Re: confused by liboctave/util/lo-ieee.cc

[John W. Eaton]

On 06/16/2017 01:17 PM, Rik wrote:
> While working on other things I stumbled upon lo-ieee.cc and I have some
> questions about the implementation.
>
> First, now that C++11 is an Octave requirement, do we really need to check
> for standard functions being in the cmath library?  It seems like a hard
> requirement that the corresponding C and C++ libraries implement what is in
> C++11.  Example code is

I suppose not, if C++11 says they should exist.  The only case we should 
have to worry about is if we need to replace one of these functions 
because of some known bug, but that's a separate issue.

> --- Code #1 ---
> int
> __lo_ieee_isnan (double x)
> {
> #if defined (HAVE_CMATH_ISNAN)
>   return std::isnan (x);
> #else
>   // Gnulib provides.
>   return isnan (x);
> #endif
> }
> --- End Code #1 ---

I guess we should just deprecate these functions now if isnan is a 
standard function that we can expect to exist on all platforms that we 
care about.

> Second, if the #if defined checks should remain, shouldn't they differ
> between float and double versions?

Yes, that's probably just an oversight.

> Finally, if we're relying on the C++ std library which contains templated
> versions of functions, why do we need to have special code for floats?

I don't think we should.  Let's just use the std:: functions directly 
now and eliminate our wrappers.

> In
> the extracted code below, the case of doubles just uses std::isfinite, but
> the float version is more complicated.

I don't know why there was a difference.

jwe