Anssi Hannula wrote:
> Klaus Schmidinger wrote:
>> Anssi Hannula wrote:
>
>>> diff -Nurp -x '*~' vdr-1.4.0/device.c vdr-1.4.0-fix/device.c
>>> --- vdr-1.4.0/device.c 2006-04-14 17:34:43.000000000 +0300
>>> +++ vdr-1.4.0-fix/device.c 2006-05-12 06:31:20.000000000 +0300
>>> @@ -627,7 +627,7 @@ eSetChannelResult cDevice::SetChannel(co
>>> // use the card that actually can receive it and transfer data from
>>> there:
>>>
>>> if (NeedsTransferMode) {
>>> - cDevice *CaDevice = GetDevice(Channel, 0, &NeedsDetachReceivers);
>>> + cDevice *CaDevice = GetDevice(Channel, Setup.PrimaryLimit,
>>> &NeedsDetachReceivers);
>>> if (CaDevice && CanReplay()) {
>>> cStatus::MsgChannelSwitch(this, 0); // only report status if
>>> we are actually going to switch the channel
>>> if (CaDevice->SetChannel(Channel, false) == scrOk) { //
>>> calling SetChannel() directly, not SwitchChannel()!
>>> @@ -1158,7 +1158,7 @@ int cDevice::Ca(void) const
>>>
>>> int cDevice::Priority(void) const
>>> {
>>> - int priority = IsPrimaryDevice() ? Setup.PrimaryLimit - 1 :
>>> DEFAULTPRIORITY;
>>> + int priority = ActualDevice() == this ? Setup.PrimaryLimit - 1 :
>>> DEFAULTPRIORITY;
>>> for (int i = 0; i < MAXRECEIVERS; i++) {
>>> if (receiver[i])
>>> priority = max(receiver[i]->priority, priority);
>>> @@ -1183,6 +1183,8 @@ int cDevice::ProvidesCa(const cChannel *
>>>
>>> bool cDevice::Receiving(bool CheckAny) const
>>> {
>>> + if (this == cTransferControl::ReceiverDevice())
>>> + return true;
>>> for (int i = 0; i < MAXRECEIVERS; i++) {
>>> if (receiver[i] && (CheckAny || receiver[i]->priority >= 0)) //
>>> cReceiver with priority < 0 doesn't count
>>> return true;
>>
>> I'm afraid this patch has a nasty side effect.
>>
>> Asssume the following scenario:
>>
>> - two devices, primary is 1
>> - all channels are available on both devices, except for channels
>> 10, 11 and 12 which can only be received by device 2
>> - channels 10 and 11 are on the same transponder, channel 12 is on a
>> different one
>> - switch directly to channel 10 -> transfer mode starts
>> - press "Up" to switch to channel 11, but that channel is skipped and
>> it switches to channel 12
>>
>> So I guess I can't accept this patch.
>
> Thank you for taking a look.
>
> However, I don't see why there is such a problem in the above scenario:
>
> - let's consider PrimaryLimit is 0 (the default)
> - SwitchChannel() gets called (device.c line 581)
> - PrimaryDevice()->ProvidesChannel() returns false (line 591)
> - GetDevice() is called (line 281)
> - for the device 1 ProvidesChannel() returns false (line 288)
> - ProvidesChannel() is called for device 2 (dvbdevice.c line 767)
> - Priority = 0, this->Priority() = -1, so hasPriority = true
> - ProvidesChannel() result = true is set in line 785
> - ProvidesChannel() return true and ndr = false (device.c line 288)
> - (device[i]->Receiving() && !ndr) evaluates true (line 290)
> - pri = 0 so device 2 is selected
> - SwitchChannel() calls another SwitchChannel (line 601)
> - SwitchChannel() calls SetChannel (line 568)
> - and so on
>
> Hopefully I'll have time to test this scenario later today.
Well, just try it - it did behave here as described.
> BTW, isn't the PrimaryLimit there so that liveview would not be
> distracted when a recording with (priority < primarylimit) is trying to
> start? Why is it used for primary device exclusively, then? If the
> liveview is in transfer-mode, shouldn't that be protected as well?
This parameter was introduced at a time when it was not yet
possible to record and replay on the FF cards at the same time,
so that a timer wouldn't prevent the user from live viewing.
Now that the FF cards can record and replay at the same time,
the primary device actually wouldn't need to be "protected" any
more, because if it has to be used for recording (for instance
because it is the only device that can provide a certain channel),
it could just fetch the live channel from a different device and
show it in Transfer Mode.
> If you think it should be used for primary device only, only the change
> in Receiving() in my patch is necessary. If you think it should indeed
> be used for whatever device the user currently is watching, there are
> some more places where transfermode-liveview priority is assumed 0, when
> it should be PrimaryLimit (at least in device.c line 591 call to
> GetDevice()).
> Or if you still think the PrimaryLimit adds only complexity for not much
> of real value, then remove it ;) and assume 0 for it's value everywhere.
I'd really prefer to drop the whole PrimaryLimit thing, because, as
we see in this thread, it makes things rather complex.
If users program so many concurrent timers that the system needs all the
devices to record them, well, then that's apparently what they wanted.
Looking back at your original posting, I believe the actual problem that
triggered all this was that streamdev would interrupt live viewing in
Transfer Mode, right? Well, what if streamdev, when selecting the device
to use, could avoid the ActualDevice()?
To test this (without modifying the cDevice API) you could introduce a
bool AvoidActualDevice = false;
right before
cDevice *cDevice::GetDevice(const cChannel *Channel, int Priority, bool *NeedsDetachReceivers)
and inside that function check that variable, and if it is true and the
checked device is the ActualDevice(), continue the 'for' loop:
for (int i = 0; i < numDevices; i++) {
if (AvoidActualDevice && device[i] == ActualDevice())
continue;
bool ndr;
...
Finally, streamdev would have to be modified so that it sets AvoidActualDevice
to 'true' before calling cDevice::GetDevice(), and back to 'false' afterwards.
If this turns out to be feasible, we could change cDevice::GetDevice() in version 1.5.x
so that it has an additional parameter for this.
Klaus
