Quick comments on draft 1.7.0

[Michel Hack <hack@xxxxxxxxxxxxxx>]

(This is a short recirculation period, so I want to trigger reply
comments early before compiling my final comments.)

I am quite pleased with the result, given that we try to avoid major
updates at this juncture.  Most of my draft 1.6.0 comments were dealt
with in a satisfying manner, except for one:  the definition of the
Invalid exception in 7.6 where the only change was to remove the 1.6.0
addition of the mention of unpredictable indications for functions of
Clause 9.

Am I really the only one to feel that the definition of Inexact as given
in 7.6  ("result would differ if range and precision were unbounded")
does not apply to quantize() and roundToIntegralExact(), and that these
two operations should therefore be mentioned explicitly in 7.6?

Ok, here possible new comments.

Page 55 line 4:  Say "For numeric results, one rounding..."

Page 57 lines 30 and 32:  change "cohort" to "quantum".

    (0+x and x are necessarily in the same cohort, except
    possibly when x is -0, and so are 1*x and x).

Page 57 line 30:  This is pretty difficult to achieve, and therefore
    possibly pointless.  x and 0+x have the same quantum only when
    the exponent of 0 is not smaller than that of x, and so this is
    unlikely to be known at compile time (unless the 0 is the one
    with the maximum exponent).

Page 57 line 32:  On the other hand, if the "1" is 1e0, i.e. the
    plain integer 1, then 1*x and x are guaranteed to have the same
    quantum -- and if the "1" has a smaller quantum (e.g. 100e-2),
    then the quantum of the product will be different unless x is
    subnormal.

    So this could simply be stated that the identity 1*x = x can be
    exploited if and only if the exponent of "1" is zero.

    Both of these clarifications should probably be qualified as
    applying to decimal.

Finally, there will be something about minNum() and friends -- either
a strengthening of the definition ("for decimal, if the operands are
numerically equal, the first operand is returned"), or a listing of
unpredictable quantum for minNum() etc. in clauses 10 and 11.

Michel.
Sent: 2008-03-15 23:36:19 UTC