Experimental Physics and
Industrial Control System

Steve Lewis <[email protected]> · Mon, 17 Oct 2005 12:40:29 -0700

At 2:14 PM -0400 2005/10/17, Kay-Uwe Kasemir wrote:

But back to the technical aspect, did NIF just pick the wrong
interfaces and thus each change requires 2-6?
Is the other OO framework talk mostly by people who
have never actually tried it on the scale of NIF?

Well, there are several aspects here. Because of the need for Ada

language bindings, there were few choices of CORBA vendors; none

offered Dynamic Invocation Interface. So a change of interface is a

very early binding.

There was one Java program that could access tier-1 (IOC) interfaces

as well as tier-2; when started at the command line, it dynamically

parsed *ALL* the IDL in the shop. You were then presented with a

pick list of 300-400 callable interfaces: you filled in the IN

parameters and back came the OUTs. (It took about 15 minutes to

start on a Sunblade 1000.)

The problem was that each object designer created whatever interfaces

they saw fit to support use cases as they came along. So in addition

to the "interesting" ones (like commands), every get/set for every

attribute required a semi-custom interface. What took the place of

the motor record, for example, had about 50 interfaces...

So the interfaces were not "wrong", but in a model where one

requirement -> one interface, then new interfaces (commands +

attributes) are an ongoing reality.

Adding new attributes to the motor record affects only a

sophisticated sequencers, say, that need to manipulate them; it does

not affect displays, archivers, alarm handlers, striptool, etc, etc,

as long as the list of channels is administratively dealt with

properly, as Kay suggests. That's very late binding.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System