Add is_inf and is_nan functions to builtin::

[leonerd]

As it is not expected that these two functions are likely to be called very often, I didn't bother making inlined-opcode variants of them as per e.g. is_bool(). Instead these are just regular XSUBs like trim() is.

Testing-wise, I'm a little unsure about the use of 1 * (inf - inf) to get a calculated NaN value, but I really couldn't think of anything else that actually works. Is it portable enough?

[tonycoz]

The NOK check seems non-perlish to me, it also breaks overloads:

tony@venus:.../git/perl6$ ./perl -Ilib ../22059.pl 
Built-in function 'builtin::inf' is experimental at ../22059.pl line 6.
Built-in function 'builtin::is_inf' is experimental at ../22059.pl line 11.
Built-in function 'builtin::is_inf' is experimental at ../22059.pl line 13.

1

1
tony@venus:.../git/perl6$ cat ../22059.pl
use v5.38;
use builtin qw(is_inf inf);
use POSIX();

package Foo {
  use overload '0+' => sub { inf };
}

my $x = bless {}, "Foo";

say is_inf($x);
say POSIX::isinf($x);
say is_inf("Inf");
say POSIX::isinf("Inf");

[tonycoz]

Another issue with this (and ceil(), floor() for that matter), is it doesn't support overloads in the same way other numeric functions do, e.g.:

use overload sin => sub { ... };

It's less of an issue if the 0+ overload returns a Inf/Nan as needed (which this code also doesn't support), but I can see it as an issue for ceil/floor with bignum-ish overloads.

[jkeenan]

Another issue with this (and ceil(), floor() for that matter), is it doesn't support overloads in the same way other numeric functions do, e.g.:

use overload sin => sub { ... };

It's less of an issue if the 0+ overload returns a Inf/Nan as needed (which this code also doesn't support), but I can see it as an issue for ceil/floor with bignum-ish overloads.

@leonerd, @tonycoz has noted some significant problems with this pull request. Do you wish to pursue it further? Thanks.