NCC#2003092385 [db-wg] uodate failure

>   I am wondering whether "inline" error messages are such a good idea?
>   I guess there's a good chance that they remain in the updated and
>   re-sent message. That could give us another set of failure modes.

do you send the diagnostic output back to a compiler?

>   _I_ would probably go for a more descriptive error message at the end,
>   including the _attribute_ that made the reference, _and_ the _string_
>   used as a reference.

then have line and character position numbers.  you _really_ need a
completely unambiguous trail to the offending text

randy