[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Issue 82 comment



It is not necessarily the case that larger-format floats will always
have larger exponents.  It is also not clear to me that quietly rolling
over into the next-larger float is the right thing to do on exponent
overflow, or that such rollover is efficiently implementable on all
machines.  On the Vax, users will often have declared a single type of
float for efficiency, and compiled this in; in such cases, we clearly
want a runtime error rather than a "helpful" coercion.  I think that it
is probably best for the white pages to specify that floating-exponent
overflows always signal an error.

Ideally, the error handler should get enough info so that the user can
supply a quietly-size-expanding handler if he wants to.  This means
that, if possible (can we require this?) it should be a correctable
error, which is passed the operation name and the original args; the
handler can then return the "answer", computed however it likes and in
any format it likes.

-- Scott