bug#5692: rfc822-addresses returns a string on occasion

[Stephen Berman <stephen.berman@gmx.net>]

On Sun, 07 Mar 2010 18:37:16 +0100 Andreas Schwab <schwab@linux-m68k.org> wrote:

> Uday S Reddy <u.s.reddy@cs.bham.ac.uk> writes:
>
>> (rfc822-addresses "foo:;@bar")
>> "(Unparsable address -- Strange character \\@ found: \"@_^_bar\")"
>
> Thanks, I've checked in a fix.
>
> Andreas.

Your fix in effect reverts a change I submitted to deal with bug#453.  I
wrote in that thread: "I'm not even sure what TRT is here: as I wrote in
my OP, my patch returns an error string as specified by
rfc822-bad-address, but the caller bbdb-rfc822-addresses expects a list,
so it still raises a wrong-type error.  This happens when the relocated
catch sexp has scope to the end of the let that immediately encloses it
(patch (i) below).  But [if] I leave the catch in its existing position
and move rfc822-nuke-whitespace to within it (patch (ii) below), then no
error is raised, but also there's no indication of a bad address; yet
stepping through the code with edebug does pass through
rfc822-bad-address.  So where should the catch be located?  Hopefully
someone familiar (unlike me) with RFC 822 and rfc822.el, and ideally
also bbdb-rfc822-addresses, can answer that question and say whether
either of these patches is suitable, or provide a better one."  Chong
Yidong decided that giving the catch wide scope was the correct fix and
installed it.  Your patch undoes that (even moves the catch lower than
it was before my patch), making it possible for bug#453 to occur again.
But again, I don't know what the best fix is.

Steve Berman