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

RE: X3T2 activities

As it so happens, X3T2, Language Compatible Arithmetic Standard, has
already included a section on Common Lisp.

One minor point was brought up:  The definition of the xxx-EPSILON
constants is not quite right.  The problem is that if these values
represent the smallest E such that (/= (+ 1 E) 1), then they will have
to change according to the rounding mode if it's an IEEE implementation.
However, I thought that any particular CL implementation could work
around this by not implementing the xxx-EPSILON constants by the usual
DEFCONSTANT mechanism.

On the other hand, Common Lisp could change its definition of those
constants to be the same as what LCAS specifies.

It defines epsilon to be radix^(1-precision).  This is the ratio of
the weight of the least significant digit of the fraction  radix^(-precision)
to  1/radix.  Another description is that epsilon is the largest relative
representation error for all normalized values.

Copies of ANSI document X3T2/91-006, LCAS V3.0, 26 Dec 1990, can probably
be obtained (I haven't asked if they're able and willing) from Brian
Wichmann, baw@seg.npl.co.uk, or Martha Jaffe, jaffe@rdvax.enet.dec.com.
If they can't provide the file via the Internet, I'm willing to make
hard copies.