svn commit: samba r20394 - in branches/SAMBA_3_0/source: include lib smbd

[James Peach]

On 29/12/2006, at 10:21 PM, Jeremy Allison wrote:

> On Fri, Dec 29, 2006 at 08:26:11PM -0800, James Peach wrote:
>>
>> On 28/12/2006, at 11:47 PM, Volker Lendecke wrote:
>>
>>> On Thu, Dec 28, 2006 at 02:24:16PM -0800, James Peach wrote:
>>>> What's the goal of this work? ie. can you give a idea of the final
>>>> design that you are working towards?
>>>
>>> Easy: Survive the Samba4 RAW-NOTIFY test.
>>
>> So are you intending that all the notification events pass through
>> the messaging system? fam and inotify are more suited to adding file
>> descriptors to the main select loop. I hope that I'll get to adding
>> notify support for BSD/Darwin soon, and I think this will suit a
>> similar model.
>
> Actually, I think the best way of doing this is to hook
> the FAM/inotify events into the main loop and get them
> to send notification messages to be picked up in the same
> way that systems without FAM or inotify will receive them.
>
> That way we keep the server receieve event -> client notification path
> the same for all systems, and only add in extra front ends
> to send "receieve event" messages into the "standard"
> event processing (if you see what I mean).

I'm not sure I see the need for any messaging if you have a kernel- 
supported notification mechanism. I don't have a lot of context here,  
but I don't really see why you can't keep the same recv -> notify  
path without using messaging.

--
James Peach | jpeach at samba.org