As long as the Romberg function faithfully implements the Romberg
algorithm, I see no reason to remove or deprecate it. The Romberg
algorithm may or may not be appropriate or useful for a particular
application, but that's up to the user, as with other functions which
are named for the *algorithm* as opposed to the abstract
*functionality*.
If it were called numeric_integrate, that would be a different matter.
In fact, it would probably be a good idea to build a
numeric_integrate, which intelligently chooses among existing
algorithms, perhaps based on a (symbolic) analysis of the form of the
integrand. That function would be our "best effort" for general
users, as opposed to sophisticated users who want tight control over
the method being used.
-s