[TUHS] man-page style

Eric Allman tuhs at eric.allman.name
Wed Nov 28 10:10:43 AEST 2018


[I sent this almost a week ago, but it never showed up, probably because
my From address didn't match my subscription address.  Apologies if this
is a dup.]

I confirm Jon's observation.  It's true, sendmail wasn't "designed" in
the waterfall model sense, because at the time the email world was a
disaster, with new networks appearing seemingly daily, each of which
seemed to feel a need to come up with a new syntax for addresses.  Some
of those syntaxes were left-associative and some right-associative, and
there was no "correct" answer --- different sites wanted to parse the
same email address differently.  For example, consider:

	decvax!research!foo at berkeley

If you're on decvax you should send this to research.  If you're at
Berkeley you should send this to decvax.  And if you're on some other
site with an ARPAnet connection but without UUCP, you should send it to
Berkeley.  I concluded that something arbitrarily flexible (which meant
Turing Complete) was necessary.  Also, since some sites didn't
understand things like "@" signs in addresses, it was necessary to
rewrite the header so "reply" would work.  And all of this had to fit
into a 16-bit address space.

Life is easier today: the world has agreed on "user at domain", the
multitude of networks (e.g., UUCP, DECnet, CSNet, Berknet, ChaosNet,
PurdueNet, and possibly my favorite, a network out of the UK that used
"user at domain" but with the domain reversed, e.g., eric at edu.berkeley.cs
instead of eric at cs.berkeley.edu) is down to effectively one, and the
only 16-bit machines out there are not-very-powerful microcontrollers.
If I were building sendmail in this very different world, it would look
very different.

eric


On 2018-11-19 08:48 , Jon Steinhart wrote:
> Warner Losh writes:
>> On Fri, Nov 16, 2018 at 2:40 PM Jon Steinhart <jon at fourwinds.com> wrote:
>>
>>> Emacs sort of
>>> violates my UNIX-sense as it does many things instead of doing one thing
>>> well.
>>
>>
>> I'd argue that's not a bad thing. When people tried to add macros to make
>> or sendmail, you wound up with crazy like imake or the crazy sendfile.m4
>> stuff. Of course, sendmail and one thing aren't mates, but sometimes you
>> need to do a few, well chosen things well to avoid the crazy that trying to
>> misuse something will bring to the table.
>>
>> Warner
> 
> Funny that you bring this up as I was just talking to Eric about this.
> I was telling him that someone had recently asked me why sendmail was
> so complicated, and I explained to them that it was because email wasn't
> always like it is today; that there were many disparate email systems and
> sendmail glued them all together.  Eric said something like yeah, and I
> would have liked to have a better syntax but memory was too constrained
> at the time to let me do anything better.
> 



More information about the TUHS mailing list