From: Junio C Hamano <gitster@pobox.com>
Date: Fri, 09 Dec 2011 18:00:44 GMT
Subject: Re: [PATCHv2 0/13] credential helpers
Message-ID: <7vzkf1fwvn.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20111209022913.GA2600@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

> It works, and it detects truncated output both ways properly (I know
> because I had to update every test, since the old output was missing the
> end-of-credential marker).
>
> It makes me a little sad, because the original format (relying on EOF)
> was so Unix-y.

It saddens me, too.  A reasonable middle ground would be to stop treating
an empty input as "no restriction" but "never matches".

I suspect that it is far more likely for a helper to fail (due to
configuration errors, for example) before it produces any output than
after it gives some but not all output lines.