With the upcoming 1.2 release of CCL (formerly called OpenMCL), the old implementation of TIMER-REQUEST is no longer available. Hunchentoot relies on this in its WITH-TIMEOUT implementation which is (only) called from FORCE-OUTPUT*. As asynchronous timeouts are generally unsafe, I would like to get rid of WITH-TIMEOUT altogether. Instead, I think it would be better to only use the underlying streams' mechanisms for timeouts, if available. CCL 1.2 will have stream timeouts, LW already has them and the other Lisp implementations will propably support some form of timeouts in I/O sooner or later.
Does anyone here have sentiments regarding WITH-TIMEOUT? If not, I'd propose that WITH-TIMEOUT be totally removed from Hunchentoot.
-Hans
On Feb 18, 2008 10:39 AM, Hans Hübner hans@huebner.org wrote:
With the upcoming 1.2 release of CCL (formerly called OpenMCL), the old implementation of TIMER-REQUEST is no longer available.
I got actually fooled by the fact that TIMER-REQUEST and friends lives in the CCL package, but is defined in the ACL-COMPAT system. I'd still want to get rid of it, so if anyone has comments, please make them.
Thanks, Hans
Hunchentoot relies on this in its WITH-TIMEOUT implementation which is (only) called from FORCE-OUTPUT*. As asynchronous timeouts are generally unsafe, I would like to get rid of WITH-TIMEOUT altogether. Instead, I think it would be better to only use the underlying streams' mechanisms for timeouts, if available. CCL 1.2 will have stream timeouts, LW already has them and the other Lisp implementations will propably support some form of timeouts in I/O sooner or later.
Does anyone here have sentiments regarding WITH-TIMEOUT? If not, I'd propose that WITH-TIMEOUT be totally removed from Hunchentoot.
-Hans
On Mon, 18 Feb 2008 10:39:40 +0100, "Hans Hübner" hans@huebner.org wrote:
Does anyone here have sentiments regarding WITH-TIMEOUT? If not, I'd propose that WITH-TIMEOUT be totally removed from Hunchentoot.
Not that I'm totally against it, but what's the reason? Right now, I'm seeing it as a kind of stopgap solution for Lisps without stream timeouts. If you don't have /any/ timeouts, you're pretty prone to DoS attacks, aren't you?
Can't WITH-TIMEOUT be implemented in OpenMCL 1.2?
On Mon, 18 Feb 2008 17:15:39 +0100, Edi Weitz edi@agharta.de wrote:
On Mon, 18 Feb 2008 10:39:40 +0100, "Hans Hübner" hans@huebner.org wrote:
Does anyone here have sentiments regarding WITH-TIMEOUT? If not, I'd propose that WITH-TIMEOUT be totally removed from Hunchentoot.
Not that I'm totally against it, but what's the reason? Right now, I'm seeing it as a kind of stopgap solution for Lisps without stream timeouts. If you don't have /any/ timeouts, you're pretty prone to DoS attacks, aren't you?
Although, having thought about this for the duration of another cigarette, I tend to think that it's probably not so elegant and clear to mix stream timeout mechanisms and "other" timeouts. Yeah, maybe we should get rid of this and rely on the streams to take care of that. Might also increase pressure on the implementations which don't have it yet.
-
Edi Weitz -
Hans Hübner