Re[3]: [Bese-devel] UCW AJAX Problem

[John Doe]

> there's a slot on the application, 'apropos and the slime inspectors
> are your friend... :)

OK ;-)

> i use sbcl head, and it works fine for me. i've only seen such long gc
> times when using lots of weak pointers (sbcl's gc is using linear
> search at a point). i've minimized weak pointer usage it ucw (in a
> recent patch), but there's still one per component. but again, you
> shouldn't experience 20 secs with the latest ucw and sbcl.

I simply forced welcome page of UCW example to refresh every second (with UCW's built-in server) with only 5 connections, and SBCL started to collect garbage every 20-30 sec. It stoped all jobs for 10-15 sec while GC. How to fix it ? May be there is an option
like (setf *use-generational-gc* t) ????

> i'm personally not interested in lw, but patches are welcome.> 

Frankly speaking If there is a way to fix/avoid SBCL's GC problem, I'd prefer to use SBCL ('cause of more efficient compiler and license).

Best regards
XSet



-----Original Message-----
From: "Attila Lendvai" <attila.lendvai at gmail.com>
To: "John Doe" <x-set at mail.ru>
Date: Tue, 26 Jun 2007 10:00:34 +0200
Subject: Re: Re: [Bese-devel] UCW AJAX Problem

> 
> > Thanks for answer, attila. When I change backend from built-in to mod_lisp, IE gets right XML answer (with increased counter and HTML elements) through AJAX, but can't render it properly. I think mod_lisp backend sends something different from built-in web server. Is it possible to turn on more detailed logging on JavaScript side ?
> 
> there's a slot on the application, 'apropos and the slime inspectors
> are your friend... :)
> 
> > Another question about lisp implementations. I see that SBCL 1.0.6 collects garbage too long (up to 20 sec. with default options), it is not acceptable for production. UCW's AJAX branch don't compile with Lispworks 4.4 for Windows (errors related to MOPP), same thing with LispWorks 5 Personal Edition for Linux. What implementation(s) do you actually use ?
> 
> i use sbcl head, and it works fine for me. i've only seen such long gc
> times when using lots of weak pointers (sbcl's gc is using linear
> search at a point). i've minimized weak pointer usage it ucw (in a
> recent patch), but there's still one per component. but again, you
> shouldn't experience 20 secs with the latest ucw and sbcl.
> 
> lw's slot-value-using-class is passing in the slot name instead of the
> effective slot instance, this may cause problems. they should read the
> spec finally... :)
> 
> i'm personally not interested in lw, but patches are welcome.
> 
> -- 
>  attila
>