[pulseaudio-discuss] Memory oddities in module-dbus-protocol

[David Henningsson]

On 01/26/2012 01:08 PM, Tanu Kaskinen wrote:
> On Thu, 2012-01-26 at 12:35 +0100, David Henningsson wrote:
>> I'm trying to trace down a memory error in module-dbus-protocol, but I
>> am confused.
>>
>> Either you use synchronous hooks, and then you can reference the
>> original core objects, or you use asynchronous subscription events, in
>> which case the pointers can become dangling at any time, so you can't
>> point into the core and have to make memory copies of *everything*. Right?
>
> Make copies, or check if the objects are still in the core before
> referencing them. I tried to make those checks where needed, but
> apparently I failed at that.
>
>> The card is handled by subscription events, and I get memory errors on
>> quit [1] due to it referencing the profile's name as a hash key (!). But
>> while investigating I also saw pa_dbusiface_card_profile objects having
>> a pointer to the profile, which is disturbing.
>>
>> If this is the only place where things are broken, I can fix it. But if
>> the same logic faults apply to all over the dbus protocol module, that
>> makes me quite uneasy with shipping it in a stable distribution,
>> especially as I don't know any consumer that makes use of this information.
>
> Now that you've pointed out this issue, I don't trust the code at all
> anymore. The only sane way forward is to convert all code in the dbus
> module to use hooks instead of the subscription system.
>
> I volunteer to do that work.

FYI, for the reasons above, module-dbus-protocol has now been disabled 
in the upcoming release of Ubuntu, 12.04. We just commented the line out 
in /etc/pulse/default.pa so it's easy to re-enable.

(Btw, the only known user of module-dbus-protocol is qpaeq, which Ubuntu 
does not currently ship.)

-- 
David Henningsson, Canonical Ltd.
http://launchpad.net/~diwic