Re: sscan record and StreamDevice positioners

[Tim Mooney <mooney@aps.anl.gov>]

Stefan Heim wrote:
> Hi,
>
> sorry for the following lengthy expose to a problem regarding completion 
> notification.
>
> I have several positioners (multiple axes devices) controlled via several 
> get/set pairs of ai/ao records per axis with a StreamDevice device support 
> (DTYP=streams).
>
> Sometimes, the underlying "hardware" protocol offers only a set method without 
> completion notification for a given axis. The StreamDevice protocol file 
> section then has s.th. like
>
>
> > setX {
> >        out "AxisAbs 1 %f";
> >        in "AxisAbs 1 movement_started";
> > }
>
>
> Here the device emits the "AxisAbs 1 movement_started" as mere confirmation of 
> the control order. There is no further reply indicating completion of the 
> positioning operation. 
>
> While moving, the position can be monitored in an event based manner, i.e. 
> a .SCAN="I/O Intr" ai record
>
> | record (ai, "$(IOCNAME):stage:$(AXIS)")
> | {
> |         field (SCAN, "I/O Intr")
> |         field (DTYP, "stream")
> |         field (INP,  "@protocol_file.proto get$(AXIS) PORT")
> | }
>
> based on a protocol
>
>
> > getX {
> >        @init {
> >                out "registerVC 2 IOC:stage:X Joe";
> >                in "Joe Crds %f %*f %*f %*f %*f %*f %*f ";
> >        }
> >        in "Joe Crds %f %*f %*f %*f %*f %*f %*f ";
> > }
>
>
> Here, the device's capability to do registration based position monitoring is 
> gluelessly linked with EPICS' monitor concept. The out command issued during 
> init causes the device to send position update messages ("Joe Crds ...") as 
> if you posted a monitor on one (or several) EPICS process variable(s).
>
> The six floats thrown away in the "in" part of this protocol ("%*f") are 
> typically read by a further set of (six) .SCAN="I/O Intr" ai records, 
> offering completely seperated control information and linear rescaling of the 
> raw values by means of the respective .ASLO and .AOFF fields. With the 
> same .ASLo/.AOFF rescaling applied to the corresponding ao record of a given 
> axis, this rescaling is transparent to the user/client.
>
> In addition, there is a state variable ("ready") available with the same 
> subscription based event notification mechanism,
>
> | record (bi, "$(IOCNAME):stage:ready")
> | {
> |         field (SCAN, "I/O Intr")
> |         field (DTYP, "stream")
> |         field (INP,  "@protocol_file.proto getReady PORT")
> |         field (ZNAM,  "not ready")
> |         field (ONAM,  "ready")
> | }
>
> with protocol 
>
>
> > getReady {
> >        @init {
> >                out "registerVC 32 IOC:stage:ready Joe";
> >                in "Joe %{not ready|ready}";
> >        }
> >        in "Joe %{not ready|ready}";
> > }
>
>
> Now if I want to use the sscan record to operate on these positioners, it 
> expects them to do be capable of ca_put_callback()-based completion 
> notification. However, e.g. travelling from 0.0 to 10.0 takes ~13 s 
>
> | sheim@txm1:~$ camonitor TXM1CL:micos:ready TXM1CL:micos:ZCOARSE &
> | [1] 7781
> ! TXM1CL:micos:ready             2007-08-31 02:09:33.733498 ready
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:11.154559 0 LOLO MAJOR
> | sheim@txm1:~$ caput -c TXM1CL:micos:setZCOARSE 10
> | Old : TXM1CL:micos:setZCOARSE        0
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.735909 0.022526
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.757608 0.039484
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.779335 0.061448
> ! TXM1CL:micos:ready             2007-08-31 02:09:44.800247 not ready
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.803464 0.087914
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.828070 0.120282
> | New : TXM1CL:micos:setZCOARSE        10
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.849671 0.164518
> | sheim@txm1:~$
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.870746 0.205295 
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.891945 0.251714
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.914267 0.303483
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:44.935013 0.36009
> | [...]
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:45.658592 4.99687
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:45.680585 5.21379
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:45.703397 5.42912
> | [...]
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:46.587214 9.98588
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:46.609446 9.99538
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:46.631317 9.99971
> ! TXM1CL:micos:ZCOARSE           2007-08-31 02:09:46.653736 10
> ! TXM1CL:micos:ready             2007-08-31 02:09:47.107716 ready
>  
> but the caput -c above returns "immediately". Furtermore, the device is 
> possibly late to give out the "not ready" notification (cf. trace above), but 
> guarantees to fire "ready" only after (all) movements finished. 
>
> While I read the slides on completion notification in Scans.ppt and tried to 
> understand the busy record, I still wonder what might be the best approach to 
> the problem. Essentially, I need a way to use StreamDevice based devices not 
> offering a synchronous setPosition command as positioners with the sscan 
> record.
>
>
> Any clues?
> -Stefan

The busy record is the simplest mechanism I know of for getting a database
like this to behave correctly when written to by a ca_put_callback() -- which
is what the sscan record does when it needs to be notified that something has
finished.  (You could write custom device support that would prevent the record
from declaring itself done (i.e., executing its forward link) until the motion
is complete, but this would put the record into a situation it doesn't have code 
to handle.)

To implement a busy-record based solution, the record that the sscan record
writes to, to start the device moving, must forward link to a record that
writes '1', using a "PP" link, to a busy record's VAL field.  The record that
detects that the motion is finished must cause the busy record's VAL field to
be set back to '0'.  (Normally, it doesn't matter what link type is used to
reset the busy record, but "CA" is the safest, because it handles the
exceptional case in which the motion completes immediately.)

--
Tim Mooney (mooney@aps.anl.gov) (630)252-5417
Beamline Controls & Data Acquisition Group
Advanced Photon Source, Argonne National Lab.