Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Thu, 14 Nov 2002 12:19:19 -0700

> The idea that it is somehow "dangerous" to provide the details
> (which can of course be obtained by sufficient study of the
> code) does not.

Well, I admit that I could be a bit defensive on this issue, and that
the root problem is that this task has not floated to the top of the
list. However, many other parts of EPICS do not have internals
documentation either, and I still feel that publishing a protocol
document is nearly tantamount to specifying an interoperability
interface guaranteeing that programs will not break if we evolve the
interface, and that an interoperability interface seems to be inevitably
what folks are after when they ask for the doc. Given that the
mechanisms for evolution of this interface are not well defined at this
point, then we do need to be careful, and I'm not sure that this is
entirely a paranoid perspective after having seen the hassles resulting
when a what-seemed-at-the-time-to-be-an-innocuous-change was made to an
interoperability interface. 

So I agree that the documentation is useful and certainly necessary
during a review process, and if there are competing versions of the
libraries. However, some work is required up front before it will
appear. This task is more than labeling and describing the bit and bytes
in the stream. We must also thoroughly describe the necessary state
transitions, the necessary approach to timing of UDP retries that will
minimize network loading, the proper buffering schemes that are
required, and the proper extension mechanisms among many other things.
No surprise there I guess.

Jeff

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System