Re: .authinfo support broken in Gnus 5.8.5?

[Chris Brierley <brierley@pobox.com>]

On 2000-04-24, Shenghuo ZHU <zsh@cs.rochester.edu> wrote:

> Fixed in the latest CVS version. Please test.

Well, it is fixed, but the behavior is still a little weird, IMO.
How do I explain this.

It seems to me that the purpose of the change to
gnus-netrc-machine to add a port parameter was to allow multiple
entries in .authinfo for the same machine but a different servers
(i.e. nntp or imap).

If you *do not* specify a port when you call gnus-netrc-machine it
will prefer the first entry from .authinfo which matches the
machine and *either* has no port specified or port nntp specified.
That is, regardless of whether there is no port or "port nntp",
the first machine that matches in the file will be returned.  This
works fine if you specify your NNTP server with "port nntp" and
your IMAP server with no port information.

However, if you specify your IMAP server in .authinfo *with* "port
imap" (in fact, if you specify any port info other than nntp)
gnus-netrc-machine called without a port parameter will never find
it.  It will prefer an entry with no port information, or port
nntp, and if it can't find those it will return nil.

This is all probably fine in terms of the way this new
functionality is intended for use.  The problem here is that
nnimap.el does not currently call gnus-netrc-machine with a port
parameter.  So, if your .authinfo looks something like this:

,-----
| machine personal login foobar port imap
| machine personal login wubba port nntp
`-----

... the entry for IMAP will never be used in the current version
of Gnus.

This is all a bit funky.  Perhaps the best fix is to change
nnimap.el to specify a port when it uses gnus-netrc-machine.
However, that would *require* users to specify port information
for any server other than nntp, as an entry with no port
information will not be returned by:

  (gnus-netrc-machine list machine "imap")

Maybe it would be better for the function to only bring port
information into play *only* if there is more than one entry which
matches the machine.  Backends would be required to call
gnus-netrc-machine with port information in case there was a
collision in the file, but users wouldn't be required to specify
the port information unless they had two entries for the same
host.

Make sense?