Re: [PATCH] clone: allow detached checkout when --branch takes a tag

2012/1/5 Jeff King [off-list ref]:

quoted

@@ -721,6 +722,14 @@ int cmd_clone(int argc, const char **argv, const char *prefix)

                      strbuf_release(&head);

                      if (!our_head_points_at) {
+                             strbuf_addstr(&head, "refs/tags/");
+                             strbuf_addstr(&head, option_branch);
+                             our_head_points_at =
+                                     find_ref_by_name(mapped_refs, head.buf);
+                             strbuf_release(&head);
+                     }
+
+                     if (!our_head_points_at) {

Hmm. The context just above your patch that got snipped does this:

   strbuf_addstr(&head, src_ref_prefix);
   strbuf_addstr(&head, option_branch);
   our_head_points_at =
       find_ref_by_name(mapped_refs, head.buf);

where src_ref_prefix typically is "refs/heads/", and clearly you are
meaning to do the same thing for tags. But the use of "src_ref_prefix"
is interesting.

It is always "refs/heads/" unless we are cloning into a bare mirror, in
which case it is "refs/". So with your patch in the non-mirror case,
doing "--branch=foo" would try "refs/heads/foo" followed by
"refs/tags/foo". Which makes sense. But in the mirror case, it will try
"refs/foo" followed by "refs/tags/foo", which is kind of odd.

I wonder, though, if the original code makes any sense. By using
"refs/", I would have to say "--branch=heads/foo", which is kind of
weird and undocumented. I think it should probably always be
"refs/heads/", no matter if we are mirroring or not.

--branch should not be used with --mirror in my opinion. --branch
changes HEAD so it's no longer an exact mirror.
-- 
Duy