Eli Zaretskii <eliz@gnu.org> writes:

> Thanks.  So it sounds like you are asking for a feature that currently
> doesn't exist, AFAIU.

I'm not sure I'd classify it as a new feature.  An existing interface is
broken under some conditions.

> I added a couple of people to this discussion who were involved with
> auth-source-pass, in the hope that they will have suggestions and
> comments.

Thank you.

Now, onto why I don't think this is a new feature:

Here's an example auth-source-search invocation that can demonstrate the
problem (assuming that the user has these a dev.gentoo.org secret on
port imaps with user arsen):

  (auth-info-password (car (auth-source-search :host "dev.gentoo.org"
                                               :port "imaps"
  					       :user "arsen")))

Following M-x epa-file-disable RET M-x auth-source-forget-all-cached RET
the above returns an encrypted string rather than its actual password.

This means that a current feature (auth-source-search) breaks under some
conditions.

I've worked out a fix, tested with the following:

  (require 'auth-source-pass)
  (setq auth-sources '(password-store))
  (auth-info-password (car (auth-source-search :host "dev.gentoo.org"
                                               :port "imaps"
                                               :user "arsen")))

I've attached the patch, though it lacks a regression test.  The reason
for this is that I want to spare the auth-source-pass developers some
triage, and that there's currently no regression tests for --read-entry.