Re: [PATCH] Update SDP storage records when a record is deleted.

[Jaikumar Ganesh <jaikumar@google.com>]

Hi Luiz:

On Fri, Oct 23, 2009 at 10:44 AM, Luiz Augusto von Dentz
<luiz.dentz@gmail.com> wrote:
>
> Hi,
>
> On Fri, Oct 23, 2009 at 12:17 PM, Jaikumar Ganesh <jaikumar@google.com> wrote:
> >
> > My mail to linux bluetooth bounced yesterday.  Sending it again.
> >
> >    The req->profiles_removed gets updated from device->uuids.
> > However, in the case, where we are paired with the remote device and
> > we unpair and then repair, since we will free device->uuids on the
> > unpair, the req->profiles_removed won't get updated after the SDP
> > query on repair. And so we would need to update the storage from
> > device->uuids.
>
> Im not sure if I get you right, but if you are saying that when device
> object is gone profiles_removed is empty that is the a new device and
> we shouldn't have any profile at that point for consistence, so either
> we remove the sdp cache on device_remove_stored or we never remove
> records from the cache. Perhaps the record cache is really useless and
> we should only cache the uuids information and query the record on
> every connect attempt (we are going to connect anyway so the cost is
> about the same.)

When the device is removed, we remove the cache and remove the device->uuids
value. So when the device is created the next time, we cannot depend on
the profiles_removed value, we need to reconstruct from device->uuids.
device_remove_drivers call is based on the profiles_removed flag and hence that
call will never be made, when a new device is created.

My change removes the dependency of SDP records and the probe_drivers,
Thus, removal of SDP records doesn't depend on the profiles_removed flag
but rather uses device->uuids directly.

profiles_added and profiles_removed flags are useful in the case where the user
calls discover services on an already created device and there are
changes in the SDP
records.

>
> > Since req->profiles_removed would be empty, device_remove_drivers
> > won't get called. Also I believe it makes sense to have 2 functions,
> > even though there is slight overhead, since the deletion of SDP
> > records from storage, and removing device_probe_drivers are 2 distinct
> > operations.
>
> Maybe if you called it on device_remove_stored too, but if we are
> going to validate that information on every connection attempt I doubt
> this is useful in the end.
>
> --
> Luiz Augusto von Dentz
> Engenheiro de Computação

Thanks
Jaikumar