Re: [PATCH v2 6/6] imap-send: correctly report "host" when using "tunnel"

[Junio C Hamano <gitster@pobox.com>]

Jeff King <peff@peff.net> writes:

> Yes, I agree that the scenario I'm giving is contrary to what the docs
> say. But IMHO it is worth preferring what the code does now versus what
> the docs say. The current behavior misbehaves if you configure things
> badly (accidentally mix and match imap.host and imap.tunnel). Your new
> behavior misbehaves if you have two correctly-configure imap stanzas
> (both with a host/tunnel combo). Those are both fairly unlikely
> scenarios, and the outcomes are similar (we mix up credentials), but:
>
>   1. In general, all things being equal, I'd rather trust the code as
>      the status quo. People will complain if you break their working
>      setup. They won't if you fix the documentation.
>
>   2. In the current behavior, if it's doing the wrong thing, your next
>      step is to fix your configuration (don't mix and match imap.host
>      and imap.tunnel). In your proposed behavior, there is no fix. You
>      are simply not allowed to use two different imap tunnels with
>      credential helpers, because the helpers don't receive enough
>      context to distinguish them.
>
>      And that is not even "two imap tunnels in the same config". It is
>      really per user. If I have two repositories, each with
>      "imap.tunnel" in their local config, they will still invoke the
>      same credential helpers, and both will just see host=tunnel. The
>      namespace for "host" really is global and should be unique (ideally
>      across the Internet, but at the very least among the hosts that the
>      user contacts).

All good points.

> Yes, there are many config schemes that would avoid this problem. If you
> are going to tie the two together, I think it would make sense to use
> real subsections based on the host-name, like:
>
>   # hostname is the subsection key; it also becomes a label when
>   # necessary
>   [imap "example.com"]
>
>   # does not even need to mention a hostname. We'll assume example.com
>   # here.
>   tunnel = "any-command"
>
>   # assumes example.com as hostname; not needed if you are using a
>   # tunnel, of course
>   protocol = imaps
>
> But I would not bother going to that work myself. IMHO imap-send is
> somewhat of an abomination, and I'd actually be just as happy if it went
> away. But what you are doing seems to go totally in the wrong direction
> to me (keeping it, but breaking a rare but working use case to the
> benefit of a rare but broken misconfiguration).

Yup.