freeing fports

freeing fports

[Aaron VanDevender]

I noticed that the fport free function doesn't return how much
it frees. This patch should take care of that.

--- libguile.old/fports.c       2003-08-26 23:35:34.000000000 -0500
+++ libguile/fports.c   2003-08-26 23:47:23.000000000 -0500
@@ -827,8 +827,11 @@
 static size_t
 fport_free (SCM port)
 {
+  size_t size;
+  scm_t_port *pt = SCM_PTAB_ENTRY (port);
+  size = pt->read_buf_size + pt->write_buf_size + sizeof(scm_t_fport);
   fport_close (port);
-  return 0;
+  return size;
 }
  
 static scm_t_bits



_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-user

Re: freeing fports

[Kevin Ryde]

[-- Attachment #1: Type: text/plain, Size: 739 bytes --]

Aaron VanDevender <sig@netdot.net> writes:
>
> I noticed that the fport free function doesn't return how much
> it frees.

Thanks.

> This patch should take care of that.

No, not quite, I think you need to allow for buffers possibly being
"shortbuf" (size 1), which is not freed.

Probably best to make the adjustment in fport_close, since a port can
be closed before it's freed.  setvbuf looks like it suffers the same
sort of leak.

        * fports.c (scm_setvbuf, fport_close): Call scm_done_free for space
        freed.

Unfortunately this still doesn't make open+close leave scm_mallocated
unchanged, so maybe there's a leak somewhere else too.  Maybe it's not
worth worrying about in the 1.6 branch.  Seems ok in the latest cvs.


[-- Attachment #2: fports.c.done_free.diff --]
[-- Type: text/plain, Size: 1111 bytes --]

--- fports.c.~1.100.2.4.~	1970-01-01 10:00:01.000000000 +1000
+++ fports.c	2003-09-22 14:51:58.000000000 +1000
@@ -191,9 +191,15 @@
 
   /* silently discards buffered chars.  */
   if (pt->read_buf != &pt->shortbuf)
-    scm_must_free (pt->read_buf);
+    {
+      scm_must_free (pt->read_buf);
+      scm_done_free (pt->read_buf_size);
+    }
   if (pt->write_buf != &pt->shortbuf)
-    scm_must_free (pt->write_buf);
+    {
+      scm_must_free (pt->write_buf);
+      scm_done_free (pt->write_buf_size);
+    }
 
   scm_fport_buffer_add (port, csize, csize);
   return SCM_UNSPECIFIED;
@@ -817,10 +823,17 @@
   if (pt->read_buf == pt->putback_buf)
     pt->read_buf = pt->saved_read_buf;
   if (pt->read_buf != &pt->shortbuf)
-    scm_must_free (pt->read_buf);
+    {
+      scm_must_free (pt->read_buf);
+      scm_done_free (pt->read_buf_size);
+    }
   if (pt->write_buf != &pt->shortbuf)
-    scm_must_free (pt->write_buf);
+    {
+      scm_must_free (pt->write_buf);
+      scm_done_free (pt->write_buf_size);
+    }
   scm_must_free ((char *) fp);
+  scm_done_free (sizeof (*fp));
   return rv;
 }
 

[-- Attachment #3: Type: text/plain, Size: 139 bytes --]

_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-user

Re: freeing fports

[Aaron VanDevender]

On Wed, Sep 24, 2003 at 09:49:22AM +1000, Kevin Ryde wrote:
> Unfortunately this still doesn't make open+close leave scm_mallocated
> unchanged, so maybe there's a leak somewhere else too.  Maybe it's not
> worth worrying about in the 1.6 branch.  Seems ok in the latest cvs.

I don't think its worth worrying about, as this is all based on
scm_must_malloc/free which has been depricated in HEAD in favour
of scm_gc_malloc. I just wrote the patch because I was working
on a new port implementation for 1.6 and noticed the fport was
always returning 0, but I've since moved on to the 1.7 way of
doing things.

cya
.sig


_______________________________________________
Guile-user mailing list
Guile-user@gnu.org
http://mail.gnu.org/mailman/listinfo/guile-user