Re: Link arrays / syntax

[Andrew Johnson <anj@aps.anl.gov>]

Marty Kraimer wrote:
> Are you saying that we should not implement block, process, and wait 
> semantics as part of a link definition?
>
> I am under the impression that the general concensus that it is a good 
> idea and we were deciding semantics.
>
> Am I wrong?

I'm *not* against providing control over synchronization and sequencing 
of links, especially when it comes to arrays of them, but I don't 
believe all of the "block, process and wait" attributes make sense when 
you're dealing with single links, especially when it comes to fields 
like SDIS and TSEL.

Now I'm probably only complaining about some aspects of block and wait 
here, since process obviously *is* a link attribute for the caPut/caGet 
link types.  I also accept that we need the ability to control whether 
get/put operations on individual links cause the source record to wait 
for the target operation to complete.  However the descriptions I've 
seen so far of what the "block, process and wait" attributes mean 
haven't been very clear.

I suspect that the process attribute of a caMonitor link type actually 
has a different meaning to that of a caGet (for caMonitor it controls 
processing of the source record when a monitor event occurs, not the 
processing of the target record) - am I right in that?  If so, I would 
really want to use a different name for it, something like updateScan?

I *am* saying that the complex sequencing and synchronization controls 
should not be part of the basic link type itself, but are more logically 
part of the array of links that the calc and sequence records etc. 
actually implement.  Instead of using an array of links, they could 
contains an array of structures, where each structure contains a link 
field plus some other fields that control the seq/sync controls for that 
link.  This would allow the sequence record to have even more complex 
controls than calc, since it wants to allow the user to sequence input 
and output links together in groups maybe, insert delays and so on.

I'd also like to see the discussion of the capabilities of individual 
link types distinguished from that of the capabilities of the generic 
link interface. A process attribute controls a capability of CA and DB 
links, but means nothing to my xyzzy link type, so please don't make my 
link have to implement stuff it doesn't need.

- Andrew
--
English probably arose from Normans trying to pick up Saxon girls.