Subject: Re: README: sendmail changes
To: Martin Husemann <martin@duskware.de>
From: Andrew Brown <atatat@atatdot.net>
List: current-users
Date: 07/22/2004 00:20:52
On Wed, Jul 21, 2004 at 11:04:23PM +0200, Martin Husemann wrote:
>Let me answer various messages from this thread in one reply:
>
>I wrote:
>
>>>I have no problem with the default setup (via sysinst, or default "empty"
>>>rc.conf file, or however) to have a 
>>>
>>>  sendmail=UNCONFIGURED
>>>
>>>and the rc.d script to interpret that as "local and outgoing mail only,
>>>via sendmail". That's the same as our rc_configured=NO default.
>
>Andrew Brown wrote:
>> yeah...that's pretty much what it does, except that since the rc.subr
>> system only deals in yes and no values, the UNCONFIGURED setting gets
>> changed to yes by rc.d/sendmail so that everything's happy.  when
>> appropriate, of course.
>
>I think I don't understand this part. If all this magic would be triggered
>by a special value in rc.conf, I would not complain at all. I don't see how
>we are forced to only use "yes" or "no" here.

you are forced to use only "yes" or "no" here by the rc.subr routine
called checkyesno.  it actually understands "true"/"false",
"on"/"off", and "1"/"0", but they're semantically the same as "yes"
and "no".

handling of a mysterious "third" setting is not trivial, so the
"default" is left as "no" for other consumers of the sendmail setting,
but the sendmail setting is actually ternary (and not binary) from the
point of view of rc.d/sendmail.

am i making any sense?

>I wrote:
>>>But the current state does not allow me to stop sendmail from running,
>>>even by an explicit
>>>
>>>  sendmail=NO
>>>
>>>I have removed /etc/rc.d/sendmail on my machines as a short term fix.
>
>and I was wrong (judging only from the commit message). It actually works
>and it is what I'm going to do now.
>
>This, of course, makes the whole change acceptable. All remaining differences
>are cosmetics.
>
>Havard Eidnes wrote:
>> Well, I guess the problem was that you didn't tell the system
>> what you wanted, and the default changed.
>
>And this is true. I can live with the default behaviour change, now that
>I've learned how to deal with it correctly.

indeed.  cvs update is the option i offer you.  :)

>I presume there will be a very prominent warning in the update part of the
>release docs.

tell me where to put it and it shall be so.

>To summarize: I fully agree with the "local and standard outgoing mail
>must work out of the box" paradigma. I think the sendmail setup that is
>invoked by a default install now is sane.

good!

>But I totally dislike the way that the "sendmail" variable in the rc.conf
>system is calculated. IMHO it should default to "NO" and stay that way,
>unless the admin set it to "YES" or the default install set it to 
>"AUTOMAGIC" (or something to that effect), which explicitly calls for
>conjuring spells that "just make it work".

"AUTOMAGIC" is the third setting that we can't use in the general
case, unfortunately.  for myself, i'd like to see a smaller piece of
more powerful "magic" popped into rc.subr to handle this more cleanly
(imho)...

>Having learend that at least the admin telling it explicitly "no magic,
>please" will do what I want, I can live with the stuff as is - but I still
>don't like it.

you are not alone.

-- 
|-----< "CODE WARRIOR" >-----|
codewarrior@daemon.org             * "ah!  i see you have the internet
twofsonet@graffiti.com (Andrew Brown)                that goes *ping*!"
werdna@squooshy.com       * "information is power -- share the wealth."