> > I didn't touch the implementation in lib/read_gitconfig.c because
> > that seems to be only used by gotd, which I don't (yet) use and so
> > couldn't easily test. Is that duplicate implementation still needed?
> > I actually made my change there first and it took me a while to
> > realize it was the wrong place.
> 
> I think this is due to the fact that gotd doesn't use the libexec
> helpers, so you end up with two "interfaces": one that runs
> got-read-gitconfig under the hood (lib/read_gitconfig_privsep.c) and one
> that doesn't (lib/read_gitconfig.c).
> 
> (i've been bitten by this separation in the past too, but there are more
> files like that in lib/ which do the same thing.)
> 
> I think however that if we change the behaviour of got-read-gitconfig we
> should keep the same in lib/read_gitconfig.  I can tweak the latter in a
> follow-up if you prefer though.
> 
> I never used git-annex and I don't know how much a server should care,
> but I think that if got(1) can work on a repository, then gotd should be
> able to do it too.

Yes, I think gotd would want to be able to interact with git-annex 
repos. Git-annex repos are also regular git repos, and are synchronized 
using git fetch or push (or perhaps more low-level equivalent git 
commands). So it's quite plausible someone would want to access 
their git-annex repo via gotd. If I understand right, gotd would 
fail to open the repo if git-annex had added one of the strange 
remotes to the config file.

If you would like to update the gotd code in a similar way it would 
be helpful since I don't yet use gotd. Alternatively, I wonder if 
there's room to refactor so that the two implementations share more 
code. I am tempted to take a look but don't know if I'll find time.

-- 
James