Re: auth-source-search

[Ted Zlatanov <tzz@lifelogs.com>]

(btw, you have the list name as "nognus" but that's only the name of the
current Gnus release)

On Fri, 04 Mar 2011 11:10:49 +0100 Richard Riley <rileyrg@googlemail.com> wrote: 

RR> Since this is probably such a required thing for anything needing
RR> passwords (erc, identica, gnus, git etc etc), would not a

RR> "auth-password" function make sense?

That's what `auth-source-user-or-password' did and it was not so good.
A simple API hides the underlying complexity of the multiple backends
and tokens that auth-source.el supports.  Since the consumers of the
auth-source API are other packages and not the end users, it makes sense
to keep the API flexible and powerful.  There is only one (interactive)
function in auth-source.el (`auth-source-forget-all-cached') currently
and that's intentional.

In your example with org-mode, it's better to add the auth-source query
in org-mode itself instead of asking the user to do the query.  The
complexity can then be hidden and abstracted in whatever way makes sense
for org-mode specifically.  org-mode should not require you, the user,
to do an auth-source query.

So maybe for user convenience I can provide a wrapper like
`auth-source-pick-first-secret' but I am seriously concerned that such a
wrapper will grow in complexity and "convenience" until its complexity
overwhelms its utility.  I would like some more feedback.
 
RR> I struggled for ages to get this to work ;) Its not my code per se above
RR> but the results of collaboration on #emacs where more than one person
RR> other than me was baffled by the complexity of docstring.

Most of the complexity has to do with creation and not searching.  I
will, as I said, provide examples of the common searches.

The #emacs crowd should feel free to submit bug reports or improvement
suggestions/patches.  I promise to consider them carefully, but can't
help anyone if their complaints are on an IRC channel.

RR> Your code above is even more complex from my point of view - and I see
RR> why since you detect the secret type. But the code above would almost
RR> certainly benefit by being part of the API.

My code is also more complex because it doesn't assume anything will
work properly :) It belongs in a package instead of at the user level.

Ted