unfortunately, I won't have time to work on
this myself in the forseeable future.

Thanks,
Florian