[avahi] when can avahi_entry_group_commit return "Bad state" ?

[Trent Lloyd]

Hi Julien,

On 09/09/2008, at 11:34 PM, Julien PUYDT wrote:

> Julien PUYDT a écrit :
>> Lennart Poettering a écrit :
>>> On Tue, 02.09.08 19:00, Julien PUYDT (julien.puydt at laposte.net)  
>>> wrote:
>>>
>>>>>> I have a bug where I get published correctly at first, but  
>>>>>> don't manage
>>>>>> to update my information later : avahi_entry_group_commit  
>>>>>> returns "Bad
>>>>>> state"...
>>>>>>
>>>>>> What should I do after/before
>>>>>> avahi_entry_group_update_service_txt_strlst so that my data is  
>>>>>> updated?
>>>>> the Avahi client will return AVAHI_ERR_BAD_STATE if you commit an
>>>>> entry group whose connection to the server is dead.
>>>> Hmmm... and how do I know if the connection to the server is dead?
>>>> Shouldn't I know from the callback client?
>>> You can find that out with avahi_client_get_state(). If the state is
>>> any of AVAHI_CLIENT_S_REGISTERING, AVAHI_CLIENT_S_COLLISION and
>>> AVAHI_CLIENT_S_RUNNING a connection tot he Avahi daemon is  
>>> available.
>>>
>>> If the state is AVAHI_CLIENT_S_RUNNING the host name is well
>>> established and you may install entry groups which refer to the host
>>> name. If the state leaves AVAHI_CLIENT_S_RUNNING you should remove
>>> your entry groups since the hostname they internally might refer  
>>> to is
>>> no longer established.
>>>
>>> Browsers/resolvers may be kept during the whole time we are in one  
>>> of
>>> the aforementioned AVAHI_CLIENT_S_xxx states.
>>>
>>> This is documented on http://avahi.org/download/doxygen/main.html
>>> under "How to Register Services".
>>>
>>
>> Registering services is ok, precisely because it's well documented.
>>
>> Updating services isn't that well ; I'm trying to use the
>> avahi_entry_group_update_service_txt function:
>> http://avahi.org/download/doxygen/publish_8h.html#9f805d25a03009baf606c14a4d7cb3dc
>>
>> And I don't manage to get it to work ; I added g_print statements to
>> follow the client and entry group state change, and to my functions
>> which deal with empty group ; here is what it prints :
>>
>> AVAHI_CLIENT_S_RUNNING
>> AVAHI_ENTRY_GROUP_UNCOMMITED
>> avahi_entry_group_add_service_strlst: OK
>> avahi_entry_group_add_service_strlst: OK
>> avahi_entry_group_commit: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_commit: Bad state
>> AVAHI_ENTRY_GROUP_REGISTERING
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_commit: Bad state
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_commit: Bad state
>> AVAHI_ENTRY_GROUP_ESTABLISHED
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_commit: Bad state
>>
>> From this I gather :
>> (1) that whenever my updating function gets called, it triggers the
>> display of :
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_update_service_txt_strlst: OK
>> avahi_entry_group_commit: Bad state
>>
>> FIX: I certainly call it too soon, and should guard against calling  
>> it
>> before AVAHI_ENTRY_GROUP_ESTABLISHED is obtained (using
>> avahi_entry_group_get_state)
>>
>> (2) after AVAHI_ENTRY_GROUP_ESTABLISHED is obtained, I don't get  
>> why I
>> still have problems.
>
> The *only* example of service updating I have is
> client-publish-service.c, which does a service renaming and hence  
> has to
>  reset and publish again. That is not my case : the name is  
> constant, I
> just change some data attached to it -- a case about which the doc  
> (yes,
> I did read it!) says what not to do (free the entry group, which
> prevents protocol optimisations), but not precisely enough what to do.
>
> Would it be possible to have a working example of a use of
> avahi_entry_group_update_service_txt ? I fear I'll have to disable  
> avahi
> in the upcoming ekiga 3.0...

You don't need to _commit an update_service_txt.. you just call it  
once-off by itself.

Regards,
-- 
Trent Lloyd