Re: basic question about push

On Sun, Jan 24, 2010 at 06:59:10PM -0500, Jay Soffian wrote:

quoted

 http://git.wiki.kernel.org/index.php/GitFaq#push-is-reverse-of-fetch

Aside, and I think we've discussed this before, but I wonder if it
would make sense to:

a) Add an option to clone such as "-p [<name>] | --push-as[=<name>]"
where <name> defaults to $(uname -n | cut -f1 -d.) This would setup
the cloned repo with a push refspec
"+refs/heads/*:refs/remotes/<name>/*". e.g.:

$ git clone -o host1 -p host2 ssh://host1/~/repo.git
$ cat repo/.git/config
...
[remote "host1"]
	url = ssh://host1/~/repo.git
	fetch = +refs/heads/*:refs/remotes/host1/*
	push = +refs/heads/*:refs/remotes/host2/*

One problem with this scheme is that it actually does two things:

  1. Tells git to push into the refs/remotes/host2/ hierarchy instead of
     refs/heads/.

  2. Overrides matching behavior to push all heads.

Point (2) is less of a problem than it is without (1) in the sense that
you are pushing to a unique namespace, so you don't have to worry as
much about your crappy half-finished branches filling up the main branch
namespace. But you might not want them publicized at all, either because
they are embarrassingly bad or simply because it is cruft that you will
encounter later and wonder "is this branch, which is missing from my
local repositor (presumably from a long-ago deletion) but present
upstream of any value at all?"

I can't think offhand of a way to get the behavior of (1) without (2).
I know we have a shorthand refspec for "matching", but this would not
strictly be matching. It is "convert refs/heads/X into
refs/remotes/blah/X, and _then_ match".

If we had a refspec that worked in that manner, I would think it's a
pretty good idea (even to turn it on by default for non-bare repos).

b) The controversial part: make this option the default the default
when cloning from a non-bare repo. There would need to be some way to
turn it off.

Hmm. Do we reliably know from the client side whether a remote is bare
or not?

Of course, I'm not sure this would be any less confusing for users.
Would they wonder why they have to merge to see their pushed changes
reflected on the pushed-to repo?

Yes, I think we would still have the FAQ of "I pushed my changes, where
did they go?" but at least the answer would be "git merge foo" and not
"what you are doing is confusing to git and you need to figure out what
kind of workflow you want, redo your push, and then git merge foo".
Which seems like an improvement to me. ;)

-Peff