Re: bug? in checkout with ambiguous refnames

Uwe Kleine-König  [off-list ref] writes:

So working copy and cache are at refs/tags/sgu/mxs-amba-uart, HEAD
points to refs/heads/sgu/mxs-amba-uart

I somehow thought that we had an explicit logic to favor an exact branch
name for "git checkout $branch" even when refs/something-other-than-head/$branch 
exists, while issuing the ambiguity warning.

Ahh, what this part of the code in builtin/checkout.c does is totally
wrong:

	/* we can't end up being in (2) anymore, eat the argument */
	argv++;
	argc--;

	new.name = arg;
	if ((new.commit = lookup_commit_reference_gently(rev, 1))) {
		setup_branch_path(&new);

		if ((check_ref_format(new.path) == CHECK_REF_FORMAT_OK) &&
		    resolve_ref(new.path, rev, 1, NULL))
			;
		else
			new.path = NULL;
		parse_commit(new.commit);
		source_tree = new.commit->tree;
	} else
		source_tree = parse_tree_indirect(rev);

It uses lookup_commit_reference_gently() that follows the usual
tags-then-heads preference order, but then uses setup_branch_path() to
prefix the raw name with "refs/heads", which is totally backwards.
It should do something like:

 - use setup-branch-path to get refs/heads/$name

 - check-ref-format and resolve it; if these fail, then we are detaching
   head at rev;

 - otherwise, if the result of the resolution is not the same as rev, what
   we have in rev is incorrect (it was taken from the usual
   tags-then-heads rule but "checkout $name" must favor local branches).
   update the rev with the result of resolving refs/heads/$name

 - parse new.commit out of rev.  we are checking out the branch $name.