Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:30878
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: unknown (pb1.pair.com: domain php.net does not designate 72.36.186.210 as permitted sender)
Message-ID: <4696C77B.2060005@php.net>
Date: Fri, 13 Jul 2007 01:29:47 +0100
User-Agent: Thunderbird 2.0.0.4 (Windows/20070604)
MIME-Version: 1.0
To: David <planetbeing@gmail.com>
CC: 'Stanislav Malyshev' <stas@zend.com>, 
 'Sebastian Bergmann' <sb@sebastian-bergmann.de>,
  internals@lists.php.net
References: <ee59b7580707111343s23b4f678q68f2c1934efaccee@mail.gmail.com> <f74tjb$ju9$1@sea.gmane.org> <46969CCA.3010608@zend.com> <000901c7c4e3$2b803030$82809090$@com>
In-Reply-To: <000901c7c4e3$2b803030$82809090$@com>
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Subject: Re: [PHP-DEV] Mid-term Update for Cycle Collector (Google Summer
 of Code Project)
From: scottmac@php.net (Scott MacVicar)

Hi,

David wrote:
> Hi all,
> 
>> -----Original Message-----
>> From: news [mailto:news@sea.gmane.org] On Behalf Of Sebastian Bergmann
>> Sent: Thursday, July 12, 2007 2:52 AM
> 
>>  As a first step, I think the code changes that convert direct accesses
>>  to the reference counters to the proper macros should be merged. These
>>  changes should be done in any case and it minimizes your patch and
>> makes
>>  it more readable.
> 
> I'll get on that. Should I create a patch for 5.2 only, or should one for
> HEAD be created as well?
> 
>> -----Original Message-----
>> From: Stanislav Malyshev [mailto:stas@zend.com]
>> Sent: Thursday, July 12, 2007 2:28 PM
> 
>> Indeed, memory usage reduction is very impressive - though I'm somewhat
>> surprised so much memory is wasted, is this some special very circular
>> case?
> 
> The Template test does indeed involve a lot of cycles, which is why it was
> recommended to me as a good test case. However any program that generates a
> large amount of objects that are self-referential can do the same thing.
> See: http://bugs.php.net/bug.php?id=33595
> 
>> However, twofold speed reduction can be bad - but maybe it can be
>> optimized.
> 
> Yes, that's very true. Although, anything that involves checking and freeing
> over a gigabyte worth of objects is going to take a long time. I'll keep
> working on it.
> 
> My current priority for optimization is reducing overhead for acyclic
> programs. Three bytes are added to the zval struct with the current
> implementation and that struct is copied around a lot. Most notably, this
> causes an annoying 5% slow down in bench.php. I'm not certain what real
> world impact this has, but data with the php.ini option zend.enable_gc set
> to 0 would be very helpful.

Could the garabage collector be started / stopped during script
executing from a userland function? zend_gc_enable/disable() in addition
to an ini option to enable it at script startup?

It would give developers the opportunity to start the checking mechanism
if they knew they were working with acyclic objects.

> 
> 
> David
> 

Scott