[INN-COMMITTERS] STABLE-2_4 inn (configure configure.in)
rra at stanford.edu
Tue Sep 2 17:11:16 UTC 2003
Forrest J Cavalier <mibsoft at epix.net> writes:
>> To: inn-committers at isc.org
>> From: Russ Allbery <Russ_Allbery at isc.org>
>> Subject: [INN-COMMITTERS] STABLE-2_4 inn (configure configure.in)
>> Simplify the check for sendmail considerably. Just look for it on PATH
>> or in /usr/sbin or /usr/lib if not found there, without requiring
>> people to use --with-sendmail if it's in a standard location. That's
>> the way that pretty much every other package does it.
> I think this change is a very bad idea. There is a long thread of
> discussion that resulted in the configure rules in INN being what they
> The summary of that discussion is "Better safe than sorry."
Yeah, sorry... I was going to bring this up here and then forgot in the
work of trying to catch up on my e-mail.
> Sendmail is too bug-prone to trust an accidentally found version that
> someone might not even know was installed.
So, I was swayed by the initial discussion, somewhat, but over the years
this has caused nothing but complaints. The current rule (look in PATH
and then look in /usr/sbin and /usr/lib) is one of two variations that
every other package I've ever compiled uses (the other is to only look in
/usr/sbin or /usr/lib), and most of them don't even provide any way to
I'm certainly willing to be persuaded that it should just look in
/usr/sbin and /usr/lib and not look in the user's PATH (either works for
me; I think that only looking in /usr/sbin or /usr/lib is mildly more
correct and kept the PATH search just so that the behavior isn't as
different from the current behavior). But I think the current behavior is
broken, after years of watching its implications. It means that
practically everyone compiling INN has to pass an explicit --with-sendmail
flag to configure, when most people normally don't even have to think
about where sendmail is located.
Note that the *only* place this configure result is used in INN is in
inn.conf, which the user should really be reviewing anyway, and it's the
first entry in the file. So if we do guess wrong, it's pretty easy to
both find and fix. And if we're using a buggy version of sendmail, that
really doesn't matter much, since we're only using it to send mail and our
demands on it are pretty severely limited.
What does everyone else think about this? Sorry that I didn't bring it up
Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
Please send questions to the list rather than mailing me directly.
<http://www.eyrie.org/~eagle/faqs/questions.html> explains why.
More information about the inn-workers