Where should the type information be: in tags and descriptors 469

Hmm. I have done some ergonomic analysis of that, and my conclusions are that, yes, it is much more attractive, even to me - but I have no idea why! Working that way is almost always more interactions, slower and less reliable. I have observed experts at work, and my conclusions are based on that as much as on my own experiences. Yes, I did some time estimates of typical tasks.

In this context, Emacs-Lisp is an XEDIT-REXX copy. The problem with both is when you want to do something that is repebreastive but too complex for the built-in operations. The overhead with dropping into REXX or LISP, programming it and coming out is significant enough to discourage this for anything that is feasible by manual repebreastion. And that ignores any need to swap out thinking about (say) C syntax which you are editing and swap in thinking about REX-LISP syntax; many people find this takes a significant time (tens of seconds).

Another, even more important, disadvantage is that bugs are essentially unreportable. There is no way to convince someone that the interface is misbehaving if you can't provide evidence, and that is why many 3270 interfaces were so buggy (even after years of use) and most GUI ones are many times worse, even after decades of use by millions of users. Most users seem to have installed a filter in their brain where they don't even notice such bugs - and I don't understand that, either!

I got one such bug in ISPF fixed by tracing the TGETs and demonstrating that the result was not consistent, but it was a foul task; there was another that involved a race condition on the wire that I baulked at even attempting. I did once work out how a GUI (say X) could be designed for supportability, but the executive conclusion was not to start from existing ones; while you could add tracing, almost no support staff would be competent to analyse it, and it couldn't be automated.

I think that you could do it properly, but you would have to include such diagnostic support in the architecture. Bolting it on simply cannot be done.

Regards, Nick Maclaren.