bug#11665: kill-buffer gives an error on killing dead buffers

[martin rudalics]

> Create a buffer "test.tmp" and create an indirect buffer:
>
>    (with-current-buffer (make-indirect-buffer "test.tmp" "test1")
>                (add-hook 'kill-buffer-hook
>                          '(lambda ()
>                             (setq kill-buffer-hook nil)
>                             (kill-buffer (buffer-base-buffer)))
>                          t t))
>
> (The above triggers the killing of the base buffer when an indirect
> buffer is killed)
>
> Now, kill the base buffer "test.tmp", which raises:
>
>   signal(error ("Buffer #<killed buffer> is not a live buffer"))
>   error("Buffer %s is not a live buffer" #<killed buffer>)
>   window-normalize-buffer(#<killed buffer>)
>   replace-buffer-in-windows(#<killed buffer>)
>   kill-buffer(#<killed buffer>)
>   (if (and (boundp (quote sub-kill-buffer-and-its-windows)) 
sub-kill-buffer-and-its-windows (fboundp (quote kill-buffer-and-its-windows))) 
(kill-buffer-and-its-windows (current-buffer)) (kill-buffer (current-buffer)))
>
>
> This doesn't occur when killing an indirect buffer.

I don't have `sub-kill-buffer-and-its-windows' so I can't repeat this
easily.  Anyway, here's what I suppose to happen:

Killing "test.tmp" implicitly calls `kill-buffer' on "test1" because
killing a base buffer kills all its indirect buffers.

   Killing "test1" calls `kill-buffer' on "test.tmp" because that's on
   the hook.

      Killing "test.tmp" calls `kill-buffer' on "test1" again.

         Killing "test1" now succeeds and returns.
        
      Killing "test.tmp" now succeeds.

  Killing "test1" now tries to continue with `replace-buffer-in-windows'
  but this fails because the buffer is no more live.

I also suppose that the

>   kill-buffer(#<killed buffer>)

is a red herring in the sense that `kill-buffer' was actually invoked
with a live buffer but when the trace was printed the buffer was already
dead while `replace-buffer-in-windows' was really called with a dead
buffer as argument.

Basically, I could exit `replace-buffer-in-windows' when the argument
buffer is not live but that's not nice.  So maybe the attached patch is
better.  Can you try it?

martin

*** src/buffer.c        2012-05-29 16:13:38 +0000
--- src/buffer.c        2012-06-10 12:52:57 +0000
***************
*** 1526,1531 ****
--- 1526,1535 ----
        UNGCPRO;
      }

+   /* If killing the indirect buffers has killed our buffer, return.  */
+   if (NILP (BVAR (b, name)))
+     return Qnil;
+ 
    /* Run replace_buffer_in_windows before making another buffer current
       since set-window-buffer-start-and-point will refuse to make another
       buffer current if the selected window does not show the current