httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jim Jagielski <...@jaguNET.com>
Subject Re: svn commit: r774454 - in /httpd/httpd/trunk: include/ap_slotmem.h modules/mem/mod_plainmem.c modules/mem/mod_sharedmem.c server/slotmem.c
Date Thu, 14 May 2009 14:16:07 GMT

On May 14, 2009, at 2:33 AM, jean-frederic clere wrote:

> Jim Jagielski wrote:
>> On May 13, 2009, at 3:40 PM, Jim Jagielski wrote:
>>>
>>> On May 13, 2009, at 3:22 PM, jean-frederic clere wrote:
>>>
>>>> jim@apache.org wrote:
>>>>> Author: jim
>>>>> Date: Wed May 13 18:04:05 2009
>>>>> New Revision: 774454
>>>>> URL: http://svn.apache.org/viewvc?rev=774454&view=rev
>>>>> Log:
>>>>> ap_slotmem_mem no longer exists. We never provide the ptr to
>>>>> the actual mem. Instead, we rely solely on getter/setter
>>>>
>>>> Why? I am using it :-(
>>>>
>>>
>>> Oops! Sorry. We can add it back in... But what does
>>> ap_slotmem_mem provide that ap_slotmem_get/ap_slotmem_put?
>>> Again, the idea is that we abstract out the memory handling.
>>> So why do you need the memory pointer if you have a getter and
>>> setter which handles accessing... But no prob putting it back.
>>>
>> patch reverted
>
> Thanks.
> The idea of ap_slotmem_mem is to provide an address in the shared  
> area that can be used for example like the proxy_worker_stat. Using  
> get/put would require 2 memcpy.... I hate memcpy :)
>

This is true... Updating would require a memcpy to read in and a memcpy
to update, whereas having a pointer means we can adjust the slot
as is (at most, one memcpy). Most slots would need to return  
APR_ENOTIMPL
because a concept of a "pointer to a slot" doesn't make sense.

Mime
View raw message