How is NCEG going to resolve this conflict...

[David G. Hough on validgh]

between exception handling requirements of various standards.


Actually it's much worse than Fred indicates.  SVR3 and SVR4 implementations
have to call matherr while ANSI-C implementations must not.  Makes
it tough to claim conformance to both by default.  X/Open and POSIX
may have muddied the waters further - I expect to learn more about that
over the next few months.  Basically the approach I expect that Sun
will eventually adopt will be a number of compile time/run time modes
that allow you to select the standard you want to conform to.   Only
one can be the default, however.  The only thing worse than amateurs
setting standards is amateurs writing standard validation test suites,
but I've complained about that often enough.

I think for NCEG the right thing to do is require that NCEG extensions
include redefining a number of "functions" to be "operators in functional
notation" including at a minimum the math.h functions and abs.
Then it will be natural to understand that exceptions for those operators
won't be specified any more than exceptions for +-*/.

Implementations conforming to the NCEG IEEE extensions will go further.
For all those operators mentioned it's fairly clear what to do about
exceptions - treat them the same as existing operators.