From: Greg Chicares (chicares_at_[hidden])
Date: 2001-05-22 18:21:07
Beman Dawes wrote:
> At 09:18 PM 5/21/2001, Greg Chicares wrote:
> > ... tests of Borland and GNU fp results with various switch settings ...
> So is the conclusion that limits_test.cpp is a standards conforming
I believe that even the original limits_test.cpp was conforming.
It used the result of std::numeric_limits<float>::quiet_NaN() when
has_quiet_NaN != false, so 18.104.22.168/46 says this was "meaningful"
regardless of the value of is_iec559.
I'm not sure whether the C++ standard prescribes that a quiet NaN
mustn't behave as a signaling NaN, even though everyone knows
it shouldn't; apparently footnote 193's reference to LIA-1 is not
> , and that Borland, Metrowerks, and Microsoft compilers all have
> bugs in their FP processing?
I don't have mw or ms; as for borland, I'd rather say that
- To put it in its best C++ conformance mode, the -A switch is
insufficient: -ff- is required as well.
- To put it in its best IEC559 conformance mode, one needs to call
a documented nonstandard library routine to change the default
math hardware settings initialized by the startup code.
Bugs, or features? I'd rather say "informed choices that I dislike".
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk