Re: git-credential-cache--daemon quits on SIGHUP, can we change it to ignore instead?

From: Jeff King <peff@peff.net>
Date: Oct 26, 2015, 21:50 UTC
Message-ID: <20151026215016.GA17419@sigill.intra.peff.net>
In-Reply-To: <xmqqfv0ylwa7.fsf@gitster.mtv.corp.google.com>

On Sun, Oct 25, 2015 at 09:58:56AM -0700, Junio C Hamano wrote:

Show 24 quoted lines

> >>> I cannot speak for the person who was primarily responsible for
> >>> designing this behaviour, but I happen to agree with the current
> >>> behaviour in the situation where it was designed to be used.  Upon
> >>> the first use in your session, the "daemon" is auto-spawned, you can
> >>> keep talking with that same instance during your session, and you do
> >>> not have to do anything special to shut it down when you log out.
> >>> Isn't that what happens here?
> >>
> >> After looking at this some more, I've discovered this is NOT what
> >> actually happens here. If I "git push" from a shell and then log out
> >> and log in again, another "git push" does NOT ask me for a password.
> >> In other words, the daemon is NOT shut down automatically when I log
> >> out. Given that, does it make sense to change the daemon to ignore
> >> SIGHUP, or is there some way to change it so that it does exit on
> >> logout?
> 
> I have a feeling that it would be moving in a wrong direction to
> change the code to ignore HUP, as I do think "logout to shutdown"
> would be the desired behaviour.  If you are not seeing that happen,
> perhaps the first thing to do is to figure out why and fix the code
> so that it happens?
> 
> I dunno.  I'll cc Peff so that he can take a look when he comes
> back.

I could see it going both ways.

If SIGHUP means "I am logging out, my session is over", then I agree it makes sense to drop any credentials. And that is what SIGHUP meant when people logged in through hard-wired terminals.

But these days, people often have several simultaneous sessions open. They may have multiple ssh sessions to a single machine, or they may have a bunch of terminal windows open, each of which has a login shell and will send HUP to its children when it exits. In that case, you have a meta-session surrounding those individual terminal sessions, and you probably do want to keep the cache going as long as the meta session[1].

This is all further complicated by bash's huponexit option, which I think is off by default. So I, for example, have never noticed this behavior even with multiple xterms, because my cache never actually gets SIGHUP. I don't know what shell Noam is using, but I wonder if tweaking that option (or a similar one if not bash) might be helpful to signal "let this stuff keep running even after I exit".

But I am also not opposed to making it configurable somehow in git, if there really are two cases that cannot otherwise be distinguished.

-Peff

[1] Of course we have no idea when that meta-session is closed. But if
    you have a script that runs on X logout, for instance, you could put
    "git credential-cache exit" in it.