Hello,

thanks for the patch; I had marked to my to-do that:

"- current nio is leaking memory: 1) foreign read buffer is not freed when async-fd is closed; 2) foreign write buffers (created in map-to-foreign) aren't freed, there's just code to (remove-if #'buffer-written buffers). "

but haven't had time to fix it. I think your patch fixes both cases.

Thanks, Risto

2006/9/28, Nonny Mouse enonymoos@gmail.com:

Hi All,

Had a look over the NIO work, running ab2(apache-bench - http://httpd.apache.org/docs/2.2/programs/ab.html) ) against it, and it seems to hold up pretty well when compared with apache/mod_lisp2 and araneida (I'll post some results later on).

I ran into a memory leak, which the following patch fixed for me. The patch is against the nio-20060204.tar.gz (doesn't look like the cvs is up-to-date?) from the http://common-lisp.net/project/nio/ website.

I'm no lisp expert, so any feedback appreciated.

Cheers,

Paul.

nio-20060204$ diff async-fd.lisp.orig async-fd.lisp 184c184

< (with-slots (write-queue read-fd write-fd) async-fd

...

(with-slots (write-queue read-fd write-fd foreign-read-buffer) async-fd

189a190

...

   (foreign-free foreign-read-buffer)

216,220c217,226 < ;; remove entries that have been fully written < (setf write-queue (remove-if #'(lambda (x) < (with-slots (size written) x < (= size written)))

< write-queue))

...

;; remove (and dealloc buffers) of entries that have been fully

written

...

(let ((unfinished-entries NIL))
  (setf write-queue
      (nreverse
       (dolist (entry write-queue unfinished-entries)
         (with-slots (size written) entry
           (if (= size written)
               (with-slots (buffer) entry
                 (foreign-free buffer))
               (push entry unfinished-entries)))))))

---

nio-devel mailing list nio-devel@common-lisp.net http://common-lisp.net/cgi-bin/mailman/listinfo/nio-devel