`with-output-to-string' leaks buffers

`with-output-to-string' leaks buffers

[Romain Francoise]

Shouldn't `with-output-to-string' reap the temporary buffer it
creates even if the code it wraps exits nonlocally?

(git.el installs a function in `after-save-hook' which uses this
macro around code that signals an error, and as a result I had a
little more than 250 *string-output* buffers slowing things down
after a few days of uptime...)

2008-09-19  Romain Francoise  <romain@orebokech.com>

	* subr.el (with-output-to-string): Make sure that the temporary
	buffer gets killed.

Index: lisp/subr.el
===================================================================
RCS file: /cvsroot/emacs/emacs/lisp/subr.el,v
retrieving revision 1.606
diff -u -r1.606 subr.el
--- lisp/subr.el	7 Sep 2008 09:15:43 -0000	1.606
+++ lisp/subr.el	19 Sep 2008 21:56:40 -0000
@@ -2586,12 +2586,13 @@
   (declare (indent 0) (debug t))
   `(let ((standard-output
 	  (get-buffer-create (generate-new-buffer-name " *string-output*"))))
-     (let ((standard-output standard-output))
-       ,@body)
-     (with-current-buffer standard-output
-       (prog1
-	   (buffer-string)
-	 (kill-buffer nil)))))
+     (unwind-protect
+	 (progn
+	   (let ((standard-output standard-output))
+	     ,@body)
+	   (with-current-buffer standard-output
+	     (buffer-string)))
+       (kill-buffer standard-output))))
 
 (defmacro with-local-quit (&rest body)
   "Execute BODY, allowing quits to terminate BODY but not escape further.

Re: `with-output-to-string' leaks buffers

[Glenn Morris]

Romain Francoise wrote:

> Shouldn't `with-output-to-string' reap the temporary buffer it
> creates even if the code it wraps exits nonlocally?

FWIW, I would say yes.

Re: `with-output-to-string' leaks buffers

[Stefan Monnier]

> Shouldn't `with-output-to-string' reap the temporary buffer it
> creates even if the code it wraps exits nonlocally?

I'd think so, yes.  Could you check the users to see if you find any
suspicious use-case?


        Stefan

Re: `with-output-to-string' leaks buffers

[Romain Francoise]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> I'd think so, yes.  Could you check the users to see if you find
> any suspicious use-case?

All the users in Emacs use it as intended and do not hold on the
buffer, so I installed my changes.

Re: `with-output-to-string' leaks buffers

[Chong Yidong]

Romain Francoise <romain@orebokech.com> writes:

> Stefan Monnier <monnier@iro.umontreal.ca> writes:
>
>> I'd think so, yes.  Could you check the users to see if you find
>> any suspicious use-case?
>
> All the users in Emacs use it as intended and do not hold on the
> buffer, so I installed my changes.

Thanks.