dired-kill-tree

dired-kill-tree

[Luc Teirlinck]

The docstring of `dired-kill-tree' says:

"Kill all proper subdirs of DIRNAME, excluding DIRNAME itself.

But you have to understand that this assumes that DIRNAME is a
directory name and not a directory file name.  Otherwise, DIRNAME _is_
killed.  Anybody having taken any look at the Dired Elisp code will
probably have no trouble guessing this implicit assumption.

But `dired-kill-tree' can be used interactively and the implicit
assumption may not be obvious to most interactive users.  Actually, I
myself find it more often useful _in interactive usage_ to
_deliberately_ pass a directory file name to kill the entire tree
including DIRNAME.  This saves me a step, because I nearly invariably
_want_ to kill DIRNAME itself.

So what about turning this bug officially into a feature, with the
following change in the docstring:

===File ~/dired-aux-diff====================================
*** dired-aux.el	05 Jun 2004 21:07:07 -0500	1.119
--- dired-aux.el	06 Jun 2004 15:52:23 -0500	
***************
*** 1834,1839 ****
--- 1834,1844 ----
  
  (defun dired-kill-tree (dirname &optional remember-marks)
    "Kill all proper subdirs of DIRNAME, excluding DIRNAME itself.
+ Note that DIRNAME itself is _only_ excluded if it is a directory name,
+ which on most systems means that it ends with a slash.  If DIRNAME is
+ a directory file name, which on most systems means that it does not end
+ with a slash, then DIRNAME is killed too.  The latter is sometimes useful
+ in interactive usage, when DIRNAME is itself a subdirectory.
  With optional arg REMEMBER-MARKS, return an alist of marked files."
    (interactive "DKill tree below directory: ")
    (setq dirname (expand-file-name dirname))
============================================================

Re: dired-kill-tree

[Miles Bader]

On Sun, Jun 06, 2004 at 04:19:03PM -0500, Luc Teirlinck wrote:
> But you have to understand that this assumes that DIRNAME is a
> directory name and not a directory file name.  Otherwise, DIRNAME _is_
> killed.  Anybody having taken any look at the Dired Elisp code will
> probably have no trouble guessing this implicit assumption.

Is there a (reasonable) reason for this rather odd behavior?

-miles
-- 
`...the Soviet Union was sliding in to an economic collapse so comprehensive
 that in the end its factories produced not goods but bads: finished products
 less valuable than the raw materials they were made from.'  [The Economist]

Re: dired-kill-tree

[Luc Teirlinck]

Miles Bader wrote:

   On Sun, Jun 06, 2004 at 04:19:03PM -0500, Luc Teirlinck wrote:
   > But you have to understand that this assumes that DIRNAME is a
   > directory name and not a directory file name.  Otherwise, DIRNAME _is_
   > killed.  Anybody having taken any look at the Dired Elisp code will
   > probably have no trouble guessing this implicit assumption.

   Is there a (reasonable) reason for this rather odd behavior?

Tecnically because:

  (setq dirname (expand-file-name dirname))

and later:

      (if (and (not (string-equal dir dirname))

(then we do _not_ delete DIR).

DIR is a directory name.  If DIRNAME is not, then DIR and DIRNAME are
not string-equal.

This bug can trivially be fixed by changing:

  (setq dirname (expand-file-name dirname))

into:

  (setq dirname (file-name-as-directory (expand-file-name dirname)))

That would be an obvious alternative patch.  (Leaving the docstring
unchanged.)  Somehow, the buggy behavior turns out (at least in my
usage) to be useful interactively.

Sincerely,

Luc.

Re: dired-kill-tree

[Luc Teirlinck]

Actually, there is a third (much more general) way to view the anomaly.

The first way to deal with the problem was to quite simply make the
behavior into a feature by documenting it into the docstring.

The second is to change:

  (setq dirname (expand-file-name dirname))

into:

  (setq dirname (file-name-as-directory (expand-file-name dirname)))

thereby getting rid of the feature.

The third way is that maybe the behavior of the "D" interactive code,
usd by `dired-kill-tree' is anomalous:

>From (elisp)Interactive Codes:

`D'
     A directory name.  The default is the current default directory
     of
     the current buffer, `default-directory' (*note File Name
     Expansion::).  Existing, Completion, Default, Prompt.

If I have a directory ~/subdir, and no other subdirectory of ~/
starting with ~/subd, then if I enter ~/subd RET, then depending on my
customizations, Emacs will, in various ways, tell me that the only
possible completion is ~/subdir/.  But if I type ~/subdir, it will
accept that as is.  That is, contrary to its documentation, the `D'
interactive code does not only accept existing directory names, but
also existing directory file names.  The two are _only_ equivalent if
the function using `D' makes them equivalent.  The completions listed
after say, ~/subd RET, do not interact very well with the first way to
deal with the problem, because they only include ~/subdir/, not
~/subdir.

Sincerely,

Luc.

Re: dired-kill-tree

[Richard Stallman]

      (defun dired-kill-tree (dirname &optional remember-marks)
	"Kill all proper subdirs of DIRNAME, excluding DIRNAME itself.
    + Note that DIRNAME itself is _only_ excluded if it is a directory name,
    + which on most systems means that it ends with a slash.  If DIRNAME is
    + a directory file name, which on most systems means that it does not end
    + with a slash, then DIRNAME is killed too.  The latter is sometimes useful
    + in interactive usage, when DIRNAME is itself a subdirectory.
      With optional arg REMEMBER-MARKS, return an alist of marked files."

I think this behavior is rather error-prone and counterintuitive.
I would rather consider it a bug, and change it to exclude
DIRNAME even if it is specified in the syntactic form of a file name.

If this makes it hard to kill a subdir, we could add a new interface
for doing that, or we could make this command ask whether
to kill directory DIRNAME as well.  That would provide an accident-proof
way to ask for either behavior.

Re: dired-kill-tree

[Luc Teirlinck]

Richard Stallman wrote:

   I think this behavior is rather error-prone and counterintuitive.
   I would rather consider it a bug, and change it to exclude
   DIRNAME even if it is specified in the syntactic form of a file name.

   If this makes it hard to kill a subdir, we could add a new interface
   for doing that, or we could make this command ask whether
   to kill directory DIRNAME as well.  That would provide an accident-proof
   way to ask for either behavior.

What about the following patch, which kills DIRNAME as well if a
prefix argument is given?

===File ~/dired-aux-diff====================================
*** dired-aux.el	07 Jun 2004 15:59:16 -0500	1.120
--- dired-aux.el	07 Jun 2004 21:04:22 -0500	
***************
*** 1852,1870 ****
  			  (> (dired-get-subdir-min elt1)
  			     (dired-get-subdir-min elt2)))))))
  
! (defun dired-kill-tree (dirname &optional remember-marks)
    "Kill all proper subdirs of DIRNAME, excluding DIRNAME itself.
! With optional arg REMEMBER-MARKS, return an alist of marked files."
!   (interactive "DKill tree below directory: ")
!   (setq dirname (expand-file-name dirname))
    (let ((s-alist dired-subdir-alist) dir m-alist)
      (while s-alist
        (setq dir (car (car s-alist))
  	    s-alist (cdr s-alist))
!       (if (and (not (string-equal dir dirname))
! 	       (dired-in-this-tree dir dirname)
! 	       (dired-goto-subdir dir))
! 	  (setq m-alist (nconc (dired-kill-subdir remember-marks) m-alist))))
      m-alist))
  
  (defun dired-insert-subdir-newpos (new-dir)
--- 1852,1874 ----
  			  (> (dired-get-subdir-min elt1)
  			     (dired-get-subdir-min elt2)))))))
  
! (defun dired-kill-tree (dirname &optional remember-marks kill-root)
    "Kill all proper subdirs of DIRNAME, excluding DIRNAME itself.
! Interactively, you can kill DIRNAME as well by using a prefix argument.
! In interactive use, the command prompts for DIRNAME.
! 
! When called from Lisp, if REMEMBER-MARKS is non-nil, return an alist
! of marked files.  If KILL-ROOT is non-nil, kill DIRNAME as well."
!   (interactive "DKill tree below directory: \ni\nP")
!   (setq dirname (file-name-as-directory (expand-file-name dirname)))
    (let ((s-alist dired-subdir-alist) dir m-alist)
      (while s-alist
        (setq dir (car (car s-alist))
  	    s-alist (cdr s-alist))
!       (and (or kill-root (not (string-equal dir dirname)))
! 	   (dired-in-this-tree dir dirname)
! 	   (dired-goto-subdir dir)
! 	   (setq m-alist (nconc (dired-kill-subdir remember-marks) m-alist))))
      m-alist))
  
  (defun dired-insert-subdir-newpos (new-dir)
============================================================

Re: dired-kill-tree

[Richard Stallman]

    What about the following patch, which kills DIRNAME as well if a
    prefix argument is given?

That might be a good interface.