[R] Cannot call R's ISNAN() from a C code in >1.7 versions.
Prof Brian Ripley
ripley at stats.ox.ac.uk
Mon May 24 21:07:01 CEST 2004
On Mon, 24 May 2004, Paul Y. Peng wrote:
> Prof Brian Ripley wrote:
> > First, there are no versions 1.7, 1.8 and 1.9.
>
> Sorry for my misuse of the version numbers.
>
> > Was your version of R compiled against MSVC++ 6.0? The binary on CRAN was
> > not, and binaries for different versions of R were compiled with different
> > versions of MinGW. The entry point isnan is part of the statically linked
> > runtime on modern MinGW.
>
> I used the binary on CRAN.
Then please use the compilers described in readme.packages which match it.
> > MSVC++ 6.0 does supply _isnan (as it really should as it is part of the
> > C99 ISO standard), and you need to link against it appropriately. Hint:
> > it may have an extra underline, since it seems it is known to C as _isnan.
> > You may need to add
> >
> > #undef ISNAN
> > #define ISNAN(x) _isnan(x)
>
> Many thanks for this suggestion. It works, as always. The extra
> underline is required.
>
> > We don't support adding extensions to R using a different compiler to the
> > one used to build R. Changes already made for future releases of R will
> > make this less likely to work in R 2.0.x.
>
> I wish that the R API entry points documented in "Writing R Extensions"
> be supported in the future versions of R, because it will make programs
> built with R more portable than directly using compiler-specific
> functions, such as testing and generating the IEEE 754 special values.
> The existing entry points have saved me trouble to chase these values
> whenever I moved to an environment with a different compiler.
Who said they would not supported? What I said is that using a different
compilers to compile different parts of R is not supported (and never has
been). And that is even less likely to work in future releases. All we
guarantee is that ISNAN and R_FINITE produce calls to functions that work
under the system used to configure R. As on Windows R_FINITE now produces
a call to a MinGW macro, that is not going to work under VC++ 6.
We are not talking about using `compiler-specific functions' here, like
_isnan. What R core is doing is using *standard* C99 functions where
available, and a problem with VC++ 6.0 is that it is far from compliant
with IEC60559 aka IEEE754. Given that the recommended compiler is freely
available, it is unreasonable to expect any support for less capable
compilers.
--
Brian D. Ripley, ripley at stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272866 (PA)
Oxford OX1 3TG, UK Fax: +44 1865 272595
More information about the R-help
mailing list