No subject

[Doug Gwyn (ACISD/MCSB)]

There *is* no standard for C++ yet, but the C standard does allow f.p.
representations to be *slightly* sloppy (generally off in the last bit
position of the *advertised* precision, which is the lowest common
denominator for the various kinds of actual representation that could
be used within a single implementation).  There is an open Defect
Report, submitted by Tom Plum, requesting an official ruling on the
actual f.p. precision requirements of the C standard.  This should
prove a lively topic.

"Normalizations" as such are certainly not specified by the C standard,
because that is a lower-level concept than is addressed by the high-
level language standard.  There must be *some* representation, and
apparently certain parameters advertised in <float.h> are meant to
describe some of the characteristics of the representation(s), but
there is no requirement that (for example) the IEEE f.p. representation
be used.  It was generally felt that it would be inappropriate to try
to constrain f.p. implementations too tightly in the generic C standard.

It is possible that some IEEE f.p. standard PLUS some programming
language standard would result in a tight constraint on the kind of
behavior you were asking about.  So far as I can determine, such a
combination of IEEE f.p. and C standards would be counter to the intent
of the C standard to allow implementations some freedom to optimize
computation.