bug#13575: 24.1; dired-mark-sexp misparses directory contents

[Wolfgang Jenkner]

On Mon, Jan 28 2013, Sean McAfee wrote:

> I recently wanted to mark all zero-size files in a dired buffer.  I
> found that dired-mark-sexp seemed to be just what I needed, but when I
> invoked it with the expression (= size 0), I found that it marked every
> file, even those of nonzero size.
>
> Here's a test directory I set up:
>
>   /home/mcafee/test:
>   total used in directory 28 available 102379756
>   drwxrwxr-x  2 mcafee mcafee  4096 Jan 27 18:31 .
>   drwxr-xr-x 44 mcafee mcafee 12288 Jan 27 18:31 ..
>   -rw-rw-r--  1 mcafee mcafee     8 Jan 27 18:31 bar
>   -rw-rw-r--  1 mcafee mcafee    12 Jan 27 18:31 baz
>   -rw-rw-r--  1 mcafee mcafee     4 Jan 27 18:31 foo
>
> In this directory, I ran dired-mark-sexp with the following expression,
> intended to exercise all of the available symbols:
>
> (progn (message
> "name=%s;inode=%s;s=%s;mode=%s;nlink=%s;uid=%s;gid=%s;size=%s;time=%s;sym=%s"
> name inodes mode nlink uid gid size time sym) nil)
-------^ inode s

I think you have to customize dired-listing-switches to something like
"-alis" so that inode number and block count are actually shown in the
buffer.

> Here are the messages that were produced:
> name=.;inode=0;s=0;mode=drwxrwxr-x;nlink=2;uid=mcafee;gid=2;size=0;time=
> mcafee  4096 Jan 27 18:31;sym=
> name=..;inode=0;s=0;mode=drwxr-xr-x;nlink=44;uid=mcafee;gid=44;size=0;time=
> mcafee 12288 Jan 27 18:31;sym=
[...]
> Clearly, the "time" variable includes more of the directory listing than
> it should, and "size" is incorrectly set to zero for all files, as I
> originally noticed.

That's a bug in dired-x (wrong use of the somewhat awkward
directory-listing-before-filename-regexp variable).  As a first attempt
to fix this I propose the following barely tested patch.

Wolfgang

diff --git a/lisp/dired-x.el b/lisp/dired-x.el
index a2c1303..73eebd1 100644
--- a/lisp/dired-x.el
+++ b/lisp/dired-x.el
@@ -1449,8 +1449,7 @@ to mark all zero length files."
         ;; to nil or the appropriate value, so they need not be initialized.
         ;; Moves point within the current line.
         (dired-move-to-filename)
-        (let (pos
-              (mode-len 10) ; length of mode string
+        (let ((mode-len 10) ; length of mode string
               ;; like in dired.el, but with subexpressions \1=inode, \2=s:
               (dired-re-inode-size "\\s *\\([0-9]*\\)\\s *\\([0-9]*\\) ?"))
           (beginning-of-line)
@@ -1472,22 +1471,18 @@ to mark all zero length files."
           ;; Karsten Wenger <kw@cis.uni-muenchen.de> fixed uid.
           (setq uid (buffer-substring (1+ (point))
                                       (progn (forward-word 1) (point))))
-          (re-search-forward directory-listing-before-filename-regexp)
-          (goto-char (match-beginning 1))
-          (forward-char -1)
-          (setq size (string-to-number
-                      (buffer-substring (save-excursion
-                                          (backward-word 1)
-                                          (setq pos (point)))
-                                        (point))))
-          (goto-char pos)
-          (backward-word 1)
+         (re-search-forward directory-listing-before-filename-regexp)
+         (re-search-backward directory-listing-before-filename-regexp)
+         (forward-word 1)
+         (save-excursion
+           (setq size (string-to-number
+                       (buffer-substring (point) (progn (backward-word 1)
+                                                        (point))))
+                 gid (buffer-substring (1- (point)) (progn (backward-word 1)
+                                                           (point)))))
           ;; if no gid is displayed, gid will be set to uid
           ;; but user will then not reference it anyway in PREDICATE.
-          (setq gid (buffer-substring (save-excursion
-                                        (forward-word 1) (point))
-                                      (point))
-                time (buffer-substring (match-beginning 1)
+          (setq time (buffer-substring (1+ (point))
                                        (1- (dired-move-to-filename)))
                 name (buffer-substring (point)
                                        (or