Re: [Qemu-devel] [PATCHv11 13/31] aio / timers: aio_ctx_prepare sets timeout from AioContext timers

[Wenchao Xia <xiawenc@linux.vnet.ibm.com>]

于 2013-8-20 17:29, Alex Bligh 写道:
>
> On 20 Aug 2013, at 08:19, Wenchao Xia wrote:
>
>>   Thanks for the explanation. It seems *timeout is always set
>> to -1 before calling GSource's prepare(), so
>> "*timeout = qemu_soonest_timeout(*timeout, 10);"
>> will always get *timeout = 10, so this call can be saved.
>
> I believe that's incorrect too. glib *currently* calls
> the API that way, but there's nothing to say a future
> or past glub couldn't call each g_source with the same
> timeout pointer, with each g_source adjusting the timeout
> downwards. Or (for instance) call it with 0 for a
   So it is an undefined value, should avoid use it?
For example, other gsource have minimum timeout of 5, and
aio_ctx_prepare() is called with 0, then returned timeout
is 0, but it is supposed to work with timeout 5 as the glib
doc described:
https://developer.gnome.org/glib/2.32/glib-The-Main-Event-Loop.html#GSourceFuncs

That is my opinion, but I haven't read other project's
code about prepare() usage, so can't be sure. Guess maintainer
can give a conclusion.


> non-blocking prepare. Therefore the implemented way
> is safer (only reducing the timeout).
>
-- 
Best Regards

Wenchao Xia