> What I don't understand is Ray's position that it's a _bad_ thing for
> client software to be able to do anything with addInfo.

Well I'm certainly sorry if that's what you inferred.

My "position" would be that we shouldn't expect clients to do it, and
shouldn't necessarily design diagnostics with that in mind.

And if there is a client developer who intends to do something real smart,
and correspondingly wants us to define a specific diagnostic in a way that
facilitate that, I'm all for it. What I don't want is for us to define alot
of complex diagnostic structures that nobody ever implements.