Let's say I have a worktree that is locked by some other command.

$ got log 
got: worktree already locked
$ tog     
tog: no git repository found

Wait, what?

That problem was introduced by
df6221c7  use a separate .cvg meta data directory for cvg(1) work trees

In got_worktree_open():

+               if (meta_dir == NULL) {
+                       for (i = 0; i < nitems(meta_dirs); i++) {
+                               err = open_worktree(worktree, worktree_path,
+                                   meta_dirs[i]);
+                               if (err == NULL)
+                                       break;
+                       }

That ignores the GOT_ERR_WORKTREE_BUSY error from .got, proceeds
to try .cvg, and reports the error resulting from a missing .cvg.
It doesn't affect got(1), because that always uses the .got directory.

Does this look like a sensible fix?

diff /home/naddy/got
commit - 2cafc7864a7ffed14fede7310877bc28c6e7ef8f
path + /home/naddy/got
blob - c3846b45dec95fcd923225f11b5f013d97fc39c0
file + lib/worktree_open.c
--- lib/worktree_open.c
+++ lib/worktree_open.c
@@ -286,7 +286,8 @@ got_worktree_open(struct got_worktree **worktree, cons
 			for (i = 0; i < nitems(meta_dirs); i++) {
 				err = open_worktree(worktree, worktree_path,
 				    meta_dirs[i]);
-				if (err == NULL)
+				if (err == NULL ||
+				    err->code == GOT_ERR_WORKTREE_BUSY)
 					break;
 			}
 		} else

-- 
Christian "naddy" Weisgerber                          naddy@mips.inka.de