🚀 go-pugleaf

> From: vjs@rhyolite.sgi.com (Vernon Schryver)
> 
> >   Hhmmm, why so it does.  Amazing.  Well, if you change the rule to:
> > 
> > 	R$+\ at\ $+	$1@$2
> > 
> > (at least on sendmail 4.12), it starts working correctly.
> 
> I tried this, I think, but can't make it work.  With the backslashes, the
> rule does not seem to match 'joe at foobar' at all.  (I'm assuming
> everything we're talking about is a blank, not a tab.)  What am I missing?
> 
> SGI's sendmail is currently 5.52 with pathalias hacks similar to IDA, but
> done earlier or at least independently.

  Uh sorry, I should have looked before I leapt.  I just tested the fact
that it would no longer match on ``user@foo.at.bar''.  I forgot to test
``user at foo''.  Very sloppy of me.

  Unfortunately, now that I've been forced to look at it harder than is
required for a flip answer, I can see what the problem is.  Rules are
read by the same code which reads addresses.  Hence the interpretation of
spaces (per RFC822).

	[RFC822, section 3.4.2, WHITE SPACE:

		Note:	In structured field bodies, multiple linear space
			ASCII characters (namely HTABs and SPACEs) are
			treated as single spaces and may freely surround
			any symbol. ...]

  Even if you do get the spaces into the rule properly by quoting them
(again as per RFC822 now that we know how sendmail parses rule set LHSs)
via:

 	R$+" "at" "$+	$1@$2

It still won't match ``user at host'' because the address string's spaces
aren't quoted!  That is, the address would have to come in as:

	To: user" "at" "host

And fundamentally we're screwed at this point because by the time our
rewriting rules get a crack at an address, this level of parsing has
already gone down on the input address so that given the line:

	To: user at host

all we ever see are the tokens "user", "at", and "host".  We don't get to
see the spaces.  What you'd like to do is somehow define a class, say o,
that contained all the delimiter tokens, then write the rule as:

 	R$+$~oat$~o$+	$1@$2

but, unfortunately, sendmail's class matching code is just a little weak,
so this doesn't work.  (Not even Sun's sendmail handle's this bit of class
matching and Sun's sendmail through Bill Nowicki's work has a lot of
fixes to the class matching code.)

  The ``user at host'' translation could probably be done through a
complicated series of tokenizations and retokenization, but I wouldn't
want to try to come up with it, or maintain it, or even look at it.  The
solution is probably just to remove the rule and let gateways that
absolutely have to do it.  Better yet, have everyone remove it and force
the offending sites generating the addresses to ditch the form.

Casey