svn commit: r351423 - in head: . sbin/ping6 sbin/ping6/tests

[Jan Sucan]

On 26. 8. 2019 1:58, Alan Somers wrote:
> Jan (please keep him CCed on replies) has been musing about the same
> thing.  That might satisfy everyone.  Jan, would it be straightforward
> to implement?
> -Alan
I forgot to answer the question whether it would be straightforward. It 
depends on how fast it should be solved. Simple and quick solution would 
be to completely duplicate the getopt loop including parsing of 
arguments of the options. With some more time I would implement it in 
two commits. The first commit for separating parsing of command line 
tokens from parsing of the option arguments, the second commit for 
translation of the options.

I'm going to do it the second way. If the time is important, let me know.

-Jan
> On Sun, Aug 25, 2019 at 5:51 PM Conrad Meyer <cem at freebsd.org> wrote:
>> Hi Alan, Hiroki,
>>
>> It would be pretty easy to install a `ping6` link to the `ping(8)`
>> binary with different option parsing (conditional on argv[0]).  That
>> removes most of the issues of code and space duplication, I think?
>> And the goal would be for the 'ping6' name to retain option
>> compatibility with historical ping6.
>>
>> It's not an uncommon pattern; for example, 'id', 'groups', and
>> 'whoami' are all a single binary with multiple linked names.  Another
>> example is Clang, which provides 'cc', 'c++', 'clang', 'clang-cpp',
>> 'clang++' and 'cpp' links to the same inode — and those have very
>> different behavior depending on argv[0].
>>
>> It's less work than forcing the ping6 compatibility crowd to create a
>> port and doesn't hurt ping(8) much, AFAICT.  Is it an acceptable
>> middle ground?
>>
>> Best,
>> Conrad
>>
>> On Sun, Aug 25, 2019 at 1:26 PM alan somers <asomers at gmail.com> wrote:
>>> On Sun, Aug 25, 2019, 2:11 PM Hiroki Sato <hrs at allbsd.org> wrote:
>>>> Alan Somers <asomers at freebsd.org> wrote
>>>>    in <CAOtMX2hLxx=SKvh1ZoiMAcagQJjPaRSvkML9J+BgpQsz5uNNbw at mail.gmail.com>:
>>>>
>>>> as> On Sun, Aug 25, 2019 at 1:22 PM Hiroki Sato <hrs at allbsd.org> wrote:
>>>> as> >
>>>> as> > Hi,
>>>> as> >
>>>> as> > Alan Somers <asomers at FreeBSD.org> wrote
>>>> as> >   in <201908231522.x7NFMLuJ068037 at repo.freebsd.org>:
>>>> as> >
>>>> as> > as> Author: asomers
>>>> as> > as> Date: Fri Aug 23 15:22:20 2019
>>>> as> > as> New Revision: 351423
>>>> as> > as> URL: https://svnweb.freebsd.org/changeset/base/351423
>>>> as> > as>
>>>> as> > as> Log:
>>>> as> > as>   ping6: Rename options for better consistency with ping
>>>> as> > as>
>>>> as> > as>   Now equivalent options have the same flags, and nonequivalent options have
>>>> as> > as>   different flags.  This is a prelude to merging the two commands.
>>>> as> > as>
>>>> as> > as>   Submitted by:     Ján Sučan <sucanjan at gmail.com>
>>>> as> > as>   MFC:              Never
>>>> as> > as>   Sponsored by:     Google LLC (Google Summer of Code 2019)
>>>> as> > as>   Differential Revision:    https://reviews.freebsd.org/D21345
>>>> as> >
>>>> as> >  I have an objection on renaming the existing option flags in ping6(8)
>>>> as> >  for compatibility with ping(8).
>>>> as> >
>>>> as> >  Is it sufficient to add INET6 support to ping(8) with consistent
>>>> as> >  flags and keep CLI of ping6(8) backward compatible?  People have used
>>>> as> >  ping6(8) for >15 years, so it is too late to rename the flags.  I do
>>>> as> >  not think the renaming is useful if "ping -6 localhost" or "ping ::1"
>>>> as> >  works.
>>>> as> >
>>>> as> > -- Hiroki
>>>> as>
>>>> as> If ping works with inet6, then why would we want to keep a separate
>>>> as> tool around?  If it's just for the sake of people who don't want to or
>>>> as> can't update scripts, would a version in ports suffice?
>>>>
>>>>   Because removing (or renaming) it causes a POLA violation.  Do we
>>>>   really have a strong, unavoidable reason to force people to rewrite
>>>>   their script now?  This is still a fairly essential and actively used
>>>>   tool, not like rcp or rlogin.  Although deprecating ping6(8) and
>>>>   removing it from the base system in the future release at some point
>>>>   may work, changing the existing interface will simply confuse people
>>>>   who have used IPv6 for a long time.
>>>>
>>>>   In my understanding, the purpose to integrate ping(8) and ping6(8)
>>>>   into a single utility is to provide a consistent CLI and reduce
>>>>   duplicate code, not to break compatibility.
>>>>
>>>> -- Hiroki
>>>
>>> Those goals are incompatible. We can't provide a consistent CLI without breaking compatibility because ping and ping6 have conflicting options.  And we can't keep ping6 around while also removing duplicate code because that would be, well, duplicate code.
>>>
>>> When would be a better time than a major version bump to make a change like this?
>>>
>>> The lack of a ping6 command in freebsd 13 should serve as a pretty obvious reminder that scripts will need updating.  I think that putting a version of ping6 in ports should be a sufficient crutch for those who need it, don't you?