comment-out R::pythag

[eddelbuettel]

Per email from BDR

[eddelbuettel]

No comments by anyone? Should be harmless -- may just merge in a few ...

[kevinushey]

We might (?) need a few more changes on our side:

inst/include/Rcpp/sugar/functions/complex.h
28:# define RCPP_HYPOT ::Rf_pythag

inst/include/Rcpp/sugar/undoRmath.h
110:#undef pythag

inst/include/Rcpp/Rmath.h
222:    inline double pythag(double a, double b)        { return ::Rf_pythag(a, b); }

I think the bit in undoRmath.h can stay, but we might want to tweak what we have in complex.h:

kevin@cdrv:~/r/pkg/Rcpp [rng-state]
$ ag -Q "RCPP_HYPOT"
inst/include/Rcpp/sugar/functions/complex.h
26:# define RCPP_HYPOT ::hypot
28:# define RCPP_HYPOT ::Rf_pythag
83:             y.r = ::log( RCPP_HYPOT( x.r, x.i ) );
90:         if( (mag = RCPP_HYPOT(z.r, z.i)) == 0.0)
144:    t1 = 0.5 * RCPP_HYPOT(x + 1, y);
145:    t2 = 0.5 * RCPP_HYPOT(x - 1, y);

If I understand correctly, R removed that function because C99 now defines ::hypot() and so R doesn't need to provide its own definition. (If I understand correctly, Rf_pythag() was R's own definition of ::hypot(), for platforms that didn't provide it back in the day)

[kevinushey]

That said, if everything compiles and checks okay we can probably leave it as is?

[eddelbuettel]

There is indeed a bit more. You first catch is good: RCPP_HYPOT is conditionally defined with a now-dead fallback. The undef is harmless. The inline we can just comment out as well.

The use of RCPP_HYPOT may need a second look as RCPP_HYPOT now has no fallback. Question is then ... would #ifdef HAVE_HYPOT ever not be true? And if so, shall we just emit a warning there?

man 3 hypot points to C99, POSIX.1-2001, POSIX.1-2008. Good enough?

Also:

edd@rob:~$ grep HAVE_HYPOT ~/svn/r-devel/config.status 
S["RMATH_HAVE_HYPOT"]="# define HAVE_HYPOT 1"
D["HAVE_HYPOT"]=" 1"
edd@rob:~$ 

[eddelbuettel]

The cleanest may just be to rewrite complex.h and do away with RCPP_HYPOT in favour of ::hypot. Let me pile that on later today.

[kevinushey]

I think HAVE_HYPOT is effectively always true since (as you pointed out) it's defined by the C99 standard and I think R requires C99 or greater during compilation of R + extensions.

[eddelbuettel]

@kevinushey That was an excellent catch. I maded some changes, when you have moment, could you peruse?

[kevinushey]

LGTM -- thanks for taking care of this!