Replacing Sendmail with Postfix in the base system

[Claus Assmann]

On Wed, Jun 14, 2006, Matthew Dillon wrote:

> :Which functionality is missing? Usually sm8 is considered the MTA
> :with the most functionality.
> 
>     For example, local user feed-through aliases using '.'.  For example,

Sorry, I don't understand what you mean here. Please explain.

>     all the stupidity with turning simple control files into DBM databases.

You can use any map type you like, even "text". What kind of "control"
do you prefer? Most MTAs use some kind of "maps" to provide controls
that depend on some key (client IP, e-mail address, ...).

>     All the crazyness about whether the strings in the DBM files are case
>     sensitive, or case insensitive, or include the terminating \0 or not.

What does this have to do with "missing functionality"? Check postfix:
Wietse does the same to deal with the various map types. Case sensitive/
insensitive is a feature that has been requested by users.

> :What did the authors refuse?
> 
>     The CF file structure is a ridiculous, unreadable mess.  As long as you

sendmail X fixes this. Some years ago we made the decision that it
is simpler to start over.

> :>     added threading without dealing with races.  They added all sorts of

> :sendmail 8 has threads only in libmilter. Where are the race
> :conditions?
> 
>     Where?  If I knew that there wouldn't be any now would there!   I

Hold on a second. You say that we "added threading without dealing
with races." So you seem to know that there are races or at least
we don't "deal" with them. Please point out where we don't deal
with them (I wrote the libmilter code, so if there's something
wrong with it, I want to know it and fix it).

>     have to use libmilter.  It's a godaweful mess.  For example, if my spam
>     filter crashes for whatever the entire sendmail system hangs.   If I

"hangs"? How? Look for F=t for a milter definition.

>     start things in the wrong order (milter process vs sendmail), the entire
>     sendmail system hangs.  There is NO robustness.  I can't even restart
>     the milter filter without sendmail going nuts.  I have to kill EVERYTHING
>     and restart EVERYTHING.

It seems you have a wierd configuration: sendmail 8 connects to a milter
for every incoming connection. Hence you do NOT need to restart the daemon.
If the milter doesn't respond within the configured timeout, then
a configurable action (tempfail/ignore/...) is taken.

>     In anycase, the NetBSD move was probably due to a new sendmail issue 
>     currently being circulated by CERT which will be announced soon.  It
>     is mainly due to yet more badly written code.  I think they finally got

As I wrote: I understand your frustration and I agree with most of
your statements.  I know that there are many problems in sm8 (and
that's why there is smX). But I neither see "race conditions" in
libmilter, nor "sendmail 8 hangs when a milter doesn't respond".
If there are problems in the parts that I wrote, I want to know
what those problems are.