RE: Channel Access Portable Server, passing by pointer in read/write methods

["Jeff Hill" <johill@lanl.gov>]

Nikolay,

For the write implementation what you have will work but you are allowing the client to change the GDD primitive type and application type structure of your service’s representation – which may or may not be appropriate in your situation. You end up with an un-typed service which could cause less predictable behavior. The client could for example change, using one write, a double value surrounded by lots of meta data into the string value “off” with no metadata for example.

For your read implementation the primary concern might be that the service wouldn’t be looking at some of the details of the read request from the client. In particular, the destination GDD comes preformatted with all of the application types that the client is fetching (i.e. all of the GDD application types in a CA DBR_XXXX – say DBR_GRAPHIC_DOUBLE), but your code snippet is substituting instead what the service is storing. Maybe this is ok as long as a superset of all of the application types is always provided (where the superset is modal depending on the DBR_XXXX data model for the value type of string, float, or enum). The client’s request would be rejected for example if the destination value has primitive type double and the substituted value is a string with value “off”. What you have suggested might be ok as long as you don’t mind leaving the error checking to the server. The server does do the necessary primitive type conversion checking of the response. If any of the application types are missing for the clients DBR_XXX type then the client’s read request is rejected.

The code in base/src/makeBaseApp/templates/caServerApp/exPV.cc shows another way to implement a read using a dispatch table for each of the applications types. I don’t claim that it’s the proper way to structure a service, but perhaps it can provide some contrast.

The GDD library also as I recall supports GDD container to GDD container assignment which would be another way to implement the read.

Jeff

From: tech-talk-bounces@aps.anl.gov [mailto:tech-talk-bounces@aps.anl.gov] On Behalf Of Malitsky, Nikolay D
Sent: Thursday, June 25, 2009 12:37 PM
To: tech-talk@aps.anl.gov
Subject: Channel Access Portable Server,passing by pointer in read/write methods

 

Hello,

 

I am considering the Channel Access Portable Server for supporting the C++ middle layer servers.

The associated PVs will maintain one-dimensional arrays of characters. Their lengths can be

significant.  So, passing by pointer in read/write methods would be a preferable way. The ‘write’ case

looks straightforward (as described in the tutorial, p.35):

 

class my PV : public casPV {

….

gdd* pValue;

};

 

caStatus myPV::write(const casCtx& ctx, gdd &value){

…

pValue->unreference();

pValue = &value;

pValue->reference();

…

}

 

The ‘read’ case looks trickier (Probably, because it is not explained).

 

In short, CORBA addresses such case by using the additional wrappers, e.g. gdd_var objects,

in the ‘read’ method. However, the signature of this method is already defined by the basis

class casPV. Fortunately, it seems we can still  use the gdd-based approach using another

mechanism, the pv-associated destructor, e.g. myPVReadDestructor:

 

caStatus myPV::read(const casCtx& ctx, gdd &value){

…

value.putRef(pValue->dataPointer(), new myPVReadDestructor(pValue));

pValue->reference();

}

 

where :

 

myPVReadDestructor::run(…){

pValue->unreference();

}

 

Any comments and suggestions about a better solution.

 

Thank you,

Nikolay