Experimental Physics and
Industrial Control System

Benjamin Franksen <[email protected]> · Thu, 11 Oct 2007 11:05:14 +0200

On Thursday 11 October 2007 09:05, Heinrich du Toit wrote:
> I think that missing data is not really a problem with records and
> thereforce device support for records.
>
> If all data is required it is most probably not a record but probably
> something like streamdevice or some custom record that can implement
> something else in the interrupt callback to make sure all data is
> processed.
>
> So I would lean towards a solution that put priority on the newest data
> out there.
>
> [...]
>
> I can't really think of a record that would mind missing data?
> The only thing I think of where this is important is for Octet with a
> communications interface - in which case StreamDevice is most likely
> used and not record device support.

I agree with Mark and Eric. It may not be a problem in /most/ situations if 
interrupt are lost. However, there may be cases where this is a problem. I 
can imagine a card wants to acknowlege completion of a long runnign 
operation by issuing an interrupt. If multiple channels per card are 
involved, and each one gets its own completion interrupt, we'll have a 
problem indeed. Always remember Murphy's law ;-)

BTW, if solution 2 gets implemented, I'd make sure the interrupt queue size 
can be configured by the user.

Cheers
Ben

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System