[Telepathy] Account.Interface.ChannelRequests

[mikhail.zabaluev at nokia.com]

Hi,

>-----Original Message-----
>From: telepathy-bounces at lists.freedesktop.org 
>[mailto:telepathy-bounces at lists.freedesktop.org] On Behalf Of 
>ext Will Thompson
>Sent: Monday, November 24, 2008 3:08 PM
>To: telepathy at lists.freedesktop.org
>Subject: Re: [Telepathy] Account.Interface.ChannelRequests
>
>Dafydd Harries wrote:
>> Ar 24/11/2008 am 13:04, ysgrifennodd Alberto Mardegan:
>>   
>>> - Performance: The ChannelDispatcher object is not 
>necessary, and the 
>>> ChannelRequest is a good thing only in theory: with its 
>Proceed() method and the 
>>> need to connect to the Failed and Succeeded signals, I'm 
>afraid it's making 
>>> things slower.

Connecting to GObject signals hardly has any impact.
Installation of watches for DBus signals is done internally by DBus-Glib. If it's done suboptimally, it's a common problem and needs to be fixed as such, but we shouldn't design our APIs to work around it.

>> Signals don't need to be connected each time, you can just 
>add a message filter
>> that match on interface + signal name, which has a one-off 
>setup cost.
>>   
>
>No, you can't, because then you get woken up about *every* channel
>request, which is what the "Request as separate object" design 
>is trying
>to avoid.

True. And it might outweigh the need to call Proceed. A needless wakeup of a "cold" process which happens to listen to the same object's signals may be very costly.
We already talked about how the client library can completely hide Proceed.

Best regards,
  Mikhail