Experimental Physics and
Industrial Control System

Marty Kraimer <[email protected]> · Thu, 16 Jun 2005 09:07:56 -0500

Please note that Andrew is on vacation until June 27.

I just want to give a little background on how V4 epicsTypes.h evolved.

After the April developer's meeting I started thinking hard about how to

implement the features described by the V4 DBD definitions. The first

problem was "What data types are needed". There are two issues:

1) How to make the information available to dataAccess, i.e. how the

data in the records can be accessed without using the header files

generated from the DBD record definitions. This means that it must be

easy to generate propertyCatalogs for accessing the data contained in a

record.

2) How to make it easy and efficient it access the data via the header

files.

V4 Design : epicsTypes is an attempt to address issue 1)
V4 Design: dbdClass is an attempt to address issue 2)

Thus fields in DBD records that will be made available to dataAccess in

a generic manner will be one of the epicsTypes. dbdClass allows record

support to get/put data in a convient and efficient way. It also makes

it possible for any code that includes dbdClass to easily and efficently

access any field of any DBD record or struct.

I have spent most of my time since the April meeting working on

epicsTypes and dbdClass.

They have both evolved considerably during this time. Andrew and I have

had may discussions about these.

For the last few weeks Andrew has been making major changes to epicsTypes.h.
He took what I started and made many changes. Two major contributions are:

1) made the EpicsBuffer even more generic than I did. I had separate

buffers for strings and for arrays.

2) made the definitions real C++ objects. He said that what I was

proposing was C objects via C++ syntax. Note that dbdClass still suffers

from this defect. I have to fix this before the July developer's meeting.

The main focus of this effort is how to manage storage for data

associated with epics records.

It seems to me that a gateway has to address the same problem except

that it may be even harder. I will guess that a gateway will be able to

use everything in epicsTypes and also need something like dbdClass. I am

really interested in hearing Ralph's, Benjamin's etc thoughts on how the

gateway will manage storage. In particular how will it handle the data

connected to an arbitrary propertyId?

I still keep thinking about Doug Murray's request to show a "Hello

World" example for dataAccess. It made me start thinking about what

would happen if the request was Show me "hello world" that included an

array and a simple structure of information. Thus I think that some

additional support is needed to make it easier to use dataAccess.

Marty

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System