Re: git-clone fails when current user is not in /etc/passwd

From: Jeff King <peff@peff.net>
Date: Dec 10, 2015, 18:34 UTC
Message-ID: <20151210183426.GA27609@sigill.intra.peff.net>
In-Reply-To: <CAKfKJYuP1E5aC7SPO91j-KS-P41pYaM7NkTJ9uG6b=VvxbOBwg@mail.gmail.com>

On Wed, Dec 09, 2015 at 05:35:52PM -0500, Taylor Braun-Jones wrote:

Show 7 quoted lines

> > Well.. reflog needs it. So either you disable reflog at clone time or
> > define name/email via config file. I don't see anything wrong with
> > this behavior.
> 
> Can't git just use "unknown" and "unknown@localhost" if the name or
> email can not be determined from xgetpwuid? This seems to be the way
> that Mercurial behaves.

Yeah, I think there is a mismatch here in the ident code. When we are making a commit (which is likely to get published and is set in stone), we ask for "IDENT_STRICT", which will barf if we can't come up with something sensible. For reflogs, we are OK putting whatever auto-generated crap in there we can come up with.

But before we even hit the strict-check, we call xgetpwuid_self(), which unconditionally dies on failure. I think that function needs to be taught a "gently" form which we use for non-strict ident lookups. Unfortunately it's a little non-trivial because the strictness will need to get passed all the way down to ident_default_name() (and we need to make sure that a non-strict check followed by a strict one does not fail; i.e., that the first does not pollute the contents of git_default_name).

-Peff