Processing a file that may already be open

Processing a file that may already be open

[Sean McAfee]

I'd like to be able to write a routine that opens a file and reads it,
and potentially modifies its contents and saves it back to disk.  The
slightly tricky bit is that if there are any open buffers visiting that
file, they should be updated immediately.  (I don't want to modify the
file behind any buffers' backs, that is.)

I've written the following macro to encapsulate what seems to be the
necessary logic:

(defmacro with-visited-file (path &rest body)
  (let ((retain (gensym))
        (buffer (gensym)))
    `(let* ((,retain (find-buffer-visiting ,path))
            (,buffer (find-file ,path)))
       (unwind-protect
           (save-excursion
             (save-restriction
               (widen)
               (beginning-of-buffer)
               ,@body))
         (or ,retain (kill-buffer ,buffer))))))

It seems to work okay, but are there any lurking gotchas?  Or, is there
any better approach?

Re: Processing a file that may already be open

[Sean McAfee]

Stefan Monnier <monnier@iro.umontreal.ca> writes:
>> (defmacro with-visited-file (path &rest body)
[...]

> - use make-symbol rather than gensym: it's faster, doesn't pollute the
>   global symbol table (aka obarray), and doesn't rely on CL.

gensym is just a thin wrapper around make-symbol; how could it pollute
the global symbol table?

I usually prefer gensym because I don't like to be obliged to give the
new symbol a name.  I don't care what it's named.  And I don't mind
relying on CL.  It's the closest I get to being able to use Common Lisp
on a regular basis.

> - avoid evaluating "file" multiple times (in case the caller does
>   (with-visited-file (setq foo (concat foo ".el")) blabla))

Good point.

> - avoid find-file: find-file is a user-level command, Elisp code should
>   use find-file-noselect instead.  find-file will call switch-to-buffer
>   to display the buffer, which can fail or pop up a new window or a new
>   frame, none of which we want to see happen here.

Also a good point.  Thanks.

There's still a few minor issues to be worked out, though.  Like, should
I save the buffer after editing it?  Obviously yes, if the buffer didn't
exist before.  But if any buffers were already visiting the file, I
should probably only save it if none of them have been modified since
the file was last saved.