I think it would be better if the procedure described in this SRFI
is called something other than just ERROR. Some implementations
(Chez Scheme, SXM and possibly others) already have their own
procedures named ERROR with different set of arguments:

(error where format-string object ...)

Although in most use cases the 'where' argument is a symbol,
other argument types, including strings, are accepted, so there's
no way to make a single procedure supporting both conventions.

I believe calling this SRFI's procedure SIGNAL-ERROR will
solve the compatibility problem and make it obvious that the
underlying semantics is the one suggested by RnRS "signal an error"
concept (opposed to, say, the one of a constructor creating
an "error" object to serve as an argument to a separate SIGNAL
procedure).

Sergei