What sort of schools of thought are there on the way that exceptions
should be chosen when one is writing libraries of code?

I am in the process of writing a library of modules for arrays,
strings, and lists, which will provide a common interface for all
three - so that the naming and calling conventions will be identical.
I want to also make the exception behaviour either identical or
symmetrical.  I'm not sure which.

I.e. should I declare different exceptions in each module, or should
all the modules share the same exceptions.

NJ SML seems to take the first aproach a bit, but it mixes the two a
bit.  I'm not really sure what I want, but the approach of

    Failure of string

is really easy to use, in practice.  Though, if there were a complex
piece of code, I can see that it could be hard to figure out WHERE the
error came from.