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

From: Jeff King <peff@peff.net>
Date: Oct 30, 2015, 00:10 UTC
Message-ID: <20151030001000.GA2123@sigill.intra.peff.net>
In-Reply-To: <CAM-tV--B3HaC1DcORfnx9bWW9-quyk0=pQDxmvonc=6dgrMOxA@mail.gmail.com>

On Tue, Oct 27, 2015 at 11:46:20PM -0400, Noam Postavsky wrote:

Show 12 quoted lines

> > I dunno. I think the auto-spawn is really what makes it usable; you can
> > drop it in with "git config credential.helper config" and forget about
> > it. Anything more fancy requires touching your login/startup files.
> > Certainly I'm not opposed to people setting it up outside of the
> > auto-spawn, but I wouldn't want that feature to go away.
> 
> Perhaps we could express the auto-spawn more explicitly, something
> like "git config credential.pre-helper start-cache-daemon". A way to
> run a command before the credential helpers start would be useful to
> our magit workaround for this issue (currently we start the daemon
> before "push", "fetch", and "pull", but it won't work with user
> aliases that run those commands).

I'm not clear on when the pre-helper would be run. Git runs the helper when it needs a credential. What git command would start it? Or are you proposing a new credential interface for programs (like magit) to call along the lines of "do any prep work you need; I might be asking for a credential soon", without having to know the specifics of the user's helper config.

You can do that already like:

  echo url=dummy://a:b@c/ | git credential approve

though I guess for some helpers, that may actually store the bogus value (for the cache, it inserts a dummy cache entry, but that's not a problem; something with permanent storage would be more annoying).

I guess the most elegant thing would be to add an "init" command to the helper interface. So magit would run:

  git credential init

which would then see that we have "credential.helper" defined as "foo", and would run:

  git credential-foo init

which could be a noop, start a daemon, or whatever, depending on how the helper operates.

I dunno. It almost seems like adding a credentialcache.ignoreHUP option would be less hacky. :)

> I'm not sure it's that widely used. Perhaps most people use ssh-agent?
> That's what I do, though I've been experimenting with credential-cache
> since it was brought up by a magit user.

I don't know. Certainly up until a few years ago, anybody sane was using ssh-agent, because the http password stuff was so awful (no storage, and we didn't even respect 401s for a long time). But these days sites like GitHub push people into using https as the protocol of choice. Mostly, I think, because there was a lot of support load in teaching people to set up ssh. But I guess a lot of those people are on non-Linux platforms.

-Peff