Re: Opportunistic STARTTLS in smtpmail.el

[Ted Zlatanov]

On Tue, 03 May 2011 03:01:40 +0200 Lars Magne Ingebrigtsen <address@hidden> 
wrote: 

LMI> Ted Zlatanov <address@hidden> writes:
>> If you insist on avoiding this file check, we could have a "Lisp data"
>> backend in addition to the file and Secrets backends.  That would be
>> pretty trivial to implement and would mirror the existing netrc parse
>> results structurally.

LMI> I think the idea of putting this stuff in auth-source is good, but I'm
LMI> just wondering whether we could have a meaningful separation of secret
LMI> credentials (i.e., passwords and user names) and non-secret credentials
LMI> (key files to be used, in this instance).

LMI> I think putting stuff like key files into ~/.authinfo is fine.  But if
LMI> the user has a ~/.authinfo.gpg file (for IMAP use, for instance), and
LMI> smtpmail wants to see whether a key file is to be used, there should be
LMI> a way for smtpmail to get at this information without typing the .gpg
LMI> password.

LMI> I don't really see how that would work in any convenient way with our
LMI> current interfaces.  smtpmail really wants to say "check whether this
LMI> exists, but don't try too hard", sort of.

What's wrong with a pure data backend?  It would be specified in
`auth-sources', basically an inlined netrc parse which could be
generated by an assistant.el walk-through!

Any non-secret data can be specified inline.  Then we won't need any
file splitting, though the users can choose to put non-secret data in
a file just the same.

For example:

(setq auth-sources '((:source (:user tzz :keyfile "mykeyfile" :host "myhost" 
:port 587))
                     "~/.authinfo.gpg"))

I think that's cleaner since the inlined data maps nicely to the netrc format.

Ted