[tbnl-devel] Re: TBNL file descriptor leak

[Robert J. Macomber]

On Sun, Jan 22, 2006 at 12:41:15AM +0100, Edi Weitz wrote:
> > [not subscribed to tbnl-devel]
>
> You should be... :)

All right; now I am.
          
> I don't really like that idea.  Instead, I now decided to send a 400
> (Bad Request) reply to the client.  Check out TBNL 0.9.5 and see if it
> works for you.

It fixes the problem I was seeing, but I've uncovered a new one.

However, this one's not really TBNL's fault.  If you send a non-UTF8
urlencoded sequence _before_ the `?' in the request, TBNL dies the
same as before, because headers are read from the bivalent stream with
read-line (that is, in character mode) but Apache(*) de-url-encodes
names before passing them on for the `script-filename' header.  It can
be fixed by either making kmrcl open the bivalent stream explicitly
with :external-format :latin-1 or by running sbcl in a latin-1 locale.

In any case, I haven't hit any more socket leaks now that that one's
been tracked down, so it looks like all's right in the world, at least
as far as TBNL's concerned.

(*) now I'm using apache2 and mod_lisp2 version 1.3
-- 
Robert Macomber
tbnl at rojoma.com