Вот в одном месте еще любопытный пассаж увидел.
P.Sterbenz, Floating point computations (1974) , p.67 писал(а):
Next, suppose that we try to perform a floating point division of zero by zero. Here, there is no a generally acceptable value to be used in further calculation. In fact, such a division is usually a signal of trouble in the program. The only reason for not terminating a program is that we may want to proceed to other parts of the program or other data that will not be contaminated by the result of this division. The error message printed for division by zero should distinguish the case
from the case
with
. Moreover, a different fixup, say
or
, should be used for
.
-- 22.07.2020, 22:25 --(Это, конечно, книга старая (но местами интересная). В современных компьютерах
--- это, конечно, NaN).