RE: IEEE 754R Meeting April 15, 2004 Raw Minutes

["Kent Karlsson" <kentk@xxxxxxxxxxxxxx>]

Gabriel Dos Reis wrote:
...
> As far as I can tell, std::numeric_limits<> was designed so that
> C++ implementations can choose to comply to LIA-1, IEEE-754
> and/or tell when they support SNaNs -- the numeric_limits<>
> view is biased toward LIA-1, but it clearly makes room for SNaNs.

There is nothing in LIA-1 that sais anything "against" (S)NaNs, nor
"against" negative zeroes or infinity values. LIA-1 does not mention
them at all normatively, but LIA-1 is formulated to be compatible
with IEEE 754 (and 854), and hence allows for (S)NaNs.

Note though that LIA-2 (published) and LIA-3 (being drafted) mention
NaNs (as well as negative zero and infinities) explicly in the normative
operation specifications. But there is no requirement for their
existence,
so there is no disaster if SNaNs are removed from 754R. Nor does LIA
take any stance on the [in]untility of SNaNs. But negative zeroes and
infinites are carefully (I hope) considered. The changes for decimal in
754R, compared to 854, are more worrying (for a revised LIA-1).

                /Kent Karlsson
                (current editor LIA series of standards)

PS
I've asked before, but I'm asking again: If any of you would like to
comment on the draft LIA-3 (Complex integer and floating point
arithmetic and complex elementary numerical functions), I'd be happy
to send you a copy for review. I would much appreciate review and
comments from you. We hope to submit a version for ISO/IEC (F)CD
balloting in about a month.