On May 24, 2011, at 15:11 , Peter Soetens wrote:

> On Tue, May 24, 2011 at 6:19 PM, Stephen Roderick <kiwi [dot] net [..] ...> wrote:
>> On May 24, 2011, at 08:41 , Peter Soetens wrote:
>>
>>> On Saturday 21 May 2011 14:47:53 S Roderick wrote:
>>>> Patch RTT v1 so that the user can debug why a slave activity can't start.
>>>> This can occur when the user accidentally tries to start a slave activity
>>>> when the master activity isn't yet started. S
>>>
>>> So far hard-real-time use. I now realize that log4cpp::TimeStamp uses
>>> gettimeofday(), which certainly causes mode switches in Xenomai applications.
>>
>> Hmmm ...
>>
>>> We'll have to refine the code in Orocos to use the
>>> TimeStamp::TimeStamp(unsigned int seconds, unsigned int microSeconds)
>>> constructor instead of the default constructor
>>
>> Modify the log4cpp::Timestamp implementation itself? That would introduce a depenancy that log4cpp needs RTT, right? Or are you saying to copy RTT's approach into log4cpp?
>
> No. I was thinking of refining the Category in OCL. The category
> constructs the log message, thus also the timestamp. We should have
> OCL use the RTT to set the correct date in the timestamp class. I
> don't think it's good to let log4cpp depend on RTT, I even would like
> to avoid that it depends on Xenomai, if possible.

+1

>>> There is an example here:
>>> http://www.xenomai.org/index.php/Porting_POSIX_applications_to_Xenomai
>>> on how to get the time of day in Xenomai applications using the posix skin.
>>
>> Why not just use clock_gettime() directly within the Timestamp implementation? We could easily add a cmake check for that.
>
> That adds a dependency on Xenomai, we'd need to rename to
> log4cpp-xenomai/-gnulinux etc. Like to avoid that...
>
>>
>>> Since we need other 'time of day' like services, it's probably better to
>>> implement it in the fosi and then use that throughout the toolchain.
>>
>> We should certainly do this also, I'm just wary of making a dependancy on RTT in log4cpp, just to avoid writing a few lines.
>
> I don't think it's gonna be a big patch... but we need to add support
> in RTT which the OCL Category can use...

Agreed
S